|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
VERB MEANINGS AND THEIR EFFECTS ON SYNTACTIC BEHAVIORS:
A STUDY WITH SPECIAL REFERENCE TO ENGLISH AND JAPANESE
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
To Chieko with love and gratitude
I have so many people to thank for the completion of this dissertation. First, I
would like to thank Dr. Ann Wehemeyer, the chair of my supervisory committee, for her
constant support, valuable advice, and her outstanding scholarship. Whenever I felt that I
lost perspective, she always came up with articles and books where I could find light to
follow. I should also thank Dr. D. Gary Miller for his abundant feedback on my
dissertation. I will never forget the magnificent view from his office at Turlington where
we had heated discussions on focal points of my dissertation. I also would like to thank
Dr. Marie Nelson and Dr. Joseph Murphy for being patient with my slow progress in my
dissertation. I should not forget to thank Dr. Mohammad Mohammad for being a
committee member before he left for University of Texas.
Among the excellent graduate students in the Program in Linguistics at the
University of Florida who constantly gave me moral support and served as informants for
this dissertation were Jodi Bray, Kim Duk-Young, Evelyne Ngauchi, Jongbum Ha, and
Philip Monahan. I also need to thank my friends John Pasden, a former student in
Beginning Japanese, and Larry Tankersky, my American mentor, for providing me with
plentiful linguistic insight. Their keen interest in linguistics made me proud that I am a
linguist. I should not forget to express my gratitude for the staff at the Electronic Thesis
and Dissertation office. Whenever I had problems with my template, ETD counselors
solved every single one of them in a split second.
I would also like to express my sincere gratitude to Robin Gibson and Jean
Gibson, whose Gibson Dissertation Fellowship at the University of Florida College of
Liberal Arts and Science enabled me to devote ample time to doing research for this
dissertation. Their generous support enabled me to complete the dissertation half a year
Finally, I would like to thank my wife Chieko for her constant support and
encouragement that she has never ceased to give to me for more than five years since we
came to Gainesville. Without her, I would not have been able to muster the courage to
come over to America, let alone complete this dissertation.
TABLE OF CONTENTS
A C K N O W L E D G M E N T S ....................................................................... .....................iii
LIST OF TABLES .................................................... ........... ............ viii
A B STR A C T................................................... ix
1 IN TRODU CTION ........................... ...................... .. ..................1
2 ERGATIVE ALTERNATION ............ ..... ................................. 11
2.1 Introduction ......................... ... ...... ......... ................................. ......................... 11
2.2 E english E rgative P airs ...................... .. .. ......... .. .......................... .............. 12
2.2.1 Syntactic A aspects .......................... .... ...... ...... ... ................... 12
2.2.2 Ergative Alternation and Other Similar Syntactic Operations...................... 14
184.108.40.206 E rgatives and passives ........................................ .......... .............. 15
220.127.116.11 E rgatives and m iddles ........................................ ......................... 16
2.2.2 Semantic Aspects ... ..... ............................................ .............. 20
2.3 Japanese E rgative Pairs ...................... .. .. ......... .. ........................ .............. 23
2.3.1 Syntactic C characteristics ...................... .. .. .............................. .............. 23
2.3.2 M orphological Characteristics ........................................ ....... .............. 24
2.4 E rgative A lternation: Issues ............................ .............................. .............. 29
2 .5 U n accu sativ ity .............................................................................. 3 0
2.5.1 The Unaccusativity Hypothesis ........................................................ 31
2.5.2 U naccusative V erbs in English.............................................. ... ................. 33
2.5.3 Unaccusative Verbs in Japanese .............................................. 35
2.5.4 Semantic Characterization of Unaccusativity ............................................ 41
2.5.5 Unaccusativity and Ergative Alternation.............................. .................... 46
2.6 D direction of D erivation .............................. ........................................... 48
2.6.1 Derivational Direction of English Ergative Pairs ........................................ 49
2 .6.1.1 C au sativization............................... ............ .............. .. ............ 49
2 .6.1.2 A nticau sativization .................. ............................... ...................... ... 5 1
2.6.2 Derivational Direction of Japanese Ergative Pairs ....................................... 53
18.104.22.168 Transitivization vs. intransitivization...................... .............. 54
22.214.171.124 Causativization vs. passivization............. ............................................. 55
3 SEMANTIC CHARACTERIZATION OF ERGATIVE ALTERNATIONS ...............59
3 .1 In tro d u ctio n ..................... ... .......................................................................... 5 9
3.2 Tw o Issues of Ergative Alternation ............. ................................... .............. 60
3.3 A Problematic Case: The English Cut and Break ............... ............. .............. 62
3.4 Conditions for Ergative Alternation: English........ ............................... 67
3 .4 .1 C change of State ........6........... ...... ........... .... .. ...... ... .. ............ ... ...... .. 67
3.4.2 A gentivity ................................................................................ .. ................ 71
126.96.36.199 Kill verbs...... ......... ............................. 73
188.8.131.52 D estroy verbs ................................... ......... ........... .. ....... ...... .. 75
3.4.3 External vs. Internal C ausation.............................................. ... ................. 77
3.4.4 Onset Causation vs. Extended Causation................................................. 79
3 .5 L exical Specification ................. .. ........................................ ................ .. 82
3.5.1 Lexicalization of Instrument ............. ............ ....... .... 82
184.108.40.206 Case theory........................................................... 82
220.127.116.11 Lexical Conceptual Structure (LCS) .................................................. 84
3.5.2 Specification of Cause or Means ....................................................... 85
3.6 Conditions for Ergative Alternation: Japanese ............................................... 87
3.6.1 C change of State ......................... .. ........... .............. ...................... .................... 88
3.6.2 Specification of Instrument and Means .................................................. 92
3.6.3 Inanim ate Subjects .................... ................. ........................ .............. 95
3.6.4 Kir-kire- 'cut/get cut' ................................. ..................................... 97
3.6.4 Polysemy ....................................................... .. ................. 101
3.6.5 Sino-Japanese Change-of-State Verbs ................................. ...... .... ... 103
3 .7 S u m m ary ............................................................................... 10 6
4 SEMANTIC RELATIONS BETWEEN ERGATIVE PAIRS ..................................108
4 .1 Introduction ................ ..... ...1 ............................. ............... 108
4.2 Lexicalist View of Japanese Ergative Pairs ...................................... ............ 111
4.3 Lexical versus Syntactic D erivations .............................................................. 114
4.3.1 Lexical Derivation........................................ 115
4.3.2 Post-L exical D erivation ......... .................................................. .............. 117
4.4 Sem antic A pproach............. ............ .. ........ ...... .. ..... ... ........... 121
4.4.1 Semantic Discrepancies between Japanese Ergative Pairs .......................... 122
4.4.2 Spurious Ergative Pairs ............................... ................ .............. 128
4.5 Dichotomous View of Japanese Ergative Pairs ...................... .................... 132
4.5.1 Idiom atization ........ ............... ....................................................... 132
4.5.2 Paradigm atic Structure (PD S) ..................................................................... 133
4.5.3 PD S for Ergative Pairs .............. ......................................................... 135
4.5.4 Distributed Morphology............................. .............. 139
4.6 Sum m ary ......... .. ....... ...................................................................... 140
5 ASPECT AND ERGATIVE PAIRS ...................................................... ................142
5.1 Introduction.. .............. ..................................................... 142
5.2 Japanese A spectual Properties: Issues ............................................................... 144
5.3 Classifications of English Lexical Aspect .................................................... 147
5.4 Japanese Verbs and their Aspectual Properties ........................................ ...... 149
5.5 Ergative Pairs and their Aspectual Properties............................ 153
5.5.1 Grammatical Functions and Thematic Roles ........................................... 154
5.5.2 Terminative Orientedness .................................. 157
5.6 Ergative Intransitives ................ .......... ....... ............... 159
5.6.1 Progressives in Ergative Intransitives in -te iru........................... 159
5.6.2 Change of state and endpoint................... ......................... .............. 162
5.6.3 D ow ty's (1979) Interval Sem antics ..................................... .............. 163
5.6.4 Syntactic Test -ow aru 'finish' .................................................................... 165
5.6.5 V-owar- 'finish V Test to the Ergative Pairs in Jacobsen (1992) .............. 167
5.7 Summ ary ................................. ................................ ........ 173
6 CONCLUSION AND FUTURE RESEARCH................................ ...............175
6 .1 C onclu sion ............................................... 175
6.2 Further Issues .................................................... ................. .......... .. 179
6.2.1 Validity of Lexical Semantics-Syntax Interface ............ ...... ............ .. 179
6.2.2 Causativization ............. .... ........... .................... .... .......... 182
APPENDIX: JACOBSEN'S (1992) LIST OF ERGATIVE PAIRS .............................185
B IB L IO G R A P H Y ..... .. .... ...... .... .......... .................................. .............. 195
B IO G R A PH IC A L SK E T C H ........................................ ............................................222
LIST OF TABLES
1: A lternations patterns of cut and break ......................................................................... 66
2: Dowty's interval-based classification of change-of-state verbs ....................................... 164
3: V-owar- test for Japanese ergative intransitives ..................................... .................169
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
VERB MEANINGS AND THEIR EFFECTS ON SYNTACTIC BEHAVIORS:
A STUDY WITH SPECIAL REFERENCE TO ENGLISH AND JAPANESE
Chairman: Ann Wehmeyer
Major Department: Linguistics
Ergative alternation is a process of valency shift observed cross-linguistically in
which verbs alternate in transitivity with little or no change in form. Verbs that
participate in the ergative alternation in English have been investigated from syntactic
and semantic perspective. Japanese ergative pairs, on the other hand, are characterized
not only by valency shift but also by their derivational opposition between transitive and
intransitive alternants. Given such characteristics, previous research has mainly focused
on describing the derivational patterns and classifying ergative pairs accordingly. This in
turn results in insufficient attention to semantic aspects of Japanese ergative pairs.
The major research goal of this dissertation is to investigate whether verbal
meanings determine the alternating behaviors of ergative verbs. Under the framework of
lexical semantics, this dissertation presents a semantic analysis of ergative pairs in
Japanese. I propose to utilize research findings reported in works on English ergative
pairs to explicate the cause of the alternating behaviors of Japanese ergative pairs. One
crucial finding is that semantic properties of verbs such as change of state and
specification of causation are key factors both in English and Japanese in differentiating
verbs that undergo the alternation from those which do not. Nevertheless, my research
shows that there is some significant disagreement between English and Japanese in terms
of the ergative altematability of verbs. I suggest that such discrepancy results in part from
lexical idiosyncrasies in each language. More importantly, however, I propose that
different alternating behaviors may reflect the way native English and Japanese speakers
perceive certain events differently. The implication of this observation will be that other
different syntactic behaviors between English and Japanese may reflect people's different
perception of events.
This dissertation presents a contrastive analysis of transitive/intransitive verb
pairs in English and Japanese. The primary goal is to investigate under the framework of
lexical semantics how semantic aspects of verbs affect the alternatability of Japanese
verbs. The transitive/intransitive verb pair, which I refer to as ergative pair1 throughout
this dissertation, is described either as a single verb which is used both transitively and
intransitively, as in English, or as a pair of morphologically related verbs, as in Japanese,
which respectively describe a transitive and intransitive situation (Levin 1985:19-20).
The transitive and intransitive member of the ergative pair is referred to as ergative
transitive and ergative intransitive, respectively. Examples of English and Japanese
ergative pairs include:
(1) TRANSITIVE INTRANSITIVE
English Japanese English Japanese
break war- break ware-
sink sizume- sink sizum-
bend mage- bend magar-
melt tokas- melt toke-
bake yak- bake yake-
Depending on the approaches adopted by linguists, "ergative" is also referred to as
"unaccusative." In Government-Binding theory, the term ergative is most commonly used
(Burzio 1986). In Relational Grammar, on the other hand, unaccusative is widely used
(Perlmutter 1978). Following Roberts (1987), I restrict the term ergative to verbs like
break and open which have transitive counterparts. This allows me to distinguish ergative
verbs from unaccusative verbs like come and appear which lack transitive counterparts.
As will be discussed in detail in Chapter 2, the distinction between ergative and
unaccusative is crucial to the present research.
Ergative pairs are also characterized by change in argument structure. Consider the
following typical case of English ergative alternation
(2) a. Tom broke the vase.
b. The vase broke.
Crucially, it is observed that the syntactic object (vase) of the ergative transitive (break)
corresponds to the syntactic subject of the ergative intransitive. This type of
configurational correlation characteristic of the ergative pair is specifically referred to as
In Western linguistics, the ergative alternation has been investigated for more than
three decades within theoretical frameworks such as Case Grammar (Fillmore 1968b,
1970), Generative Semantics (Lakoff 1968a, 1970), Government and Binding Theory
(Burzio 1986), Cognitive Grammar (Langacker 1991), and Lexical Semantics (Levin
1993, Levin and Rappaport Hovav 1994, 1995). In particular, Fillmore (1970) focuses not
only on verbs that participate in the ergative alternation but also on verbs that do not,
attempting to account for the cause of the difference semantically. What is striking about
Fillmore is that he laid the foundation for the subsequent lexical semantic approaches to
this issue. Building on a more in-depth analysis by Smith (1978) of the ergative
alternatability of verbs, Levin and Rappaport Hovav (1995) devote a whole chapter
(Chapter 3) to investigating the semantics-syntax interface involved in the English
ergative alternation (Levin and Rappaport Hovav refer to it as "causative alternation"),
2 The ergative alternation is also referred to as causative alternation (Haspelmath 1993,
Levin and Rappaport Hovav 1995), transitivity alternation (Hale and Keyser 1987), and
unaccusativity alternation (Kiparsky 1998), among others.
arguing that whether a given verb participates in the alternation depends to a great extent
on the semantic properties lexically inherent in the verb.
In Japanese linguistics, on the other hand, previous research on ergative pairs has
mainly focused on their distinct suffixal forms as shown in (1) above, proposing a range
of classifications accordingly (Sakuma 1936, Teramura 1982, Shibatani 1990, Jacobsen
1992). This may partly explain why there have been only sporadic and incomplete
semantic analyses provided for Japanese ergative pairs (cf. Okuda 1978, Jacobsen 1982a,
1992, Hayatsu 1987, Mitsui 1992, Kageyama 1996). With a thorough lexical semantic
approach presented in the present study it will be shown that Japanese ergative pairs can
be semantically characterized to a large extent. This in turn suggests the possibility that
the alternatability of a verb will be better accounted for in reference to the semantic
properties of the verb. Furthermore, such a lexical semantic approach will suggest a need
to propose a morpho-semantic classification distinct from the morphology-based
classifications proposed in the past (cf Hayatsu 1989).
One question that has been continuously addressed in the literature on lexical
semantics over the decades is whether the syntactic behaviors of verbs are predictable
from the semantics of the verbs (Perlmutter 1978, Wasaw 1985, Pinker 1989:104ff, Levin
1993a, Levin and Rappaport Hovav 1995:1, 1998, Ono 1997).3 In particular, researchers
have sought to uncover semantically coherent verb classes which are fairly constant with
respect to argument structure. Under this verb-class model, let us consider the English
3 As for the view that the relationship between the lexical-semantics and syntax is not
necessarily consistent, see S.T. Rosen (1996), Lemmens (1998), and Rosen and Ritter
(1998). See also Chapter 6 of the present dissertation.
verbs happen and cut. The verb happen belongs to a semantic verb class called 'verbs of
occurrence' and can only occur in intransitive constructions (cf. Levin 1993a:260-261):
(3) a. The accident happened yesterday.
b. *My brother happened the accident yesterday. (On the interpretation that
'My brother caused the accident yesterday')
By contrast, the transitive verb cut, which belongs to the semantic class 'verbs of
cutting' fails to occur in ergative intransitive constructions (Levin 1993a:156).
(4) a. The butcher cut the meat.
b. *The meat cut. (On the interpretation that 'The meat got cut')
Given the correlation between verbs and their syntactic behaviors just described,
Levin and Rappaport Hovav (1994, 1995) conduct detailed analyses of the semantics of
English ergative pairs, seeking to determine which semantic properties contribute to
verbs' altematability. Drawing on a representation of verb meaning referred to as
predicate decomposition, Levin and Rappaport Hovav (1995:94) propose the following
lexical semantic representation template for ergative verbs:
(5) [[x DO-SOMETHING] CAUSE [y BECOME STATE]]
Accordingly, the ergative verb break is represented as follows (Levin and Rappaport
(6) break: [[x DO-SOMETHING] CAUSE [y BECOMEBROKEN]]
One key feature in the representation in dealing with the issue of the lexical semantic-
syntax interface, according to Levin and Rappaport Hovav, is the primitive BECOME. On
their view, this primitive represents the semantic property 'change of state,' which Levin
and Rappaport Hovav view as one of the crucial elements which have to do with the
ability of verbs to alternate in transitivity (for detailed discussion of this issue, see
In Japanese the ergative alternation is typically represented as follows:5
(7) a. Taroo-wa kabin-o wat-ta. (< war- + ta)
Taro-TOP vase-ACC break-PAST
'Taro broke the vase.'
b. Kabin-ga ware-ta.
'The vase broke.'
One characteristic associated with the Japanese ergative alternation in comparison to the
English equivalent is that the former marks the difference in transitivity morphologically,
as readily observed in the contrast between war- and ware- (Tsujimura 1990b:938). Since
there are a number of distinct suffixal forms associated with ergative transitives and
intransitives, one primary research goal researchers have attempted to achieve is to
classify this particular group of verbs based on their derivational features (cf. Chapter 2;
see also Shimada 1979 for a detailed outline of previous studies).
Compared to the rich tradition of morphological research on verb classifications
involving ergative pairs, their syntactic and semantic aspects have not received much
attention from researchers. As for syntactic characterization of ergative pairs, Okutsu
(1967) first introduced the so-called dynamic (dootai-teki) approach to the derivational
verbal morphology based on generative grammar. The dynamic approach differs from the
4 Another key assumption of the template is that the underlying semantic structure of break is a transitive
one. On this view, the ergative intransitive use of the verb is derived by virtue of detransitivization, a
completely opposite process to causativization which is widely held to account for the ergative alternation
in general. See Chapter 2 for more discussion of this issue.
5 In this dissertation the following abbreviations are used: ACC=accusative particle, ASP= te-iru aspect
marker, DAT=dative, EMP=emphatic marker, GEN=genitive, GER=gerund, INF=infinitive,
NEG=negation marker, NOM=nominative particle, OBL=oblique, PAST=past tense marker,
PRG=progressive, TOP=topic particle.
(1967) first introduced the so-called dynamic (dootai-teki) approach to the derivational
verbal morphology based on generative grammar. The dynamic approach differs from the
so-called static (seitai-teki) approach in that, according to Okutsu, the former approach
explores the syntactic derivation of transitive verbs (i.e., transitivization) and intransitive
verbs (i.e., intransitivization) mediated by the addition of derivational suffixes, while the
latter approach focuses on classifications ofergative pairs based solely on morphological
distinctions. Most importantly, Okutsu adopts the idea of 'embedding' from generative
syntax, arguing that ergative transitive constructions are bi-clausal structures where an
intransitive clause is embedded into a transitivizing clause containing the transitivizing
suffix -as. Inoue (1976) further develops a generative syntactic approach to ergative pairs
With the growing interest in lexical semantics over the past three decades,
researchers have investigated the relationship between verbs' ability to alternate in
transitivity and their meanings. Okutsu (1967) is among the first scholars who point to
the significance of semantic-based analyses and encourage further research along these
lines. Following the earlier attempts by Miyajima (1972) and Nishio (1978), a series of
work by Hayatsu (1987, 1989, 1995) attempts to determine the semantic properties of
verbs that have to do withthe verb's ability to alternate in transitivity. As with English,
one semantic property thus isolated is change of state that is brought about on a Theme
argument. Furthermore, Hayatsu notes that the subjects of ergative intransitives have a
strong tendency to be inanimate. In other words, ergative intransitives containing animate
Themes are most likely to resist alternating with transitive counterparts in Japanese.
Overall, Hayatsu's approach is notable in that it is comparable to the lexical semantic
approach that has been intensively discussed in the Western linguistics.
Drawing more on the findings in lexical semantics in Western linguistics,
Kageyama (1996) extends a long discussion about the relationship between verb
meanings and the syntactic realizations of their argument structures. In keeping with
Lexical Conceptual Structure (LCS) analogous to Levin and Rappaport Hovav's lexical
semantic representation as in (5) above, Kageyama conducts a comparative analysis of
the English and Japanese ergative alternations. On his terms, the alternation is not only a
morphological or/and syntactic phenomenon but also a semantically explicable process.
Kageyama basically agrees with other scholars like Smith, Haspelmath, and Levin and
Rappaport Hovav, among others, in arguing that a given verb participates in the ergative
alternation when a Theme argument is perceived to bring about a change on its own
without any intervention of an external causer or agent (Kageyama 1996:158ff). What is
noteworthy about Kageyama, however, is that he attempts to fuse Western lexical
semantic approaches and traditional affix-based approaches together to account for
Japanese ergative pairs and their alternating behaviors. Specifically, Kageyama assumes
that the extent to which a given event is perceived to occur spontaneously or to be
intervened by an external causer or agent is signaled in a fairly consistent manner by (the)
suffixal forms added to Japanese ergative verb pairs.
Another line of semantic approach to the Japanese ergative pairs is proposed and
discussed in Jacobsen (1982a, 1992). Under the assumption that there are regular
correlations between morphological markedness and semantic markedness, Jacobsen
associates morphological patterns shown by ergative pairs with speaker's empirically-
based event views. Based on his markedness theory, Jacobsen argues that if one member
of a transitive-intransitive pair is more marked than the other, then the morphological
markedness reflects the atypicality of a change-event.6 For instance, if an intransitive
member is more marked than its transitive counterpart, the change of state denoted by the
intransitive verb is perceived to occur less naturally in an autonomous situation than in a
coerced situation. Furthermore, Jacobsen defines the "typical realization of a concept"
(1992: 10) as prototype, maintaining that in spite of some major historical changes in
verbal morphology, fairly regular markedness patterns observed in many of the Japanese
transitive/intransitive verb opposition still reflect a strong correlation between
morphology and semantics. In short, Jacobsen holds that morphological and semantic
markedness is a reflection of universal linguistic principles, suggesting the universality of
prototype held among human beings.
It has been noticed in the literature that verbs that alternate in transitivity in one
language are likely to undergo the same phenomenon in other languages (1993:92, Hale
and Keyser 1998:89). Cross-linguistically, for instance, verbs equivalent to the English
break undergo the ergative alternation, whereas verbs equivalent to the verb laugh do not
(Pinker 1989:134). Given this observation, this dissertation research purports to
6 Jacobsen's position that the correlation between formal markedness and semantic
markedness observable among ergative pairs is based on the idea of prototype rather than
that of accumulation of information is notable when compared to the general principle
that languages show formal markedness patterns which reflect proportional accumulation
of semantic information (Dowty 1979:46). In this general principle, a formally marked
lexical item or grammatical structure is provided with more semantic properties than a
formally unmarked counterpart is. Cross-linguistically, for instance, plural nouns are
formally more marked than singular nouns since they, in addition to the basic category
'nominal,' contain 'plurality,' which is seen as a more marked feature than 'singularity'
(Greenberg 1963, Haspelmath 1993:87).
supplement data from Japanese for further research on the lexical-semantics and syntax
relationship. Furthermore, in view of insufficient cross-linguistic data in this area of
research (Levin 1985:61), a study of the Japanese alternation that this dissertation
presents from a lexical semantic point of view will serve to further explicate the
mechanism of the lexical-semantic and syntactic interface cross-linguistically.
The organization of the dissertation is as follows. In Chapter 2, I first outline
characteristics of ergative alternations in English and Japanese. Then, I discuss
unaccusativity and middle constructions, which are assumed to be relevant to ergative
alternations. I analyze morphological opposition which characterize the Japanese
ergative alternation. In particular I focus on the issue of the direction of derivation
regarding the alternation. I will suggest that the directions of Japanese ergative pairs
basically follow morphological markedness patterns.
Chapter 3 deals with semantic aspects of ergative pairs. I first discuss semantic
properties of English and Japanese verbs that appear to be responsible for the syntactic
expressions of arguments occurring with the verbs. Then, I demonstrate that a given
change-of-state verb is paired with the transitive/intransitive counterpart when a means or
an instrument responsible for the change of state is unspecified. In this view, lack of the
intransitive counterparts of Japanese transitive verbs like kar- 'cut (with a sickle, scissors,
etc.)' and hik- 'grind, mince' will be accounted for on the basis of the fact that both verbs
lexically specify a means or an instrument whereby the events are brought about.
In Chapter 4, I review in detail the classification of ergative pairs presented by
Jacobsen (1992). Adopting the concept of semantic transparency (Aronoff 1976), I
propose that the ergative pairs under the classification of Jacobsen be separated into those
which are semantically transparent and those which are not. Drawing on the theory of
Distributed Morphology (Halle and Marantz 1993), I argue that ergative pairs that are
semantically transparent are derived post-lexically, whereas those which are semantically
opaque are listed as separate lexical items in the lexicon.
In Chapter 5, I turn to the issue of aspect of the Japanese ergative pairs. I first
illustrate that intransitive members of ergative pairs have a perfect meaning and transitive
members have a progressive meaning, respectively, in combination with the aspectual
marker -te iru. I propose that the type of an argument in the syntactic subject position
determines the aspectual realization of a predicate suffixed by -te iru. Furthermore, given
that certain intransitive members can be interpreted as progressive in -te iru
constructions, I suggest that the semantic property 'change of state' should be sub-
categorized according to 'length of time' and 'definite endpoint' necessary for a change
to be completed.
In Western linguistics, the phenomenon of alternation has been regarded as one of
the prominent syntactic behaviors regarding verbs and their co-occurring arguments. In a
broad sense, the ergative alternation is classified under the macro-category called
diathesis alternation. Diathesis alternations are concerned with "alternations in the
expression of the arguments of verbs" (Levin 1993b:80), subsuming, in addition to the
ergative alternation, other types of alternations such as locative alternation and dative
(1) Locative Alternation:
Jack sprayed paint on the wall.
Jack sprayed the wall with paint. (Levin 1993a:51)
(2) Dative Alternation:
Bill sold Tom a car.
Bill sold a car to Tom. (Levin 1993a:46)
One characteristic that differentiates the ergative alternation from the dative and locative
alternations is a change in valency. As evident in (1) and (2), the dative and locative
alternations have to do with a change in the syntactic arrangement of arguments of verbs.
On the other hand, the ergative alternation involves not only a change in the arrangement
of verbs' arguments but also a change in the number of syntactically-realized arguments,
as shown in (4) in Chapter 1, repeated here below:
(3) a. Tom broke the vase.
b. The vase broke.
In addition to the fact that the Theme vase is post-verbal in (3a) but pre-verbal in (3b) in
relation to break, the transitive ergative predicate typically occurs with two arguments
(i.e., dyadic), whereas the intransitive ergative predicate occurs with a single argument
(i.e., monadic). In this respect, the ergative alternation is also characterizable as a
The ergative alternation is observed cross-linguistically (Nedyalkov and Silnitsky
1973, Haspelmath 1993) and has been extensively investigated in languages such as
English (Levin and Rappaport Hovav 1994, 1995), French (Labelle 1992), Japanese
(Okutsu 1967, Jacobsen 1992, Kageyama 1996), and Korean (Croft 1990), among others.
In this chapter, I first outline the ergative alternation in English and Japanese, focusing on
differences and similarities in the ways in which it is represented in both languages. Then
I go on to discuss unaccusativity with respect to alternatability of verbs. Finally, I discuss
the issue of direction of derivation regarding the Japanese ergative alternation.
2.2 English Ergative Pairs
2.2.1 Syntactic Aspects
It is well known that there are an enormous number of ergative verb pairs in
English (Langacker 1991:387). As noted in Chapter 1, the English ergative alternation is
expressed in a majority of cases by one single verb with no morphological change. Such
identity in form between a transitive use and an intransitive use results in the fact that the
distinction in transitivity regarding an ergative verb is discerned solely by virtue of the
context where each member occurs (Nedyalkov and Silnitsky 1973:3). More specifically,
only a shift in the valency of verbs' argument structure indicates the difference in
transitivity in English The transitive use of break is a case of dyadic valency, consisting
of an Agent and a Theme, while the intransitive use represents a monadic valency, taking
a Theme argument only. Typologically, alternations in which no changes in verb forms
occur between ergative transitives and intransitives are referred to as labile alternations.
Other languages that predominantly show this alternating pattern include Chinese,
German, German, and Greek (Teramura 1982:305, Haspelmath 1993:102).
In addition to the shift in valency just described, the ergative alternation needs to
meet the following two conditions. Firstly, the object or "internal argument" of the
transitive alternant should always be realized as the subject of the intransitive alternant
(Comrie 1985:322). As evident in (3) above, the predicate break and the argument vase
follow this constraint. The rationale for imposing such a constraint on the configurational
relation between a predicate and its internal NP argument is that it is necessary to
distinguish genuine ergative verb pairs from verbs which undergo the so-called
unspecified object alternation (Levin and Rappaport Hovav 1994:37).
(4) a. Tom smokes cigarettes.
b. Tom smokes.
At first glance, smoke seems to undergo the ergative alternation, given that it alternates in
transitivity without any morphological mediator, just like break. According to the
constraint noted above, however, the transitive and intransitive uses of smoke are not
eligible for an ergative pair in that the object (cigarettes) of the transitive use in (4a) is
not realized as the subject of the intransitive use; rather, it is syntactically unexpressed or
"unspecified" in the original object position.
Secondly, the object of the transitive alternant and the subject of the intransitive
alternant should have an identical thematic role, a thematic and syntactic correlation that
Langacker refers to as 'ergative pattern' (1991:387; cf Levin and Rappaport Hovav
1995:79). In a sense, this constraint is self-explanatory given that while Theme arguments
are syntactically realized differently, they are both base-generated as internal arguments.
Again the internal argument of break in (3) meets this requirement: The object (vase) of
the transitive use in (3a) and the subject of the intransitive in (3b) share the same Theme
role. This thematic relation accounts for why smoke in (4) fails to constitute an ergative
pair; the subject (Tom) of the intransitive construction is an Agent, while the object of the
transitive is a Theme. Given these observations, the ergative alternation would be
schematized as follows:
(5) a. NPi Vtr NP2 (transitive construction)
b. NP2 Vintr (intransitive construction)
In short, the process of ergative alternation will be summarized as (1) a shift in valency of
the arguments of the verb (2) parallelism between the subject of the transitive
construction and the object of the intransitive construction and (3) the preservation of the
thematic role assigned to the verb's internal argument.
2.2.2 Ergative Alternation and Other Similar Syntactic Operations
In the preceding section, we observed that the ergative alternation undergoes a
series of syntactic operations such as valency shift and configurational parallelism of
arguments. Note that such syntactic behaviors are not restricted to the ergative
alternation; rather, they are readily observable in other transformational operations. In
this section, we outline syntactic and semantic characteristics of passivization and middle
formation, demonstrating that in spite of striking similarities, the two syntactic
formations should be distinguished from the ergative alternation.
18.104.22.168 Ergatives and passives
Passivization is characterized in general as transforming a transitive verb into an
unergative (i.e., intransitive), along with advancement of the object of the transitive to the
subject of the intransitive (Perlmutter 1978:181ff; cf Shibatani 1985:822). Put more
informally, passivization involves a NP movement from a transitive internal argument
position into a passive external argument position whereby the syntactic subject of the
passive corresponds to the syntactic object of the transitive. Given this description,
Larson (1988:366) argues that passivization and ergative alternation share a crucial
parallelism in terms of NP movement, as illustrated below (t stands for "trace")
(6) a. The boat was sunk t.
b. The boat sank t.
I argue that the two syntactic processes should not be considered identical for the
following two reasons. Firstly, English passivization necessarily involves an overt,
regular morphological change in verbs, whereas the ergative alternation does not. Jaeggli
(1986) maintains that in passivization, the role of the passive suffix -en is seen as
'absorbing' the external theta-role which is originally associated with the external
argument of a transitive verb. In this respect, passivization is characterized as the
interaction of morphological and syntactic process, while ergative alternation is simply a
syntactic process in English.1
1 Not every scholar accepts Jaeggli's formulation of the passive. See Miller (1993:186-8)
for his pre-movement string analysis.
Secondly, the ergative alternation is distinct from passivization in that the latter
always implies that an Agent that brings about an event can be syntactically indicated in
passive structures (cf. Siewierska 1984:78, Roeper 1987:268, Haspelmath 1993:90,
Ackema and Schoorlemmer 1994:69). Evidence for this claim is observed in the fact that
passives can co-occur with a by-phrase, whereas ergative intransitives cannot.
(7) a. The boat was sunk by Bill.
b. *The boat sank by Bill. (Roeper 1987:268)
Even when no by-phrase is syntactically present, passives still imply that an Agent is
present implicitly. This is exemplified in the following:
(8) a. The ship was sunk to collect the insurance.
b. *The ship sank to collect the insurance. (Roeper 1987:268)
In short, that the passive be V-en can co-occur with a purpose infinitive clause as in (8a)
substantiates our view that passives invariably posit the presence of agentivity, whether it
is overtly expressed or not.
22.214.171.124 Ergatives and middles
Middle formation is a process of suppressing the subject of a transitive and
moving the object of the transitive into the subject position of a middle verb, as typically
illustrated below (Hoekstra and Roberts 1993:183-184):2
(9) a. Someone bribed the bureaucrats.
b. Bureaucrats bribe easily. (Keyser and Roeper 1984:381).
2 Ackema and Schoorlemmer (1995) claim that no movement takes place in middle
formation. In their terms, the grammatical subject in middles are base-generated in the
VP internal subject position.
(10) a. The butcher cuts the meat.
b. The meat cuts easily. (Levin 1993a:26)
In the examples, sentences (9b) and (10b) are middle constructions and the verbs (bribe
and cut) occurring in these constructions are referred to as middle verbs or simply
Numerous studies analyze middle formation in reference to ergative alternation
(Fiengo 1980, Keyser and Roeper 1984, Hale and Keyser 1987, Fagan 1988, 1992,
Condoravdi 1989, Fellbaum and Zribi-Hertz 1989, Levin 1993a, Fujita 1994, Kitazume
1996, Nakamura 1997). One syntactic characteristic of middle formation has to do with
the application of the syntactic rule Move ac to the internal argument of a transitive
predicate so that the argument is externalized in the middle construction, as illustrated in
(9) and (10) above. Furthermore, the thematic role of the internal argument remains the
same whether it is in the transitive construction or in the middle construction. Recall that
these are among the characteristics involved in the ergative alternation as well.3 Given
such similarities, Hale and Keyser (1986, 1987, 1988) maintain that middle formation
and ergative alternation are fundamentally identical, providing a single lexical rule for
both middle and ergative formations (cf. Fujita 1994:73):
(11) The Ergative-Middle Alternation
[x cause [y "undergo change"], (by...)]]
[y "undergo change", (by...)] (Hale and Keyser 1987:20)
3 Rapoport (1993:173-4) also points out a semantic similarity between ergatives and
middles. Just like ergatives, middles can be observed in verbs which lexicalize a change
of state (cf Chapter 3, Section 3.4.1).
According to Hale and Keyser, the only distinction between the ergative alternation and
the middle alternation is that the LCS of the latter contains a means clause as depicted by
the parenthesized by-phrase (see Chapter 3, Section 3.5 for further discussion of the
relationship between means clause and ergative alternatability).
In spite of the similarities just described, substantial evidence has been presented
that middles and ergatives are essentially distinct from each other for reasons that follow.
Firstly, as Keyser and Roeper (1984) note, middles are stative verbs, meaning that they
cannot describe events. Ergatives, on the other hand, are event-verbs. This contrast is
illustrated by the fact that middles cannot occur in progressive constructions, while
(12) a. The boat is sinking.
b. *Bureaucrats are bribing easily. (Keyser and Roeper 1984:385)
Keyser and Roeper also suggest that middles cannot occur in situations which describe
particular events in time, whereas ergatives can. This is evident in the following example
where the ergative sink occurs in the past tense, while the middle bribe does not.
(13) a. The boat sank in less than an hour.
b. ?Yesterday, the mayor bribed easily, according to the newspaper.
(Keyser and Roeper 1984:384)
Keyser and Roeper observe that (13b) is marginally grammatical since middles only
describe situations which are generic or "held to be generally true" (1984:384).
It has been noted in the literature that while ergatives imply no Agent, middles
always entail Agent implicitly (Fiengo 1980:57, Keyser and Roeperl984:404-405, Hale
and Keyser 1986:15-16, 1987:18, Condoravdi 1989, Nakamura 1997:119, Fujita
1994:87). By "implicit Agent" it is meant that the middle construction always implies an
Agent in the eventuality denoted by the middle verb but the thematic role of Agent can
have no phonetic or structural realization. 4 Put differently, Agent is present semantically
but absent syntactically in the middle construction (Nakamura 1997:123; see also
Ackema and Schoorlemmer 1994:69).5 One way of illustrating that middles express the
existence of Agent implicitly but not syntactically is to see if they may co-occur with
agent-oriented adjuncts like a by-phrase or a to-infinitive phrase. The ungrammaticality
of the following sentences points to the validity of this statement.6
(14) a. *The official bribes easily by managers. (Keyser and Roeper 1984:406)
b. *Bureaucrats bribe easily to keep them happy.
(Keyser and Roeper 1984:407)
In short, the presence of an Agent at the underlying semantic level of middles may
crucially differentiate middles from ergative intransitives, which are assumed to involve
4 Miller (1993:178) suggests that the implicit argument in middles may well be
interpreted as a Benefactive rather than Agent.
5 Stroik (1992:131) argues that the Agent role in middles can be expressed syntactically,
mainly in an adjunct position as illustrated below:
(i) a. That book read quickly for Mary.
b. No Latin text translates easilyfor Bill.
6 Keyser and Roeper (1984:407) attribute the ungrammaticality to the notion of "control."
That is, the implicit Agent of middle cannot control the lower clause, making the whole
sentence ungrammatical, while the implicit Agent in passive as in (i) below, which is
optionally expressible (i.e., by Bill), controls the lower clause, making the whole sentence
grammatical (Keyser and Roeper 1984:407).
(i) The bureaucrats were bribed (by Bill) to keep them happy.
Given the optionality of by-phrase in passives, Keyser and Roeper consider the notion
"implicit" to have a different implication for middles and passives, respectively
no identifiable external Agent (Pinker 1989:130; cf. Marantz 1984:180; see Chapter 3 for
more detailed discussion of agentivity and ergative verbs).
In summary, we observed that although the ergative alternation resembles
passivization and middle formation in a number of ways, the former is crucially distinct
from the latter in terms of the presence or/and implication of agentivity. Accordingly, the
ergative alternation investigated in the present study exclusively refers to the pattern
schematized in (5) above.
2.2.2 Semantic Aspects
Another aspect that needs to be analyzed regarding ergative pairs is a semantic
distinction between transitive and ergative members. The semantic relation between
ergative pairs has often been represented through the schemata called Lexical Conceptual
Structure (LCS) in the literature (Guerssel et al. 1985, Hale and Keyser 1987, Levin and
Rappaport Hovav 1995). The LCSs of break, for instance, are typically represented as
(15) a. ergative transitive break: x CAUSE [y BECOME broken]
b. ergative intransitive break: y BECOME broken
What is schematically striking about the LCS of the transitive alternant is its complex
structure, consisting of a causing subevent and a central subevent (Hale and Keyser
One apparent semantic property that distinguishes ergative transitives from
ergative intransitives is the presence of CAUSE in the transitive LCS, which is, in a less
technical schematization, paraphrased into 'cause to V-intransitive' (Parsons 1990:106).7
The representation in (15) indicates that the Agent argument x is associated with CAUSE,
whereas the Theme argument y is associated with the state "broken." Given that CAUSE is
not present in the ergative intransitive, uncertainty remains as to how the event denoted
by break takes place. Most researchers assume that the events denoted by ergative
intransitives occur with no intervention of an Agent (Haspelmath 1993, Matsumoto
2000a). In other words, they are perceived to occur spontaneously.8 Thus, when we say
The vase broke in English, the vase just broke on its own accord, meaning that there is no
Agent involved in the event.
The view that the ergative intransitive entails no Agent appears to raise a problem.
Consider the following sentence:
(16) Tom hit the vase with a bat, and it broke.
Evidently, the example indicates that the breaking of the vase did not occur
spontaneously; instead, it is Tom, namely an Agent, who brought about the event.
Evidence like (16) seems to suggest that it is necessary to recognize an Agent at the
semantic level of the ergative intransitive. Levin and Rappaport Hovav(1995) assume
that Agent or Cause is indeed involved in the event denoted by the ergative intransitive.
The argument carrying this thematic role is simply invisible syntactically, according to
Levin and Rappaport Hovav, because it is unspecified in surface constructions. By
7 In light of the presence of CAUSE in the transitive LCS, Pinker (1989) and Levin and
Rappaport Hovav (1994, 1995) refer to the alternation typified by break as 'causative
8 Due to this spontaneous nature of ergative intransitives, they are often referred to as
underspecification Levin and Rappaport Hovav mean that ergative intransitive
constructions can imply the presence of an Agent or a Cause. It is simply that they do not
identify the type of the Agent or the Cause (cf. Davidse 1992:109). The unspecification
of Agent or Cause characterizing the ergative intransitive also suggests the verb does not
lexicalize agentivity or cause. It should be noted here that underlying the claim by Levin
and Rappaport Hovav is their view that our real world knowledge makes it difficult to
imagine that events like the breaking of a vase would occur without an external cause
(1995:93). Their agentive or causative analysis of ergative intransitive is reflected in part
in the single causative lexical semantic representation of break (cf Chapter 1) in which
ergative intransitives are derived from causative transitives through the process of
Nevertheless, I claim, for the reasons that follow, that an argument can be made
for the spontaneity of the ergative intransitive. Firstly, returning to (16), the event (Tom
hit the vase i/ i/h a bat) leading to the rupture of the vase is not essential to the ergative
intransitive break; in other words, it is not a fundamental component of the LCS of
ergative intransitive break. It follows, therefore, that the Agent Tom participating in the
event of hitting the vase has no essential involvement in the resultant state of the vase.
Secondly, as Levin and Rappaport Hovav (1995:107) note, even if it is claimed that break
in (16) involves an Agent, the Agent is interpreted, at best, as a passive participant. Based
on Levin and Rappaport Hovav' s view, I assume that the passive nature of a participant
still enables us to conceive the breaking of the vase in (16) as occurring spontaneously
9 Researchers such as Oosten (1977), Levin and Rappaport Hovav(1995), and Kageyama
(1996) hypothesize that Theme is identified with Agent or Cause in ergative intransitive
(for relationship between passive participant and unspecification of causing event, see
Chapter 3, Section 126.96.36.199; see also Levin and Rappaport Hovav 1995:107).
2.3 Japanese Ergative Pairs
2.3.1 Syntactic Characteristics
Ergative pairs are also commonly observed in Japanese. As illustrated above, the
English ergative alternation involves the parallelism between the object of a transitive
predicate and the subject of an ergative predicate. The same configurational relationship
also holds for Japanese ergative pairs where the subject-object correlation is indicated by
case-marking rather than word order (Okutsu 1967:49, Miyajima 1972:684, Hayatsu
1987:81, Mitsunobu 1992:85).
(17) a. Sono otokonoko-ga mado-o wat-ta.
the boy-NOM window-ACC break-PAST
'The little boy broke the window'
b. Mado-ga ware-ta.
'The window broke'
Note that the object mado in (17a), which is marked with the accusative marker -o, is
realized as the subject with the nominative marker -ga in (17b), forming the object-
subject relationship characteristic of the ergative alternation. 10 In addition, the NP mado
10 The correlation between the accusative case marker -o and the transitivity of verbs has
been discussed by scholars for more than a century. Otsuki (1897), who was strongly
influenced by western grammar, first suggested that transitive verbs should take -o as the
accusative marker. Matsushita (1923:18) further developed Otsuki's view, maintaining
that all verbs that occur with -o should be considered transitive (cf Nomura 1982, Morita
1994, Suga and Hayatsu 1995). Thus, in Matsushita's view, verbs like ik- 'go' and hasir-
'run,' which are usually viewed as intransitive when occurring optionally with the
directional particle -e or -ni 'to,' were considered to be transitive verbs when occurring
(i) Waga miti-o iku
I-GEN way-ACC go
bears the same thematic role ("Theme") in both constructions, following another criterion
on whether a verb is considered to participate in the ergative alternation Based on the
correspondence between (17a) and (17b), Okutsu (1967) provides a configurational
template for the Japanese ergative alternation, which is fundamentally identical to the
English corresponding template given in (15) above (cf Suga 1981:122, Jacobsen
(18) NPi-ga NP2-o Vtr
On Okutsu's view, the subject (NP1) of a transitive predicate is deleted when a verb
undergoes intransitivization and added when a verb undergoes transitivization.
2.3.2 Morphological Characteristics
One key difference between Japanese and English ergative alternating verbs is
that the former overtly exhibits a shift in transitivity by means of derivational
morphology, as exemplified earlier and repeated below. 1
'I will go my own way'
(ii) Yamamiti-o hasit-ta
mountain path-ACC run-PAST
'I ran along the mountain path'
While basically following Matsushita's proposal, Okutsu (1967) maintains that the verbs
in (i)-(ii), which Okutsu subsumes under the category called motion verbs (idoo-doosi),
should be regarded as intransitive since the case marker -o occurring with those verbs is
not an accusative case marker but a "directional particle."
11 A small number of verbs such as hirak- 'open' and mas- 'increase,' which Martin
(1975) refers to as "ambivalent pairs," show no morphological distinctionbetween
intransitive and transitive uses (cf. Morita 1994:168-170). There are more verbs among
Sino-Japanese verbs (cf Jacobsen 1992), a type of complex verbs made up of Chinese
stems and Japanese native verbal suffix -suru (e.g., idoo-suru 'move' and shuuryoo-suru
'end'), which are identical in form regardless of their transitivity. For these ambivalent
pairs, transitivity can only be denoted configurationally.
(19) TRANSITIVE INTRANSITIVE
Japanese English Japanese English
war- break ware- break
sizume- sink sizum- sink
mage- bend magar- bend
tokas- melt toke- melt
yak- bake yake- bake
In other words, while English represents a change in transitivity by syntactic means
alone, Japanese marks the same process morphologically as well as syntactically
(Teramura 1982:305). 12 While the number varies from scholar to scholar, there are
numerous morphologically-related ergative pairs in Japanese. While Jacobsen(1992)
provides a list of about 371 verbs, Hayatsu (1989) identifies almost 600 ergative pairs in
the Japanese lexicon.
It has been noticed that morphological distinctions as observed in (19) are not
inconsistent. Rather, most of the ergative pairs in Japanese are found to follow certain
regular derivational patterns, which are fairly discernible and are limited in number. In
view of this fact, one of the main goals of inve stigating Japanese ergative pairs has been
to identify the derivational patterns and classify them accordingly (cf Shimada 1979).
12 The morphological opposition characterizing Japanese ergative pairs date back to Old
Japanese (cf Shimada 1979, Kageyama 1996:179-180, Komatsu 1999). While several
derivational suffixes have undergone phonetic changes over time, the changes are highly
consistent. For instance, many of the intransitive members with the ending -e(r) paired
with the transitive -(y)as- (Jacobsen's Class IX) used to be spelled -yu. Thus, the modem
Japanese ergative pair hier-/lhoii)- was hIn- hlnyi in Old Japanese. Komatsu (1999:101)
notes that the -yu intransitives implied that events occur spontaneously, whereas the -
(y)as- transitives indicated that events are brought about by intentional agents. Kageyama
(1996) speculates that ergative pairs in Old Japanese might have held closer links
between affixal forms and their semantic functions (cf. Dubinsky 1985:259). According
to Kageyama, the pair tumor-/tum-, which we considered to have opaque semantic links
(Chapter 4), might have had a closer semantic correlation.
Shibatani (1990), for instance, provides a five-group classification of ergative pairs as
itam- 'be damaged'
are- 'be mined'
okure- 'be late'
ta(y)e- 'be extinct'
hage- 'tear off
ore- 'be broken'
One question that classifications like those above inevitably pose is whether
particular suffix forms reflect verbs' transitivity; in other words, is it possible to predict
13 Most recently, Jacobsen (1992) proposes a sixteen group of ergative pairs in Japanese
(see Appendix; cf. Chapter 4).
14 Morita (1994:166) and Kageyama (1996:180-1) note that the pair hage-/hag- is made
more complicated due to another semantic and morphologically similar pair hagare-
/1hug,'t- 'tear off' Other multiple pairs like hage-, hagare-/hag-, hagas- include tizim-,
tizimar-/tizimer-, tiziras- 'shrinkvi/shrinkvt,' usure-, usurag-, usumar-/usumer-
nakas- 'make cry'
hag- 'tear off'14
merely by suffixal form whether each member of a given pair is transitive or intransitive?
If this is the case, then it follows that the significance of syntactic configuration will be
enormously diminished in dealing with the ergative alternation in Japanese. The
classification in (20) seems to suggest that there is such a relationship between suffixal
form and transitivity. That is, the suffixes -ar in Group (20a) and -as in Group (20c, d)
are constantly associated with intransitive and transitive, respectively.
The correspondence between suffix form and transitivity just described does not
always hold for all groups of Japanese ergative pairs. As pointed out by Okutsu (1967),
Jacobsen (1992), and among others, the suffix -e exhibits a conflicting behavior in terms
of transitivity. That is, it functions as a transitivizer in Group (20b), and as an
intransitivizer in Group (20d). Such conflicting functions associated with-e naturally
make the transitivity of a given verb containing the suffix unpredictable. Jacobsen (1992)
exemplifies this point by providing a hypothetical Japanese verb pair harik- and harike-.
Given that -e can be either a transitivizing or an intransitivizing suffix, there is simply no
telling which form of the pair is transitive. In fact, the inconsistent behavior of the suffix
seems to have puzzled researchers who attempted to prove an inherent relationship
between suffix form and transitivity.15 At the very least, the dual functions of the suffix -e
suggest that the transitivity of morphological pairs is not completely predictable from the
form of a derivational suffix.16 This in turn suggests that the syntactic configuration in
15 Kitagawa and Fujii (1999), while emphasizing the regularity of the derivational process
of morphological pairs in most of Jacobsen's classes, had to adopt a semantic approach to
the morphological pairs in Group (b) and (d) in order to elucidate their elusive behaviors.
16 The different morpho-semantic behaviors observable between -as and -e might be
better explained under Aronoff s view of structural transparency and semantic coherence
(1976:20-21; cf. Tyler 1999:80). In keeping with his terminology, it might be argued that
(18) above still plays an important role in determining the transitivity of ergative pairs in
Additionally, this research project raises a question of semantic coherence
between morphologically-related verb pairs. Specifically, I propose to attempt to
determine whether or not morphological opposition reflect changes in the semantics of
the pairs. As Chapter 4 will demonstrate, numerous modem Japanese ergative pairs hold
little or no semantic affinity. This indicates that shared verb stems between ergative pairs
do not necessarily mean that the pairs are semantically correlated. As Ichihashi (1992:18-
19) points out, ergative pairs like ak- 'open' vs. akas- 'reveal' and bake- 'disguise
oneself vs. bakas- 'bewitch' are hardly considered semantically related. I will
demonstrate later that such semantically tenuous pairs are ubiquitous in the Japanese
ergative pairs. Given this fact, the present study focuses mainly on Japanese ergative
pairs whose shared verb stems hold semantic affinity.
To summarize, I will refer to morphologically-related Japanese verb pairs as
ergative pairs only when they (1) share a common root, (2) hold very close semantic
affinities, and (3) have the same noun phrase as the subject in the intransitive
construction and the object in the transitive construction (cf. Hayatsu 1987, Ichihashi
transitive verbs containing -as are not structurally transparent. In other words, there is no
rule for determining which variant should be attached to its verb stem, but these variants
are considered semantically transparent because they are always associated with
transitive meanings. In contrast, verbs containing a non-variant suffix -e might be said to
be structurally transparent. They are not, however, semantically coherent because no
constant semantic association can be attributed to this suffix.
2.4 Ergative Alternation: Issues
Several issues have been raised surrounding the ergative alternation (cf. Levin
1985:18). One has to do with whether the ergative alternation involves a lexical rule or
syntactic rule (Keyser and Roeper 1984; cf. Chapter 4, Section 4.3). Another question
that needs to be addressed is whether verbs which alternate in transitivity consist of a
single lexical entry or two separate lexical entries (Pinker 1989:71-72). Furthermore, the
direction of derivation involving the ergative alternation has also been discussed in the
literature (Comrie 1985, Croft 1990, Dixon 1991, Jacobsen 1985, Haspelmath 1993).
While those issues call for further investigation, the present chapter is more
concerned with the obvious fact that not all verbs participate in the ergative alternation.
More specifically, there are intransitive verbs which do not occur in transitive
constructions and transitive verbs which do not occur in intransitive constructions (Pinker
1989:130; cf Ritter and Rosen 1998). As mentioned previously, the intransitive verb
laugh cannot be used transitively in English.
(21) a. Tom laughed.
b. *Bill laughed Tom (on the interpretation "Bill caused Tom to laugh")
Along the same lines, the transitive verb destroy lacks its anti-causative, intransitive
(22) a. The storm destroyed the house.
b. *The house destroyed. (on the interpretation "The house came to the state
of being destroyed")
The most common way to grammatically achieve the interpretation as instantiated within
the parenthesis in (22b) is to use a passive construction. 17
(23) The house was destroyed.
Given (22) and (23), the obvious question is why verbs vary regarding their
alternatability. As will be discussed in Chapter 3, I will follow Wasaw (1985), Pinker
(1989), and Levin and Rappaport Hovav (1995), among others, in suggesting that the
syntactic behaviors of verbs are to a large extent dependent on the semantics of the verbs.
Moreover, given that verbs that participate in the ergative alternation belong to the same
semantic classes across languages (Levin 1985:22), it is important to conduct a thorough
analysis of languages other than English to corroborate this admittedly expansive
generalization My ultimate goal is to find whether the semantics-syntax approach
proposed for the English ergative pairs can be adapted to Japanese ergative pairs.
It has been noted in the literature that unaccusativity plays an important part in the
ergative alternation (Levin and Rappaport Hovav 1995, Arad 1998). The effect of
unaccusativity on the ergative alternation is discerned most noticeably in the fact the
intransitive members of ergative pairs are usually identified as "unaccusative" (Borer and
Wexler 1987:158, Tsujimura 1990b:935, 1996:323-4, Levin and Rappaport Hovav,
1995:80, Ono 1997:168).18 This seems to suggest that a close examination of
17 D. Gary Miller (p.c.) pointed out to me that The house self-destructed would be another
way to express the same meaning inchoatively.
18 In keeping with Perlmutter (1978) and others, I utilize the term unaccusative in the
current section regardless of whether unaccusative verbs alternate with transitive uses.
unaccusativity may help us better understand the mechanism of the ergative alternation.
In this section I first review the syntactic and semantic characteristics of unaccusatives.
And then I discuss how the semantic properties associated with unaccusativity may
account for the ergative alternation system in English and Japanese.
2.5.1 The Unaccusativity Hypothesis
In 1978 Perlmutter proposed that intransitive verbs do not constitute a
homogeneous group but instead consist of two distinct types of verbs: unergative and
unaccusative verbs. Under the theory of the Unaccusative Hypothesis, Perlmutter
maintained that the arguments that are subjects of unaccusative verbs such as arrive and
fall are in fact objects at the initial level of representation. On a Government-Binding
approach, unaccusative verbs are assumed to generate their sole argument as the direct
object in D-structure. Following Burzio's generalization that the unaccusative verbs fail
to assign case to their interml argument in D-structure (Burzio 1986, cf Grimshaw 1987),
Tom in (24) is considered an argument without case.
(24) Unaccusative Verb: arrive
0 amve Tom
For the purpose of acquiring a subject case, Tom moves to the position external to VP,
which is unoccupied in the underlying structure.
Tomi amve ti
On the other hand, unergative verbs such as run or laugh have their argument in the
subject position at both levels (i.e. D- and S-structures). As a consequence, the type of
movement as illustrated in (25) does not occur to the unergative verb run in (26) since the
NP Tom originates in the subject position in D-structure:
(26) Unergative Verb: run
Tom ran (Spencer 1991:260)
The split intransitivityjust described is observable across languages. In particular, for the
purpose of demonstrating the universality of unaccusativity, a number of unaccusative
diagnostics have been proposed in different languages. 19 In what follows, we look at how
unaccusativity is represented in English and Japanese.
2.5.2 Unaccusative Verbs in English
As discussed in the preceding section, the distinction between unaccusativity and
unergativity is summarized broadly as the former having an underlying object and the
latter having an underlying subject (cf. Levin and Rappaport Hovav 1995:3). In this
section we examine English split intransitivity, focusing on diagnostics which reflect the
differing underlying argument structures of unaccusativity and unergativity.
In keeping with the view that unaccusativity is observed cross-linguistically,
several diagnostics have been proposed in English (L. Levin 1988, Levin and Rappaport
Hovav 1992a). In particular, the resultative construction is considered in the literature to
be a diagnostic which provides evidence that the surface subject of the unaccusative
verbs is in fact the underlying object (Simpson 1983, Levin and Rappaport 1989,
Kageyama 1996). It is well known that unaccusative verbs can occur in the resultative
construction, whereas unergative verbs do not, as illustrated below:
(27) a. The vase broke into pieces.
b. *Tom talked hoarse. (on the interpretation 'After Tom talked too much,
his voice became hoarse.')
Sentence (27a) is grammatical in that the resultative attribute (into pieces) is predicated
of the subject (the vase). In contrast, sentence (27b) is ungrammatical in that the attribute
hoarse cannot form a resultative relationship with the subject Tom. The only way
sentence (27b) can be grammatical is when the attribute hoarse is not resultative but
19 For instance, Impersonal Passivization in Dutch (Perlmutter 1978) and Ne-cliticization
in Italian (Burzio 1986) are such diagnostics.
descriptive on the interpretation of 'Tom talked, while his voice was hoarse.' Given the
contrast in (27), Simpson (1983:146) argues that:
The controller of a resultative attribute must be an OBJECT, whether that object
is a surface OBJECT, as in transitive verbs, or an underlying OBJECT, as in
passives and intransitive verbs of the Unaccusative class.
What is noticeable about Simpson's view is that the parallelism between a controller and
an OBJECT agrees with the view that the subject of an unaccusative verb is invariably
the underlying object. Thus, the subject vase in (27a) is considered to originate in the
internal position where it serves as a controller juxtaposed to the attribute (into pieces).
The underlying syntactic configuration between vase and the verb break will be indirectly
illustrated by the transitive construction involving the verb break, as illustrated in (28)
(28) Tom broke the vase into pieces.
Another unaccusative diagnostic that appear to point to the underlying structure of
the unaccusativity verb is adjective passive formation (Grimshaw 1987:245, Levin and
Rappaport 1986, 1989:327).
(29) a. fallen leaves
(cf Leaves fell on to the ground)
b. the stolen car
(cf The car was stolen)
c. *the walked man
(cf The man walked along the beach)
Levin and Rappaport (1989:327) point out the correspondence between unaccusative
subjects and transitive objects, maintaining that adjectival passives are predicated of both
unaccusative subjects and transitive objects, but not of unergative and transitive subjects.
2.5.3 Unaccusative Verbs in Japanese
It is maintained that as in English, sentences containing unaccusative verbs
involve movement of an internal argument in Japanese (Miyagawa 1989a:23; cf.
Kageyama 1993:46). As described earlier, the internal argument of an English
unaccusative verb, which occurs within a VP, has its thematic role Theme assigned
directly by the verb. Since the verb fails to assign case to any internal argument, the NP
moves to the position external to the VP in order to acquire a subject case via inflection.
The same is true of Japanese unaccusatives as well, except for the fact that in Japane se
the moved NP is overly marked with the case marker -ga. The template of Japanese
unaccusativity is diagrammed as follows (Miyagawa 1989a:85):
NP, ga VP
In (30), the NP is first assigned the thematic role Theme by an unaccusative verb (V).
Since the V assigns no case (Miyagawa 1989a:89), the NP moves to the subject position
to receive case, leaving a trace behind. As will be discussed in more detail later, it is
important to ensure that the NP in (30) has already received its thematic role Theme from
the verb inside the VP prior to the movement, since the case marker -ga is incapable of
assigning a thematic role
Evidence that unaccusativity also exists in Japanese has been presented in the
literature (Tsujimura 1991, Miyagawa 1989a, Kageyama 1996). At least two
unaccusative diagnostics appear to have direct relevance to Japanese ergative pairs. The
first diagnostic concerns Numeral Quantifiers (NQ) and the constraints on their
correlation with NP referents. The Japanese NQ consists of a numeral and a classifier
(CL; e.g., -satu for 'book,' -nin for 'person,' etc.), usually placed after an NP that is
(31) a. Taroo-wa hon-o ni-satu kat-ta.
Taro-TOP book-ACC two-CL buy-PAST
'Taro bought two books.'
b. Tomodati-ga san-nin uti-ni ki-ta.
Friend-NOM three-CL (my) house-to come-PAST
'Three friends came to my house.'
Miyagawa (1989a) postulates that the NP and the NQ constitute syntactically separate
phrases; in other words, the NQ is neither a specifier nor a complement of the NP.
Furthermore, the NQ is not an argument of the verb in that the verb does not assign a
thematic role to it. Rather, the relation between the NP and the NQ is more like that of
predication in which the NQ is predicated of the NP (Miyagawa 1989a:22). In this
respect, the NQ is analogous to a small clause like raw in the following sentence, where
raw is predicated of its antecedent meat (cf Williams 1980).
(32) John ate the meat raw. (Miyagawa 1989a:22)
In order to form a proper predication relationship, the NQ and the NP should always be in
a mutual c-command relation (Miyagawa 1989a:27ff). This is illustrated in the following
(33) a. Tomodati-ga 2-ri Sinzyuku-de Tanaka-sensei-ni at-ta
friend-NOM 2-CL Shinjuku-in Prof Tanaka-DAT meet-PAST
'Two friends met Professor Tanaka in Shinjuku.'
b. *Tomodati-no kuruma-ga 3-nin kosyoo-sita
friend-GEN car-NOM 3-CL broke down
'Three friends' car broke down' (Miyagawa 1989a:29)
In sentence (33a), the NQ 2-ri 'two' and its referent NP tomodati 'friend' are in a mutual
c-command relation outside the VP. The relation is diagrammed as below(Miyagawa
NP NQ VP
tomodati-ga 2-ri Tanaka sensei-ni at-ta
By contrast, tomodati and san-nin in sentence (33b) do not form a mutual c-command
relationship because tomodati fails to c-command 3-nin. This is illustrated in the
following tree diagram (Miyagawa 1989a:30):
NP NQ VP
tomodati-no kuruma-ga kosyoo-sita
It is interesting to note, however, that there are some cases in which the mutual c-
command does not seem to obtain for the relation between an NQ and its referent in
Japanese. For instance, the following example shows that the NP doa 'door' and the NQ
huta-tu 'two' are not in a mutual c-command relation because the VP (kono kagi-de huta-
tu ai-ta) that dominates the NQ does not dominate the NP. Nevertheless, sentence (36) is
(36) Doa-ga [vp kono kagi-de huta-tu ai-ta]. (< ak- 'open' + ta)
door-NOM this key-with two-CL open-PAST
'Two doors opened with this key.' (Miyagawa 1989a:43)
Miyagawa explains the grammaticality of (36) by hypothesizing that the NP doa 'door'
originates inside the VP and leaves a trace after moving outside the VP. It is this trace,
according to Miyagawa, that enables the NP doa and the NQ huta-tu 'two' to form a
mutual c-command relation. The tree diagram of this derivation is illustrated as follows
do i-ga PP NP 1NQ
kN P i j huta-tu ai-ta
kono kagi e
The grammaticality of sentence (36) implies that the verb ak- 'open' is an unaccusative
verb. Such mutual c-command relationship between a NP trace and an NQ does not
obtain for unergative verbs like hasir- 'run' as diagramed below (Tsujimura 1996:273-4):
(38) *Gakusei-ga [vp awatete inu-to sannin hasit-ta].
student-NOM hurriedly dog-with 3-CL run-PAST
'Three students ran hurriedly with the dog.'
Adv PP NQ V
awatete sannin hasit-ta
In (38), the subject NP gakusei 'student' originates and remains in the position external to
the VP, preventing itself from forming a mutual c-command relation with the referent NQ
Resultative constructions also serve as a diagnostic of Japanese unaccusativity
(Tsujimura 1991). As with English, the controllers of resultative attributes can be
predicated of the subjects of unaccusative verbs as well as the objects of transitive verbs
in Japanese (Tsujimura 1991:97).
(39) Hanako-no kami-gai [vp ti nagaku nobi-ta].
Hanako-GEN hair-NOM long grow-PAST
'Hanako' s hair grew long'
The subject NP kami 'hair' in (39), which is marked so with -ga, is linked to its trace
inside the VP. Thus, it follows from the generalization given above (Section 2.5.3) that
the resultative attribute nagaku 'long' is predicated of the underlying OBJECT kami. On
the other hand, as expected from the behavior of the English verb talk as described above,
the Japanese equivalent hanas- 'talk' does not occur in resultative constructions.
(40) *Taroo-wa [p kutakuta-ni hanasi-ta].
Taro-TOP exhausted talk-PAST
'Taro talked exhausted'
Since the subject (Taro) is a base-generated subject, it violates Simpson's requirement
that the controller should be OBJECT. Thus, sentence (40) proves to be ungrammatical.
Based on the unaccusative diagnostics just described, given below are Japanese
intransitive verbs which are identified as unaccusative in the literature (Miyagawa
1989a:43, 97-99, Tsujimura 1990a:283, 1990c:264, 1996:276, 323).20
(41) simar- 'close' ak- 'open'
koware- 'broke' taore- 'fall'
katamar- 'solidify' ku- 'come'
hair- 'enter' agar- 'go up, rise'
tuk- 'arrive' kuzure- 'collapse'
kire- 'be cut' more- 'leak'
toke- 'melt' tamar- 'accumulate'
ware- 'break' suber- 'slide'
kie- 'turn off ukab- 'float'
tuk- 'turn on' de- 'come out'
ik- 'go' oti- 'fall'
kaer- 'return' tat- 'depart'
korogar- 'roll' hazum- 'bounce'
To summarize this section, we have observed that a range of diagnostics both in
English and Japanese has attested to the view of split intransitivity. Note, however, that
these diagnostics in essence reflect the fact that the distinction between unaccusative and
unergative is syntactically coded (cf Levin and Rappaport 1989:9). In other words, these
diagnostics do not tell us what semantic aspects, if any, of verbs result in such different
syntactic behaviors. As will be discussed below, Perlmutter (1978:161) had already
suggested that the distinction between unaccusative and unergative can be explained
semantically (for further work on semantic implications of unaccusativity, see Levin
20 Kishimoto (1996:264-5) gives the following additional list of unaccusative verbs:
obore- 'be drowned,' sin- 'die,' umare- 'be born,' nemur- 'sleep,' tissoku-suru 'smother,'
hurue- 'tremble,' kuru- 'go wrong,' kom- 'be crowded,' moe- 'burn,' yowar- 'weaken,'
katamuk- 'lean,' mahi-suru 'paralyze,' nak- 'cry,' korob- 'fall down,' kumor- 'get
cloudy,' naor- 'heal.'
1985). In the section that follows, I will conduct a semantic analysis of split intransitivity,
suggesting that the unaccusative/unergative distinction is characterizable semantically as
well as syntactically.
2.5.4 Semantic Characterization of Unaccusativity
In the preceding section, I have demonstrated that the subjects of unaccusative
verbs originate in the object position at the underlying level both in English and Japanese.
In so doing, I outlined several diagnostics which are proven to attest to split intransitivity.
One difficulty, however, with the analysis of unaccusativity on the basis of such
diagnostics has to do with the fact that unaccusative diagnostics in general depict the core
structure of unaccusative verbs (e.g., [collapse tent]) in indirect ways. In other words, all
these diagnostics deal with deep unaccusativity, where the argument appears only in the
subject position (cf. Bresnan and Zaenen 1990).
There are, however, some diagnostics that seem to point to surface unaccusativity
in English (Levin and Rappaport Hovav 1995:19):
(42) a. There appeared a ship on the horizon
b. Into the room came a man
Sentences (42a-b) are examples of there-insertion construction and locative inversion,
respectively (L. Levin 1988, Hoekstra and Mulder 1990). Note that both diagnostics
overtly indicate the original position of the internal arguments (ship and Tom) of the
unaccusative verbs (appear and came). Given (42), one may well claim that
unaccusativity is indeed a syntactic phenomenon, which makes no semantic account
necessary (cf. C. Rosen 1984). The problem with the surface unaccusative diagnostics,
however, is that they are applicable to a limited number of unaccusative verbs. For
instance, the ergative intransitives melt permits neither there-insertion construction nor
(43) a. *There melted lots of ice that morning. (L. Levin 1988:23)
b. *On the streets of Chicago melted a lot of snow.
(Levin and Rappaport Hovav 1995:224)
Another problem with unaccusative diagnostics has to do with inconsistency with
respect to the selection of unaccusative and unergative verbs. Levin and Rappaport
Hovav (1995:13) suggest that different unaccusative diagnostics may bring about a
conflicting result as to whether a given verb is unergative or unaccusative. For instance,
as discussed by Borer (1998:61), the Dutch unaccusative verb vallen 'fall' can occur with
the unaccusative diagnostic impersonal passive if the verb is provided with the additional
meaning of "intention" (see Levin and Rappaport 1989 for further discussion of
In view of the syntactic uncertainty involving unaccusativity just described, Van
Valin (1990) argues that the unaccusative-unergativity distinction is better explained
semantically (cf. Tsujimura 1994a, Kishimoto 1996). In fact, Perlmutter had initially
suggested the possibility of explaining the unaccusative-unergative distinction
semantically, stating that "[I]nitial unergativity vs. unaccusativity is predictable from the
semantics of the clause" (1978:161; cf Van Valin 1990:221). In 1984, Perlmutter and
Postal formulate the Universal Alignment Hypothesis, assuming that semantically-
characterizable verbs of unaccusative type in one language are also unaccusative in all
other languages (cf. Mulder and Wehrmann 1989:117):
(44) Universal Alignment H)plathe'i'
There exist principles of universal grammar which predict the initial
relation borne by each nominal in a given clause from the meaning of the
(Perlmutter and Postal 1984:97)
Building on the hypothesis, Perlmutter and Postal categorize English unergative and
unaccusative verbs semantically as follows (1984:98-99):
(45) Unergative verbs
a. Predicates describing willed or volitional acts:
work, play, speak, talk, smile, grin, frown, grimace, think, meditate,
cogitate, daydream, skate, ski, swim, hunt, bicycle, walk, skip (voluntary),
jog, quarrel, fight, wrestle, box, agree, disagree, knock, bang, hammer,
pray, weep, cry, kneel, bow, courtesy, genuflect, cheat, lie (tell a
falsehood), study, whistle (voluntary), laugh, dance, crawl, etc.
whisper, shout, mumble, grumble, growl, bellow, blurt out, etc.
Predicates describing sounds made by animals:
bark, neigh, whinny, quack, roar (voluntary), chirp, oink, meow, etc.
b. Certain involuntary bodily process21
cough, sneeze, hiccup, belch, burp, vomit, defecate, urinate, sleep, cry,
(46) Unaccusative verbs
a. Predicates expressed by adjectives in English:
a very large class, including predicates describing sizes, shapes, weights,
colors, smells, states of mind, etc.
b. Predicates whose initial unclear term is semantically a Patient:
burn, fall, drop, sink, float, slide, slip, soar, flow, ooze, seep, trickle, drip,
gush, hang, sway, wave, tremble, shake, languish, thrive, drown, stumble,
roll, succumb, dry, blow away, boil, seethe, lie (involuntary), sit
21 If bodily processes are perceived as involuntary, they might well be unaccusative (cf
melt, freeze, evaporate, vaporize, solidify, crystallize, dim, brighten,
redden, darken, yellow, rot, decompose, germinate, sprout, bud, wilt,
wither, increase, decrease, reduce, grow, collapse, dissolve, disintegrate,
die, perish, choke, suffocate, blush, open, close, break, shatter, crumble,
crack, split, burst, explode, burn up, burn down, dry up, dry out, scatter,
disperse, fill, vanish, disappear, etc.
c. Predicates of existing and happening:
exist, happen, transpire, occur, take place, and various inchoatives such as
arise, ensue, result, show up, end up, turn up, pop up, vanish, disappear, etc.
d. Involuntary emission of stimuli that impinge on the senses:
shine, spark, glitter, glisten, glow, jingle, clink, clang, snap (involuntary),
crackle, pop, smell, stink, etc.
e. Aspectual predicates:
begin, start, stop, cease, continue, end, etc.
last, remain, stay, survive, etc.
In general, unergative verbs appear to reveal a unified, constant characteristic: a majority
of them, with the exception of verbs of involuntary bodily process, involve volition, or
'agentivity' of the argument.22 By contrast, the argument of the unaccusative entails no
agentivity (Dowty 1991:605ff, Borer 1994:21). More importantly, however, the thematic
role 'Patient' in (46b) indicates that unaccusativity involves 'change of state or location'
as one of its key semantic properties. In particular, it is readily noticed that the inchoative
verbs are associated with this semantic property (cf Levin and Rappaport Hovav
1995:80). As will be discussed in more detail later in Chapter 3, change of state plays a
crucial role in determining not only unaccusativity of verbs but also the ability of verbs to
alternate in transitivity.
22 Dowty (1991:607) characterizes volition as involving 'sentience.'
Another semantic property that appears to have been noticed in the recent
literature as a defining characteristic ofunaccusativity is that of aspect. In particular,
telicity or a natural endpoint in time, which typically characterizes the aspectual classes
of Achievements and Accomplishments, is also considered to characterize unaccusativity
(McClure 1990, Dowty 1991:607, Levin and Rappaport Hovav 1992a:260, 1995:166-7,
Tsujimura 1996:329, Arad 1998).23 The correlation between unaccusativity and telicity is
demonstrated in part by the fact that most of the unaccusatives do not combine with
adverbials of duration likefor an hour.
(47) a. *Tom arrived for an hour.
b. *The accident happened for an hour.
By contrast, unergatives, which are characterized as atelic, are compatible with the same
(48) a. Tom walked for an hour.
b. Tom cried for an hour.
The telic nature of unaccusativity is also observed in Japanese. Tsujimura
(1991:97) points out that the Japanese aspectual marker -te iru means resultative state
when occurring with unaccusative verbs, whereas it means progressive when occurring
with unergative verbs. Thus, as illustrated below, the accusative predicate ai-te iru (< ak-
'open' + te iru) expresses a resultative state.
(49) Mado-ga zenbu ai-te iru.
window-NOM all open-ASP
'All the windows are open' (Tsujimura 1991:98)
23 Tenny (1987, 1994) uses the term "delimitedness" rather than telicity to describe the
same aspectual situation. Arad (1998:18) estimates that 95% of unaccusatives are telic.
By contrast, the unergative predicate arui-te iru (< aruk- 'walk' + te iru) expresses
(50) Masao-wa ima eki-no mae-o arui-te iru
Masao-TOP now station-GEN in front of walk-ASP
'Masao is walking in front of the station now'
Interestingly enough, if unergatives co-occur with adverbials that designate the endpoint
(i.e., telicity) of the activities denoted by the verbs, then the -te iru constructions with
such unergatives no longer express progressive but rather resultative state (cf. Tsujimura
(51) Masao-wa go-maim arui-te iru
Masao-TOP five miles walk-ASP
'Masao has walked five miles'
Examples (50) and (51) seem to show that telicity has to do with the distinction between
unaccusativity and unergativity in Japanese (the issue of aspect in relation to
unaccusativity will be discussed in more detail in Chapter 5).
2.5.5 Unaccusativity and Ergative Alternation
As mentioned at the beginning of this section, many of the intransitive altemants
of the English ergative alternating verbs are identified with unaccusative verbs. Arad
(1998:13) further assumes that the ability of verbs to participate in the ergative
alternation serves as one of the unaccusative diagnostics (cf. Levin and Rappaport Hovav
1995:31). This is readily illustrated by the intransitive verbs of the following alternating
pairs, which are found in the list of the unaccusative verbs in (46) above.
(52) a. Tom broke the vase.
The vase broke.
b. The heat melted the ice in no time.
The ice melted in no time.
c. Tom opened the door.
The door opened.
If the contention that the intransitive members of ergative pairs are unaccusative is
correct, then one may well say that the ergative alternation overtly shows the underlying
position of the single argument of the unaccusative verb. Thus, the transitive altemants in
(52a-c) can be claimed to show the position where the subject of the unaccusative verb is
assumed to originate. Furthermore, if we follow the view that the intransitive break is
unaccusative, then the transitive construction could serve as evidence that the subject of
an unaccusative verb is identified as an internal argument, originating in the object
While most of the intransitive alternants of the ergative alternating verbs are
unaccusatives, it is nevertheless important to note the fact that not all the unaccusative
verbs participate in the alternation (Grimshaw 1987:251, Pinker 1989:42-43, Ono
1997:168). In other words, only a subset of the unaccusative verbs can alternately occur
in transitive constructions. Pinker lists the following verbs as unaccusatives which do not
(53) fall, come, appear, arrive, enter, ascend, die, vanish, exist
Additionally, among the unaccusative verbs given in (45) above, the following seem to
fail to enter into the ergative alternation (cf Pinker 1989:131-132, Levin 1993a):24
24 Some of the non-alternating unaccusative verbs may be used as transitive verbs
depending on the meaning.
(i) a. The sun shines.
b. *The scientists shines the sun
(ii) a. The flashlight shines.
b. Tom shines the flashlight.
D.G. Miller (p.c.) points out that since the verbs in (i) and (ii) have different past tense
forms (shone and shined respectively), they should be considered different verbs.
(54) soar, flow, ooze, seep, trickle, gush, tremble, languish, flourish, thrive,
stumble, succumb, seethe, sit; evaporate, yellow, rot, decompose, germinate,
sprout, bud, wilt, wither, collapse, disintegrate, perish, blush, disappear;
happen, transpire, occur, take place; arise, ensue, result, show up, end up,
pop up; shine, spark, glitter, glisten, glow, crackle, smell, stink; last, remain,
The question remaining to be addressed is: what aspects of the unaccusative verbs in (53)
and (54) make them unable to alternate with a transitive use? Given that unaccusativity is
determined semantically (cf. Levin and Rappaport Hovav 1995:21), is it possible to
further account for such differing behaviors among unaccusative verbs on semantic
grounds? While many of the non-alternating unaccusatives might be semantically
characterized as verbs of existence and appearance or verbs of emission (cf Levin 1993a),
are there any semantic properties shared by these two classes of verb whereby we can
predict the non-alternating behaviors of these verbs? As stated earlier, the position that I
take throughout my dissertation is that the syntactic behaviors of verbs are generally
explicable semantically. Thus I assume that the distinction between alternating and non-
alternating unaccusative verbs should be made in terms of their semantic properties. We
will discuss this issue in more detail in Chapter 3.
2.6 Direction of Derivation
One of the issues addressed in the literature with regard to the ergative alternation
has to do with the direction of derivation between ergative transitives and ergative
intransitives. In other words, research is concerned with which member of the ergative
pair is basic and which one is derived. Earlier research based on derivational morphology
and markedness theory has revealed that the direction of derivation between ergative
pairs varies from language to language (Nedyalkov and Silnitsky 1973, Haspelmath
1993). As noted previously, Japanese is among those languages in which morphological
pairs are formally distinguishable and ample research has been conducted on the issue of
derivational direction (Okutsu 1967). For languages like English, however, in which the
distinction between transitives and ergatives is not overly marked, the direction of
derivation has been a moot point (Levin and Rappaport Hovav 1995, Hale and Keyser
1998). In this section, I review how the issue of direction of derivation has been
addressed in the literature on the ergative alternation in English and Japanese.
2.6.1 Derivational Direction of English Ergative Pairs
The view most commonly held by researchers regarding the direction of
derivation is that transitive ergatives are derived from intransitive bases, a process usually
referred to as causativization (Chomsky 1970:215, Williams 1981, Keyser and Roeper
1984, Guerssel et al. 1985, Pinker 1989, S.T. Rosen 1996). On this view, the ergative
transitive break, for instance, is derived from the intransitive break as a result of the
addition of the semantic element CAUSE to the latter.25
(55) intransitive break cause (+ intransitive break) transitive break
Keyser and Roeper (1984) and Guerssel et al. (1985) consider the process of
causativization to take place in the lexicon, representing the LCS of the ergative verb
break as follows:
(56) a. LCS for intranstive break: y come to be BROKEN
b. LCS for transitive break: x CAUSE (y come to be BROKEN)
25 Hale and Keyser (1986) consider the process of causativization to be a crucial property
of ergative alternation in that it clearly distinguishes the ergative alternation from the
middle alternation. Hale and Keyser argue that a middle verb is derived from a dyadic
LCS, while an ergative intransitive is derived from a monadic LCS (1986:11).
The representation in (56) demonstrates that the transitive LCS (x CAUSE (y come to be
BROKEN)) is derived from the ergative LCS (y come to be BROKEN) by virtue of what
they call a causative rule. On this view, the ergative alternation reflects a syntactic
realization of the intransitive and transitive LCSs of break. In short, Guerssel et al. argue
that while the basic argument structure of the ergative verb is monadic, both transitive
and intransitive uses of ergative pairs are listed in the lexicon. This implies that the verb
break has two separate lexical entries in English(cf. Chapter 4, Section 4.3.1).
In support of the causativizing process of the ergative alternation, Hale and
Keyser (1998:100, 111) simply posit the concept of basic argument structure, maintaining
that "[I]n the absence of morphological evidence, the direction is always from the simpler
structure (the intransitive or inchoative) to the more complex (the transitive)."
Haspelmath (1993) expresses a similar causativization view from a semantic point of
There are independent semantic reasons to think that the causative member of an
inchoative/causative alternation is semantically derived, while the inchoative
member is semantically basic...on purely semantic grounds we seem to be forced
to conclude that causative verbs are derived from inchoatives. (1993:89)
As evidenced in the arguments by Hale and Keyser and Haspelmath, they rely on their
own intuition concerning the direction of derivation with little concrete evidence.
Drawing on grammatical judgment, Chomsky (1970) notes that it is the ungrammaticality
of *NP grow children and the grammaticality of children grow that enables us to say the
ergative use of grow is basic. This observation, however, does not account for the
contradicting case where it is grammatical to say He broke the rule but ungrammatical to
say *The rule broke (for further discussion of this issue see Chapter 3, Section 3.2).
While hardly any conclusive evidence is provided for the theory of
causativization, the view that ergative intransitives are derived from their corresponding
ergative transitive bases has gained substantial ground recently. In contrast to
causativization, this process is referred to as anti-causativization (Zubizarreta 1987) or
detransitivization (Levin and Rappaport Hovav 1995). In keeping with the proposal by
Guerssel et al. (1985), Keyser and Roeper (1984) view ergative pairs as being generated
in the lexicon. Crucially, however, Keyser and Roeper differ from Guerssel et al. in that
they assume that ergative uses are derived from the transitive bases via ergative rule. The
anticausative process presented by Keyser and Roeper is characterized by movement of
the argument via Move ac from the object position inside the transitive VP. Keyser and
Roeper illustrate this process by the verb sink (Keyser and Roeper 1984:402).
(57) sink / [s NP [p [NP]]]
[s NP, [w [t,]]]
Levin and Rappaport Hovav (1995) basically follow Keyser and Roeper in
suggesting that the LCS of break is a single causative representation and the ergative
intransitive counterpart is derived through what they call detransitivization In support of
their view of detransitivization, Levin and Rappaport Hovav provide evidence that
possible subjects of ergative intransitives are a subset of possible objects of the
corresponding transitive ergatives due to selectional restrictions imposed on the former.
That is, ergative transitives appear to allow a wider rage of objects than the
corresponding intransitives allow subjects. This is illustrated below, in which the
syntactic objects table and skirt of the transitive ergatives in (58a) and (59a) fail to occur
in the subject positions of the intransitive counterparts in (58b) and (59b).
(58) a. The waiter cleared the table.
b. *The table cleared.
(59) a. The dressmaker lengthened the skirt.
b. *The skirt lengthened (Levin and Rappaport Hovav 1995:86)
On Levin and Rappaport Hovav' terms, causativization approach does not provide a
satisfactory explanation as to how transitives can be derived from non-existent
intransitive counterparts. Building on their claim that "the basic use of the verb will
impose less stringent restrictions on its arguments," Levin and Rappaport Hovav
conclude that ergative transitive constructions are basic (1995:86).
Moreover, as Levin and Rappaport Hovav point out, cross-linguistic
morphological evidence appears to support their transitive-based and intransitive-derived
view (1995:87-88). Citing a study by Haspelmath (1993) in which ergative alternating
patterns in more than twenty languages are analyzed, Levin and Rappaport Hovav stress
the fact that the ergative intransitive form of break, seemingly the prime example of the
ergative verb among researchers, has a strong tendency to be more marked
morphologically than the transitive counterpart. In light of the evidence of derivational
morphology, it is reasonable to assume, even in languages like English where there are
hardly any morphological distinctions between ergative transitives and intransitives, that
ergative transitives are basic and the corresponding intransitives are derived via
2.6.2 Derivational Direction of Japanese Ergative Pairs
As mentioned previously, the ergative pairs in Japanese are striking in that almost
all the pairs are distinct in form. In a majority of cases, the formal distinctions are
asymmetrical, that is either transitive ergatives are more marked than the corresponding
intransitives or vice versa. Furthermore, as illustrated in Section 2.3.2 above,
morphological derivations between pairs appear to follow relatively constant patterns.
Given such consistent correspondence between derivational forms and transitivity, it has
been proposed that the derivational directions of Japanese ergative pairs are basically
predictable (Okutsu 1967, Jacobsen 1982, 1985, 1992, Kanaya 2000). There is a caveat,
however, about relying too much on derivational morphology to determine the direction
of derivation. In fact, in light of a wide range of morphological devices whereby one
member of an ergative pairs is derived from the other, Levin (1985) argues that "it is not
possible to assume that one member of the pair is derived from the other on the basis of
the nature of the morphological relation between the verbs involved" (21). Since Levin
does not seem to take Japanese data into account in her analysis, it remains to be seen
whether her argument is applicable to ergative pairs in Japanese. In what follows, we will
investigate how accurately the derivational patterns of Japanese ergative pairs may allow
us to determine the direction of derivation.
26 It is important to note here that researchers are not always certain about the
directionality of derivation. Dixon (1991:291), for instance, expresses his uncertainty
about his position, suggesting that the direction is determined depending on the ergative
verb type. Building for the most part on native speakers' intuition, Dixon assumes that
break, crush, and smash are basically transitive, burst and explode are basically
intransitive, and tear and chip are indecisive.
188.8.131.52 Transitivization vs. intransitivization
Based on the view that morphologically marked forms are derived, Okutsu (1967)
proposes three derivational patterns associated with ergative pairs in Japanese. The first
pattern is referred to as tadooka ('Transitivization'), which involves a derivational
process of deriving transitive forms from intransitive bases by means of the Transitivizer
(60) INTRANSITIVE TRANSITIVE
ugok- 'move' ugokas- 'move'
tob- 'fly' tobas- 'let fly'
wak- 'boil' wakas- 'boil'
The second pattern is zidooka ('Intransitivization'), which shows a derivational pattern
completely opposite to tadooka; it derives transitive forms from the intransitive bases by
virtue of the Intransitivizer -ar.
(61) TRANSITIVE INTRANSITIVE
hasam- 'put between hasamar- 'get caught between'27
tunag- 'connect' tunagr- 'get connected'
husag- 'block' husagar- 'get blocked'
The two opposing derivational patterns just described may raise a question about
describing the direction of derivation regarding the English ergative alternation as either
causativization or anticausativization The issue may be further complicated by the third
pattern called ryookyokuka ('Polarization'), in which both transitive and intransitive
forms appear to be derived from some hypothetical roots.28
27 When there is no intransitive counterpart in English, I use 'get + past participle,'
following a suggestion by Croft (1991:267-8), to differentiate it from the prototypical,
28 Haspelmath (1993) notes that the Japanese ergative pairs are characterized as non-
directed alternation in that both transitive and intransitive uses are derived from the same
(62) TRANSITIVE INTRANSITIVE
naos- 'fix' naor- 'get fixed'
hirak- 'open' hirak- 'open'
For the pair naos- and naor-, the transitive suffix -s and the intransitive suffix -r are
considered equal in terms of markedness, making it impossible for us to determine the
derivation of direction. For the ambivalent pair hirak-, on the other hand, since there is no
change in form between the transitive and intransitive, Polarization would be the most
appropriate classification (Okutsu 1967:58). In short, given the morphological evidence
for multiple derivational patterns of Japanese ergative pairs, one may well reconsider the
question of derivational direction involving the English ergative alternation
184.108.40.206 Causativization vs. passivization
It is important to notice that in Japanese the process of transitivization and
intransitivization just described is often considered comparable to the process of
causativization and anticausativization, respectively (Shibatani 1973, Noda 1991,
Jacobsen 1992, Kitagawa and Fujii 1999). This observation arises in part from
morphological evidence. That is, the transitivizer -as bears a morphological resemblance
to the Japanese causative suffix -(s)ase, a causative morpheme which attaches to either
intransitive or transitive verbs almost freely to create the meaning 'cause somebody to do
stem. One typical pair representing the non-directed alternation is atumar-/atume-
'collect, gather," a derivational pattern comprising Jacobsen' s Class III. In this pair, the
transitive and intransitive uses are equipollently derived from the stem atum- by virtue of
the derivational suffixes -ar and -er, respectively. The problem with Haspelmath's claim,
however, is that the number of verb pairs (31) surveyed is very small compared to the
number of ergative pairs (134) given by Jacobsen (1992). More importantly, his analysis
of derivational suffixes is rather vague. For instance, while Haspelmath labels the pair
yurer-/yuras- as non-directed, he considers a similar pair kawak-/kawakas- as
causativization. It seems that at the very least Haspelmath analyzes these particular pairs
mainly from a isomorphic point of view without taking into account the semantic
function of the Transitivizer -as.
(something).' Likewise, the intransitivizer -ar is viewed as resembling the Japanese
passive suffix -rare, a passive morpheme whose most common function is to passivize
transitive verbs.29 Given such similarities, Noda (1991) concludes that there is little
difference between causativization/passivization and transitivization/anticausativization.
The only difference, according to Noda, should be the degree of productivity involved in
each process; causativization/passivization is more productive than
transitivization/anticausativization in that the former is applied to a larger number of
Moreover, the parallelism between transitivization and causativization, on the one
hand, and the parallelism between intransitivization and passivization, on the other, has
been demonstrated syntactically in the literature (Ichihashi 1992). As schematized
previously, repeated here in (63), the configurational property characteristic of the
ergative pairs is that the object of the transitive verb corresponds to the subject of the
(63) NPi-ga NP2-o Vtr (transitive construction)
NP2-ga Vintr (intransitive construction)
The identical correlation is observed between the non-causative (i.e. intransitive) and
causative constructions mediated by -(s)ase.
29 Diachronical surveys seem to suggest that the transitivizer -as was derived from the
causative morpheme -(s)ase and the intransitivizer -ar was derived from the passive
morpheme -are (cf Shibatani 1990:236).
(64) a. NPi-gaNP2-o Vi + -(s)ase (causative construction)
Akira-wa Taroo-o go-maim aruk-ase-ta30
Akira-TOP Taro-ACC five-mile walk-CAUS-PAST
'Akira made Taro walk five miles'
b. NP2-ga Vi (intransitive construction)
Taroo-wa go-mairu arui-ta (< aruk- + ta)
Taro-TOP five-mile walk-PAST
'Taro walked five miles'
Along the same lines, the syntactic configuration of passivization is comparable to (64).
(65) a. NPi-gaNP2-o Vtr (active (i.e. transitive) construction)
Taroo-ga hon-o nusun-da
Taro-NOM book-ACC steal-PAST
'Taro stole the book'
b. NP2-ga (NPi-ni) Vtr + -(r)are (passive construction)
Hon-ga (Taroo-ni) nusm-are-ta
book-NOM (Taro-by) steal-PASS-PAST
'The book was stolen by Taro'
In spite of the striking morphological and syntactic similarities just described,
many researchers agree that the processes of transitivization and intransitivization should
be distinguished from causativization and passivization. Shibatani (1976a) discusses the
disparity between transitivization and causativization from a structural point of view.
More specifically, transitive verbs with the transitivizer -as are lexicalized, contributing
to a mono-clausal structure as a whole. Causative verbs with the causative morpheme -
(s)ase, on the other hand, are assumed to constitute an embedded structure, just like
complex structures associated with the analytical causative verbs make or have in English
(Shibatani 1976a:244; see also Kuroda 1993).
30 The choice between -ase and -sase entirely depends on whether the stem of a verb ends
with a consonant (kak- 'write' kak-ase- 'make someone write') or a vowel (tabe- 'eat'
- tabe-sase- 'make someone eat).
(66) a. Taroo-wa isi-o ugokas-ita
Taro-TOP stone-ACC move-PAST
'Taro moved the stone'
b. Taroo-wa Akira-ni isi-o ugok-ase-ta
Taro-TOP Akira-DAT stone-ACC move-CAUSE-PAST
'Taro made Akira move the stone'
As for the contrast between ergative intransitives and passives, Noda (1991:225-
6) suggests that it is the lack of agentivity that differentiates ergative intransitives from
passives. According to Noda, the ergative ware- in (65a) implies that a balloon blew up
spontaneously, whereas the passive war-are- in (67b) implies that somebody blew up the
balloon, whether on purpose or by accident.
(67) a. Huusen-ga ware-ta
'The balloon burst'
b. Huusen-ga war-are-ta
'The balloon was burst'
Jacobsen (1992) mentions non-productive characteristic of morphologically-based
transitivization and intransitivization in Japanese, arguing that as a whole causativization
and passivization are productive enough for native speakers never need to memorize each
causativized and passivized form individually (cf. Chapter 4, Section 4.2)
SEMANTIC CHARACTERIZATION OF ERGATIVE ALTERNATIONS
The fundamental hypothesis behind the present research is based on a view of
lexical semantics in which the semantics of verbs is responsible to a great extent for the
syntactic realizations of arguments surrounding the verbs (Wasaw 1985, Grimshaw 1987,
Pinker 1989, Levin and Rappaport Hovav 1995, Rappaport Hovav and Levin 1998). The
rationale behind the syntax-lexical semantics interface is that the alternatability of the
arguments of verbs, as represented by ergative alternation, dative alternation, and locative
alternation, among others, is predictable from the meaning of the verbs. According to
Levin (1993a), for instance, the semantic group of English verbs called break-verbs, such
as break, crack, shatter, and tear, undergo the ergative alternation, whereas the semantic
group of verbs called hit-verbs, such as batter, hammer, hit, andpound, do not.
One main research goal regarding ergative pairs has been to elucidate "the
conditions under which verbs allow alternation in transitivity" (Ritter and Rosen
1998:135). As mentioned in Chapter 1, Pinker (1989) notes that semantic properties that
appear to be pertinent to the alternating behaviors of verbs are similar across languages:
The same alternations in other languages are prone to applying to the same kinds
of verbs and being constrained by the same kinds of criteria and shifts in
interpretation as one finds in English (1989:97)
Specifically, Pinker points out that verbs with the meaning 'breaking' have a strong
tendency to participate in a lexical ergative alternation cross-linguistically, whereas verbs
with the meaning 'laughing' invariably lack a lexical ergative alternant or variant
(1989:134; see also Marantz 1984:181-2, Hale and Keyser 1998:89). Haspelmath
conducts an extensive cross-linguistic survey of derivational morphology between
ergative pairs, generalizing that ergative pairs in a majority of languages contain 'change
of state' as an essential semantic component (1993:92-93). In this chapter, I will provide
a lexical semantic analysis of the ergative alternation in English and Japanese. It will be
demonstrated that while verbs which undergo this particular alternation appear to belong
to similar semantic classes in both languages, there are yet crucial differences that need to
3.2 Two Issues of Ergative Alternation
There are two issues surrounding the ergative alternation that need to be
addressed here. One is that there are transitive and intransitive verbs which do not
alternate in transitivity under any circumstances. As mentioned above, the English verb
hit simply lacks an ergative intransitive use.
(1) a. Tom hit the boy.
b. *The boy hit.
The Japanese equivalent tatak- 'hit' also fails to occur in intransitive constructions.
(2) a. Taroo-wa Akira-o tatai-ta. (< tatak- + ta)
Taro-NOM Akira-ACC hit-PAST
'Taro hit Akira'
b. *Akira-ga tatakat-ta.1 (tatakar- + ta)
Akira-NOM get hit-PAST
'Akira got hit'
1Since there is no intransitive counterpart to tatak-, I have coined a hypothetical form,
upon which native speakers would agree as the most probable form. In what follows,
whenever I need to coin a hypothetical form, I follow this principle.
Another complex aspect of the ergative alternation is that verbs which alternate in
transitivity are not always consistent with their behaviors (Van Voorst 1995:513,
Lemmens 1998:37). For instance, while the English verb break normally alternates in
transitivity, when the verb takes contract as a Theme argument, it fails to undergo the
(3) a. He broke the contract.
b. *The contract broke. (Levin and Rappaport Hovav 1995:85)
Van Voorst (1995:514) lists more cases of such inconsistent behaviors of ergative verbs.
(4) a. He cracked the plate The plate cracked
b. He cracked the security code *The security code cracked
(5) a. He crashed his car into a tree His car crashed into a tree
b. He crashed the party *The party crashed
(6) a. He thickened the sauce The sauce thickened
b. He thickened the line *The line thickened
(7) a. He twisted the cord The cord twisted when he pulled on it
b. He twisted his hair with his fingers *His hair twisted
(8) a. He stretched the elastic band The elastic band stretched
b. He stretched his leg *His leg stretched
2 Brousseau and Ritter (1991:60) note that the same phenomenon is observed in French.
(i) a. Jean a brisesa l'accord
'John broke the agreement
b. *L'accord s'est brise(e)
'The agreement broke'
Given these examples, as will be discussed later in more detail, the alternatability of
verbs appears to depend not only on the verbs themselves but also on the constructional
environments where these verbs occur.
The same observation applies to a number of Japanese ergative verbs. The
intransitive verb tat- 'to stand' is paired with tate- 'set up,' alternating in transitivity
according to the configurational template given in Chapter 2, Section 2.3 above:
(9) a. Boo-ga tat-te iru
'The pole stands'
b. Kodomo-ga boo-o tate-te iru
child-NOM pole-ACC put up-ASP
'The child is putting up the pole'
However, if the intransitive verb tat- occurs with an animate subject, then the transitive
counterpart becomes ungrammatical in the construction that follows the same template:
(10) a. Kodomo-ga rooka-ni tat-te iru
child-NOM hallway-in stand-ASP
'A child stands in the hallway'
b. *Taroo-wa kodomo-o rooka-ni tate-te iru.
Taro-TOP child-ACC hallway-in stand-ASP
'Taro stands a child in the hallway'
Given the examples from English and Japanese, the objectives of this chapter are (1) to
explain why some verbs alternate in transitivity while others simply do not and (2) to
elucidate the conditions under which typical ergative verbs like break fail to show the
3.3 A Problematic Case: The English Cut and Break
One way of explaining how the semantics of verbs determine the syntactic
structures of arguments occurring with the verbs is to compare verbs which are
semantically distinct. The contrast between break and hit is a case in point. As will be
discussed later in detail, the transitive use of break entails the semantic properties 'cause'
and 'change of state,' while hit does not. The weakness of this approach might be that if
there is more than one semantic property that differs among verbs as in break and hit,
then it becomes more difficult to pinpoint the true cause of the differing syntactic
behavior shown by each verb. In this respect, the contrast between the English break and
cut has been considered more appropriate to the study of the syntax-lexical semantics
interface due to their semantic similarities (e.g., the transitive uses of the verbs entail both
'cause' and 'change of state'), on the one hand, and yet some crucial syntactic
differences, on the other. In this section I outline the semantic differences and similarities
between the two verbs and their syntactic behaviors that are assumed to reflect the
The verb break is one of the verbs which are subsumed in general under a large
semantic verb group called "change of state" (Levin 1993a). In particular, the change of
state described by break is considered so prototypical that it is often used as the label of a
subclass of change-of-state verbs (e.g., verbs of breaking, break-type verbs, etc.) such as
shatter, crack, split, etc. The essential semantic property associated with this particular
verb group would be summarized as 'a change in the material integrity' (Hale and Keyser
1987:7) or, more elaborately, as a 'change that involves an initial condition of being
whole and a final condition of being (able to be) separated into several detachable
portions' (Ravin 1990:222). It is this semantic property that distinguishes break-verbs
from other change-of-state verb groups such as "verbs of bending," "verbs of cooking,"
and "verbs of change of color" (See Levin 1993a for detailed study of this issue). The
verb cut also involves a similar kind of change of state as break, namely 'a separation in
its material integrity' (Hale and Keyser 1986:11, 1987:7). It is interesting to note,
however, that under the semantic-model approach, the verb cut, which comprises another
semantic group of verbs referred to as "contact-effect" (e.g., crush, pierce, slash, bite,
shoot, etc.), is normally separated from verbs of breaking (Levin 1995:67).
Given that break and cut share change of state as one of their essential semantic
properties, it is necessary to account for why they should be classified into separate
semantic verb groups. One may seek an explanation for this question in the syntactic
behaviors that each verb shows. In fact, the syntactic structures where cut and break can
occur have been discussed extensively in the literature, particularly under the scope of
argument structure alternations (Kilby 1984, Guerssel et al. 1985, Hale and Keyser 1986,
1987, 1988, Pinker 1989:104-109, Levin and Rappaport Hovav 1992b:136, 1994:62-3,
1995:103, Levin 1993a:5-11, Miller 1993:180). In the remainder of this section, we
compare the syntactic behaviors of the two verbs.
Firstly, both verbs typically undergo the middle alternation as illustrated below
(cf. Chapter 2, Section 2.2).
(11) a. This bread cuts easily.
b. This glass breaks easily. (Pinker 1989:106)
On the other hand, they both fail to undergo the so-called "contact locative" alternation, a
syntactic behavior which is typified by verbs like hit and bump.
(12) a. I hit the wall with the bat.
I hit the bat against the wall.
b. I cut the bread with the knife.
*I cut the knife against the bread.
c. I broke the egg with a spoon
*I broke a spoon against the egg. (Pinker 1989:107)
What is more striking is certain alternating behaviors associated with cut, but not with
break. One such behavior is conative alternation, in which cut takes the preposition at
implying that "the subject is trying to affect the oblique object but may or may not be
succeeding" (Pinker 1989:104; see Guerssel et al. 1985:50, Miller 1993:180, Levin
1995:67 for further discussion).
(13) a. Margaret cut the bread.
b. Margaret cut at the bread. (Pinker 1989:104)
(14) a. Janet broke the cup.
b. *Janet broke at the cup. (Levin 1995:67)
Another syntactic behavior exclusive to cut is part-possessor ascension, in which "the
'possessor' appears as the direct object and the body-part noun appears in a locativee
prepositional phrase'" (Fillmore 1970:126).
(15) a. Sam cut Brian's arm.
b. Sam cut Brian on the arm. (Pinker 1989:105)
(16) a. Jim broke his leg
b. *Jim broke him on the leg. (Fillmore 1970:126)
The most frequently discussed alternating behavior, however, in relation to cut and break
is ergative alternation. As mentioned previously, break is among other change-of-state
verbs that undergo the ergative alternation (see Chapter 2, Section 2.2 above). The verb
cut, on the other hand, does not undergo this alternation 3
3 The verb cut rarely occurs in a seemingly intransitive construction as follows:
(i) This bread cuts.
(17) a. Margaret cut the bread.
b. *The bread cut. (Levin 1995:66)
The argument structure alternations in which cut and break do and do not participate are
summarized as follows:
Table 1: Alternation patterns of cut and break
causative conative middle contact-locative part-possession
alternation alternation alternation alternation alternation
cut + + +
break + +
Once argument structure alternations are understood to revolve around verbal
predicates, one may say that cut and break show a good deal of syntactically distinct
behaviors. The question that still remains to be addressed is what factors of the verbs
trigger such different syntactic behaviors. Or put differently, since I conform to the
position that the semantics of verbs determine the syntactic behaviors of the verbs, what
are the semantic properties that are responsible for different alternating behaviors as
exemplified by cut and break? In the next section, I will extend our discussion and
explore semantic properties of verbs in English and Japanese, especially focusing on the
verbs' meanings that might have to do withthe different behaviors in terms of the
Hale and Keyser (1987:19), however, call the sentence an "unadorned" middle,
differentiating it from the genuine ergative intransitive construction The vase broke.
3.4 Conditions for Ergative Alternation: English
3.4.1 Change of State
One verbal semantic property that is viewed as playing a crucial role in
determining alternatability of verbs is that of "change." In linguistics literature, verbs
undergoing the ergative alternation have been understood to indicate some sort of change
brought about on a Theme argument. As early as the 1920s, Jespersen had pointed out
that verbs that bring about a change in a person or a thing tend to be "doublefaced" or
alternate in transitivity in English (Jespersen 1927:332-3).
(18) Change-class verbs
break the ice the ice breaks
boil water the water boils
burst the boiler the boiler burst
improve an invention his health has improved
Levin (1985:18-19) specifies the types of change associated with verbs, maintaining that
verbs of change of state and position undergo the ergative alternation in English.
(19) Change of state: break, crack, open, close, melt, freeze, harden, dry
Change of position4: roll, bounce, move, float, drop, turn
Traditionally, however, change of state or, more elaborately, "changes in the
physical shape or appearance of some entity' (Levin and Rappaport Hovav 1995:80), has
been viewed in the literature as the key semantic property determining whether a given
verb may participate in the ergative alternation (Fillmore 1970, Smith 1978, Levin 1985,
1993a, Pinker 1989, Haspelmath 1993, Levin and Rappaport Hovav 1992b, 1994, 1995,
Van Voorst 1995, S.T. Rosen 1996). The role of change of state in verbs' alternatability
will be made more explicit when compared to verbs which inherently involve no change
4 In Levin (1993b), change of position is referred to as "manner of motion."
of state at all. Fillmore (1970) clearly illustrates this point by comparing the verbs break
(20) a. John/A rock broke the stick.
The stick broke.
b. John/A rock hit the tree.
*The tree hit. (Fillmore 1970:122-123)
According to Fillmore, the reason why a contact verb like hit fails to alternate in
transitivity is that unlike break, it does not inherently entail any effect or change of state
on a co-occurring Theme argument. This point will be illustrated more explicitly in (21).
(21) a. I hit the vase with a hammer, but it did not break; it was made of iron
b. *I broke the vase with a hammer, but it did not break; it was made of iron
Based on this observation, Fillmore concludes that break participates in the ergative
alternation because it lexically entails a change of state.
The essential role of change of state in verbs' alternatability is further illustrated
by the verb bake in a unique manner (Atkins et al. 1988, Levin 1993a:175 & 243-4,
Kageyama 1996:161-2, Ono 2000:8ff). The verb is normally categorized as a change-of-
state verb and in this sense, as expected, it undergoes the ergative alternation.
(22) a. She baked the potatoes.
b. The potatoes baked. (Kageyama 1996:161)
Interestingly enough, if we replace 'potato' with 'cake,' then the intransitive construction
(23) a. She baked the cake.
b. *The cake baked.5
Levin and Rappaport Hovav (1992a:259) explain that the meaning of bake in (23) is
slightly different from the one in (22) in that baking a cake implies 'creation' in addition
to the basic change-of-state meaning, roughly phrased into 'create by means of change of
state bake.' In short, as pointed out by Levin and Rappaport Hovav(1992b: 139), only
when bake implies change of state, does it exhibit the ergative alternation. That the
additional semantic property of creation may hinder bake from alternating in transitivity
is further supported by the fact that other verbs of creation like make, produce, build,
assemble, etc. do not undergo such alternation
Note, however, change of state is not necessarily the sole factor in determining
whether a given verb will undergo the ergative alternation. Firstly, as Levin and
Rappaport Hovav (1994:41) point out, some groups of verbs such as verbs of emission
(sound or light) and position, which are not readily identified with verbs of change of
state, do alternate in transitivity as illustrated below.
(24) a. Tom beamed the flashlight.
The flashlight beamed.
b. Tom hung the photo on the wall.
The photo hung on the wall. (Levin and Rappaport Hovav 1994:42)
More importantly, there are many verbs of change of state in English which do not
alternate in transitivity (Levin and Rappaport Hovav 1992b:133n). As noted earlier,
5 Kageyama (1996:161) gave the sample sentences She bakeda cake/*A cake bakedwith
no explanation about why he changed the article (from the to a) as well as the noun. In
order to minimize additional complexities, I chose to use 'the cake' in my sample
contact-effect verbs to which cut belongs provide support for this statement. Levin refers
to this group of verbs simply as cut verbs, listing the following members (1993a:156):
(25) chip, clip, cut, hack, hew, nip, saw, scrape, scratch
Recall that while cut is normally classified in a different semantic class than break, it
crucially involves change of state (Fillmore 1970, Levin and Rappaport Hovav 1994,
1995).6 Nevertheless, the transitive verb cut fails to occur intransitively, as illustrated
(26) a. Elsa clipped the article out of the paper.
b. *The article clipped out of the paper.
(27) a. I sawed the board in half.
b. *The board sawed in half.
(28) a. Jane scraped the carrot thoroughly.
b. *The carrot scraped thoroughly.
(29) a. Vandals had slashed most of the seats on the train.
b. *Most of the seats on the train slashed.
(30) a. I hurriedly snipped the string.
b. *The string hurriedly snipped.
Another semantic group of verbs that needs to be mentioned here is what Levin refers to
as destroy verbs (Levin 1993a:239).
(31) Destroy verbs:
annihilate, blitz, decimate, demolish, destroy, devastate, exterminate,
extirpate, obliterate, ravage, raze, ruin, waste, wreck
6 For the purpose of illuminating the difference between break and cut, Levin (1993a:9)
refers to break as a 'pure' change of state verb.
Like cut verbs, the destroy verbs involve change of state but fail to alternate in
(32) a. The bomb destroyed the whole city.
b *The whole city destroyed.
(33) a. The bulldozer razed the building.
b. *The building razed.
Levin (1993a:239) notes that the destroy verbs do not participate in the ergative
alternation since they uniformly denote the total destruction of entities (see Section
220.127.116.11 below for more detailed discussion of destroy verbs). In short, the examples in
(26) (30) and (32) (33) suggest that change of state is not the sole determining factor
of the altematability of verbs in English.
Another semantic property that needs to be considered surrounding the ergative
alternation is agentivity. When an entity is agentive, the entity or 'agent' always involves
volition or intention (Talmy 1976, Delancey 1984).7 As discussed in Chapter 2,
agentivity is most typically associated with the subjects of unergative verbs.
7 One difficulty with agentivity is its definition In particular, the difficulty has to do with
the question of whether agentivity can be characterized by one single semantic feature
such as 'animacy' or 'volition' (Cruse 1973, Hopper and Thompson 1980, Delancey
1984, Schlessinger 1995; see Somers 1987 and Abdul-Roaf 1998 for detailed reviews of
discussion on this issue). Due to the difficulty in pinpointing a single semantic property
associated with agentivity, it has become more common to take a multiple-element
approach to this issue in recent years. Foley and Van Valin (1984:32) consider a
combination of animacy, volition and control to be the contributing factor to the agentive
interpretation of an "actor." Oosten (1980:482) proposes the most comprehensive view of
agentivity, arguing that four semantic properties-intentionality, volition, control, and
responsibility-combine to make an entity agentive.
(34) a. I walked along the river.
b. Tom runs five miles every day.
Agentivity also characterizes the subjects of causative transitive verbs. Returning to the
contrast between break and cut, it has been assumed in the literature that the impossibility
of cut in the ergative intransitive construction is due to its indispensable implication of a
volitional agent that takes control of the use of a sharp instrument (O'Grady 1980:63,
Kilby 1984:44, Haspelmath 1993:93-94, Lemmens 1998:37). In other words, the agent-
oriented verb cut is incompatible with ergative intransitive constructions in which, as
mentioned in Chapter 2, Section 2.2, the thematic roles of Agent and Cause are
suppressed and not in central focus. Such semantic constraints on the subjects of cut are
reflected in selectional restrictions on the choices of arguments as subjects. That is, the
ungrammaticality of *The lightning cut the clothesline is due to the fact that no natural
forces such as lightning can be viewed as being volitional (Levin and Rappaport Hovav
By contrast, agentivity is not necessarily an essential semantic property for break.
While the verb can take an agentive entity as subject in transitive constructions as in Tom
broke the vase, it is important to note that the event denoted by break can be brought
about in varying manners. This is illustrated by the fact that unlike cut, the transitive
ergative use of break takes non-agentive inanimate entities like an instrument and natural
force as subject (cf Langendoen 1970:72-3, Talmy 1976, 1985, Levin and Rappaport
8 Nishimura (1993:503) suggests the possibility of assigning the Agent role to the
(35) a. The ax broke the window.
b. The earthquake broke the window.
c. The falling stone broke the window.
Langacker (1991:332) points out that there are many other ergative verbs in English like
open and wake up that allow a wide range of causers or causing events as subjects. In
short, under the assumption that causative change-of-state verbs that undergo the ergative
alternation are lexically unspecified about agentivity, one might attribute the
unaltematability of cut to its entailment of a volitional agent as part of the inherent
meanings of the verb.
18.104.22.168 Kill verbs
While the concept of agentivity seems to account for the contrast between break
and cut in terms of alternatability in transitivity, the analysis based on agentivity poses
several problems. Firstly, agentivity does not necessarily determine correctly whether a
given change-of-state transitive verb ergativizes. Compare killwith break. Like the
transitive use of break, the verb kill contains CAUSE in its semantic representation and its
internal argument appears to undergo change of state.9 As with break, the verb occurs
with non-agentive as well as agentive causes.
(36) a. An arrow through the heart killed Max
b. Malaria killed Nigel. (Foley and Van Valin 1984:32)
Nevertheless, kill does not participate in the ergative alternation as shown below.
(37) a. *Max killed (due to an arrow through the heart).
b. *Nigel killed (due to Malaria).
9 Fontenelle and Vanandroye (1989:19) refer to as kill as a purely causative verb,
differentiating it from ergative verbs like break.
Instead, the inchoative meaning of death is expressed with the completely distinct lexical
item die or the verbal phrase pass away.
(38) a. Max died due to an arrow through the heart.
b. Nigel died/passed away with Malaria.
The peculiarity of kill in terms of the ergative alternation will be made more explicit
when compared to other verbs with similar meaning such as assassinate, murder,
slaughter, etc., which also fail to be used intransitively.
(39) a. The terrorist assassinated/murdered the senator.
b. *The senator assassinated/murdered.
(40) a. The terrorists slaughtered many civilians.
b. *Many civilians slaughtered.
The inability of the verbs in (39) and (40) to alternate in transitivity will be accounted for
by our preceding discussion on agentivity because this set of verbs apparently lexicalizes
agentivity. 10 This is readily illustrated by the fact that none of the verbs can take non-
agentive subjects. As mentioned above, the subject of kill can be an inanimate object or
natural force which, for lack of volition, could not be the subject of assassinate and
murder (cf. Fillmore 1968:28).
10 Intention is also an essential lexical semantic component of murder and assassinate.
That these verbs entail intention as an indispensable feature is shown by the fact that it is
redundant for them to have the phrase on purpose, whereas the verb kill can occur with
(i) a. #The man murdered the senator on purpose
b. #The man assassinated the senator on purpose
c. The man killed the senator on purpose
Lack of intention in the verb kill is further illustrated by the fact that the verb can occur
with the adverb phrase by accident, whereas assassinate and murder cannot.
(ii) a. The man killed his brother by accident.
b. *The man assassinated/murdered his brother by accident.
(41) a. *The explosion assassinated/murdered the senator.
(Levin and Rappaport Hovav 1995:102)
b. *The earthquake assassinated/murdered the senator.
(42) a. The explosion killed more than one hundred people.
b. The earthquake killed more than one hundred people.
The question that remains to be addressed, therefore, is whether it still is possible to
account for the unalternatability of kill without relying on agentivity.
Levin and Rappaport Hovav argue that transitive verbs that always require "an
animate intentional and volitional agent as subject" never detransitivize (1995:102).11
Such verbs include, other than assassinate and murder, verbs of creation like write and
build. In short, agentivity does account for the alternatability of certain transitive change-
of-state verbs. Nevertheless, given the non-alternating behavior of the non-agentive kill
there seems to be a need to reconsider the validity of adopting agentivity to the analysis
of English ergative pairs.
22.214.171.124 Destroy verbs
Another group of verbs that raise a question about viewing agentivity as a
contributing factor in verbs' non-alternatability are destroy verbs. Maruta (1998:94)
analyzes the verbs destroy and devastate, arguing that these verbs are similar to kill in
that while they can take non-agentive subjects like natural forces as subjects, they do not
alternatively occur transitively and intransitively.
11 By "volitional" or "intentional" it is meant that an agent is interpreted as "deliberately
performing an action which brings about a change of state in a patient" (Delancey
1984:5). The obligatory presence of the volitional agent as the causer of an event is also
considered an essential property which determines whether a given transitive event is
prototypical or not (Delancey 1985, Lakoff 1977, Hopper and Thompson 1980).
(43) a. The avalanche destroyed several houses.
b. *Several houses destroyed.
(44) a. Hurricanes devastated the region
b. *The region devastated.
Levin (1993a) suggests that an additional semantic property may provide an account for
such syntactic behaviors of destroy verbs. According to Levin, it is the degree of change
of state that results in the non-alternatability of the destroy verbs listed in (31). In other
words, destroy verbs imply that the Themes (house and region) of these verbs undergo a
complete demolition of their physical structures, entirely losing their original functions or
uses (Levin 1993a:239; cf Dixon 1991:112). This will be illustrated more clearly by the
awkwardness of sentence (45b), compared to (45a):
(45) a. I broke the glass, but I was able to drink some water with a remaining
b. #1 destroyed the glass, but I was able to drink some water with a
Another way of demonstrating that the destroy verbs indicate the totality of destruction is
that they do not occur with the particle up, which, combined with deformation/separation
verbs, emphasizes the completeness of a change of state. In other words, it is redundant to
attach up to a verb which already means a complete destruction lexically (Jackendoff
1990:116, Kageyama 1996:222). For instance, the verb wreck and waste, which are
among the destroy verbs in Levin (1993a:239), does not occur with up.
(46) a. The earthquake wrecked up hundreds of old buildings.
b. *The war wasted up the country.
This is not the case with break, which does not inherently involve such totality of
destruction. Thus, the use of up with break as in Tom broke up the vase is grammatical.12
In short, the impossibility of destroy verbs to occur in ergative intransitive constructions
as in (43b) and (44b) above again suggests that agentivity associated with the subject of
the causative transitive is not the sole factor in determining the verb's alternatability.
To summarize, although change of state and agentivity are crucial semantic
properties that have much to do with the alterability of verbs, they do not provide an
adequate explanation about the two key semantic groups of verbs-cut and destroy
3.4.3 External vs. Internal Causation
We have observed that the two semantic properties-change of state and
agentivity-do not convincingly account for the alternatability of verbs in English. At the
very least, these semantic properties did not adequately account for the syntactic
behaviors exemplified by certain causative transitive verbs like kill and destroy. Given
such problems, another approach to the ergative alternation proposed in the literature
over the last two decades is to focus on the type of causation relevant to the realization of
an eventuality. More specifically, this approach suggests that causation is not a uniform
concept but rather characterized by its multifaceted properties. The crucial difference
between causation and agentivity is that the former places more emphasis on the
connection between agents/causers and caused events, as frequently discussed in the
causal chain model (cf Croft 1990).
12 In terms of aspect, destroy is classified into Accomplishment, while break is
Achievement (Brinton 1988:29, Smith 1991; for more discussion of aspectual
characteristics of Accomplishment and Achievement, see Chapter 5 below).
Smith (1978) recognizes two semantic features associated with ergative verbs,
namely 'independent activity' and 'external control. 13 In other words, Smith argues that
verbs that participate in the ergative alternation express activity or change of state that
can occur relatively independently but at the same time has the possibility of being
controlled by an external agent (Smith 1978:101-2; cf. Davidse 1992:109). For instance,
verbs like break and open alternate since the events denoted by these verbs can occur
independently or be externally controlled. By contrast, verbs like destroy or buildwhich
denote activities or changes of state that are controlled only by external agents do not
alternate in transitivity. Along the same line, verbs like shudder or laugh do not alternate
with transitive uses in that control over the se activities cannot be relinquished entirely to
external agents (Smith 1978:107; cf Levin and Rappaport Hovav 1995:90). In short,
Smith assumes that the alternation between an ergative intransitive and an ergative
transitive reflects the dual features ('independent activity' and 'external control')
associated with the ergative verbs.
Levin and Rappaport Hovav(1994, 1995) follow and expand on Smith's view of
change in relation to causation, proposing a distinction between external causation and
internal causation, which is nearly equivalent to Smith' s independent activity and
external control. In their terms, verbs which fail to undergo the ergative alternation
represent internally caused eventualities, which result either from the volition or will of
agents that perform activities like play and speak or from the inherent properties of the
arguments that undergo the events represented by verbs like blush and tremble. By
13 Independent activity might be identified with spontaneous event. For the correlation
between ergative intransitives and spontaneity, see Chapter 2, Section 2.2.2 above.
contrast, verbs like break which participate in the ergative alternation describe
eventualities that are necessarily caused by an external cause, whether it is an agent, a
natural force, or an instrument (Levin and Rappaport Hovav 1995:91-3).
One major difference between the contentions of Smith and those of Levin and
Rappaport Hovav has to do with how we perceive the eventuality denoted by the ergative
use of a verb. For instance, with the sentence The vase broke, Smith claims that the vase
can break spontaneously with no intervention of an outside cause. Levin and Rappaport
Hovav, on the other hand, argue that our real world knowledge tells us that the vase could
not break without an external cause (1995:93). In their views, even in sentences like the
following, the eventualities are perceived as being brought about by some external cause,
which Levin and Rappaport Hovav identify with the Theme arguments plate and door
themselves (cf Kageyama 1996).
(47) a. The plate broke by itself.
b. The door opened by itself. (Levin and Rappaport Hovav 1995:88)
I would argue that the existence of an external cause is highly unlikely in sentences like
(47). Rather, it seems more natural to assume that, as Smith argues, eventualities
represented by ergative verbs could occur independently or spontaneously. In short, I
would claim that Smith's dual semantic features of independent activity and external
control differentiate ergative verbs from non-alternating causative change-of-state of
verbs like cut and destroy more properly.
3.4.4 Onset Causation vs. Extended Causation
Shibatani (1973a) notes that two types of causation are observed cross-
linguistically. When causation serves as an initial impulse so that the event denoted by a
given verb follows, this type of causation is referred to as ballistic. On the other hand, if
causation is responsible not only for the instigation of the event but also for its entire
process, then the causation is referred to as controlled. Drawing on the English verbs
send and bring/take, Shibatani describes the former as an example of ballistic causation
and the latter as that of controlled causation. While Shibatani does not specifically
mention the two types of causation in relation to ergative alternation, his theory lays the
foundations for subsequent studies of ergative pairs along this line (cf. McCawley 1976,
Van Voorst (1993, 1995) and Kiparsky (1997) utilize the two types of causation
to account for the alternatability of verbs. When transitive verbs are available for the
ergative alternation, the external causers of the verbs are ballistic, a causation type
characterized by Talmy (1985b) as onset causation. Under the concept of onset causation,
the agents or external causers occurring with ergative transitives merely initiate events,
lacking full control of the events ensuing after the initiation 14 By contrast, when
transitive verbs fail to alternate with intransitive uses, the verbs' external causers are
likely to control the events that follow, the other causation type characterized as extended
causation (Talmy 1985b). Under the concept of extended causation, agents or external
causers continue to participate in the entire process of an eventuality. 15 Based on the two
14 Due to the nature of the causation, the agents or external causers may also be referred
to as "instigator" (Wilkins and Van Valin 1993).
15 Maruta (1998:100) schematizes the LCSs of onset causation and extended causation as
(i) a. [x ACT ON y] INITIATE [y... ] (x = Initiator)
b. [x ACT ON y] CAUSE [y...] (x = Extended Causer)
What characterizes Maruta's templates are the use of 'INITIATE' as the connector
between the two sub-events and the characterization of the external argument x as
'Initiator,' while the connector for non-alternating causative verbs is simply 'CAUSE.'
differing types of causation, Kiparsky (1997) analyzes the contrasting behaviors
exhibited by smear and splash (cf. Hale and Keyser 1993, 1997).
(48) a. Mary smeared paint on the wall.
b. #Paint smeared on the wall.
(49) a. Mary splashed paint on the wall.
b. Paint splashed on the wall. (Kiparsky 1997:494)
In Kiparsky' s terms, smear is grammatically anomalous in intransitive constructions in
that smear denotes "a process requiring the initiation and continuous participation of a
causing Agent" (1997:495), while splash does not entail such extended causation. 16
Kiparsky adds that the same analysis also applies to the lack of the intransitive use in
other verbs like shelve, paint, ring, put, push, and kick.
In light of the distinction between onset and extended causation, the two external
causation verbs break and cut seem to be distinguishable with respect to the availability
of the ergative alternation. That is, the event of breaking is seen as that of onset
causation, continuing independently or autonomously after the initial physical force by an
external agent or cause. The event of cutting, on the other hand, is seen as that of
extended causation since some external agent or instrument needs not only to initiate the
event but also to continue participating in the entire process of the event. The concept of
onset and extended causation might also explain the lack of the intransitive use of destroy
since the verb requires incessant intervention of an external agent or causer until the
completion of the event (cf Maruta 1998:100).
16 Hale and Keyser (1993, 1997) analyze smear and splash in terms of the lexical
licensing of manner component (cf Section 3.5.2).
3.5 Lexical Specification
In this section I propose that the notion of agentivity is still a key factor in
accounting for the altematability of verbs. Instead of limiting our attention to
decomposing agentivity into multiple semantic components (cf. Section 126.96.36.199),
however, I suggest that agentivity is a minimal unit simply meaning 'volition.' Based on
this view, I would argue that agentivity can be associated with a variety of inanimate
entities such as instruments as long as they are used by volitional agents (Levin and
Rappaport Hovav 1995). Such association enables agentive cut verbs, which inherently
entail the use of instruments, to take instruments as subjects. Consequently, I will propose
that the specification of an instrument or a means involved in the realization of an event
is another semantic factor in determining whether a given verb of change of state can
undergo the ergative alternation.
3.5.1 Lexicalization of Instrument
188.8.131.52 Case theory
It has been noticed in the literature that the verb cut lexicalizes the use of an
instrument ('cutting device') during the course of the event denoted by the verb (Guerssel
et al. 1985:51-2, Hale and Keyser 1987:5, Brousseau and Ritter 1991, Levin and
Rappaport Hovav 1992:137, 1995:107, Levin 1993a, 1993b, Rappaport Hovav and Levin
1998:100ff). Traditionally, the use of an instrument associated with verbs has been
considered an important component of thematic roles. In a framework of Case Grammar,
Fillmore (1968b) recognizes a need to include the case of Instrumental in addition to the
cases of Agent and Object when schematizing transitive event verbs like open.17
17 In more modem terminology, Instrumental, Agent, and Object are comparable to
instrument, agent, and patient/theme, respectively.
(50) open: + [_ (I) (A)]
In (50), it is crucial that Instrumental and Agent, which are in parentheses, are optional
cases. Consequently, the following constructions are made possible.
(51) a. John opened the door with a key. ([ O I A])
b. John opened the door. ([_ O A])
c. The key opened the door. ([ O I ])
d. The door opened. ([_ ])
One crucial point of Fillmore's schematization is that the absence of a case results in the
absence of the argument associated with the case and accordingly the absence of the
meaning associated with the argument as well. Thus, the absence of the Instrumental key
as in (51b) and (51d) makes it almost improbable for us to interpret the whole sentence as
involving the use of a key.
One problem arises when Fillmore's schematization is applied to verbs like cut. In
fact, Langendoen (1970:72) describes the basic case structure of cut as follows:
(52) cut: Patient, Result, Instrument, (Agent)
Notice that in (52) Instrument is an obligatory case. Since Langendoen appears to follow
Fillmore in assuming that obligatory case means obligatory syntactic realization, the case
frame in (52) postulates that sentences containing the verb cut need to occur with some
form of instrument overtly. Thus, in considering the sentences The boy cut the cloth i/ ih
the scissors and The scissors cut the cloth to be grammatical, Langendoen follows this
principle. However, whether deliberately or not, he fails to mention that a sentence like
The boy cut the cloth is also grammatical, although no instrument is syntactically
expressed. In short, Fillmore and Langendoen are of the opinion that there are no
ambiguous relations between case roles and their syntactic realizations of arguments.
184.108.40.206 Lexical Conceptual Structure (LCS)
Instead of utilizing thematic roles, Guerssel et al. (1985) draw on the framework
of LCS, describing the indispensable association of an instrument with cut as follows:
(53) cut LCS: x produce CUT on y, by sharp edge coming
into contact with y
(Guerssel et al. 1985:51)
On their view, the use of a sharp instrument is posited as part of the inherent lexical
meanings of cut (See also Hale and Keyser 1986, 1987).18 In other words, even if a
cutting device does not syntactically occur with cut, the sentence implies underlyingly
that the event of cutting is caused by the use of a certain cutting device.
(54) a. Tom cut the cake in half.
b. I cut my face while I was shaving.
c. The phone wires were cut by the thieves.
18 In the schematization of cut, it is important to make a distinction between instruments
and natural forces. This may be illustrated by the fact that a cutting device can be the
subject of the verb cut, whereas natural forces cannot.
(i) a. The knife cut the bread.
b. *The lightning cut the clothesline. (Levin and Rappaport Hovav 1995:103)
Levin and Rappaport Hovav assume that agentivity is still the key to the eligibility of a
syntactic subject of cut. In their term, the use of a cutting device like a knife is naturally
associated with a volitional agent who has control over it. Schlesinger (1995) attributes
the acceptability of the instrumental subject to the additional feature CAUSE associated
While no instruments are overtly expressed in (54), each sentence implies the use of
some form of cutting device, such as a knife in (54a), a razor in (54b), and perhaps a pair
of scissors or a knife in (54c).19
On the other hand, the LCS of break does not contain such a delineation of a
cutting device (Guerssel et al. 1985:55):
(55) break LCS: x CAUSE (y come to be BROKEN)
Given the sentence Tom broke the vase, therefore, it is not obvious how Tom caused the
event; Tom could have broken the vase by hammering it or by knocking it off from a
table by accident, or perhaps he might have broken it by smashing it against the wall. The
crucial point to be made underlying the LCS in (55) is that in order for the event denoted
by break to take place the use of an instrument is not obligatory.
3.5.2 Specification of Cause or Means
As observed above, the key difference between Case Grammar and LCS
regarding the lexical representations of verbs is that LCS views the lexicalized semantic
features as being preserved throughout derivations. This position is more clearly
expressed by Levin and Rappaport Hovav (1994, 1995) under the notion of specification
(cf Van Voorst 1995). It is noted that verbs sharing certain lexically specified properties
syntactically behave differently from those, even if they appear semantically close, that
lack such specified properties. For instance, as pointed out by Gropen et al. (1991), the
verbsfill and pour are semantically relatively similar in that both describe, roughly
19 Ravin (1990:214) maintains that in a sentence like the following, cut does not specify
the use of any instrument:
(i) The broken window cut John's finger (on the interpretation of "John
brushes his finger against the glass")
Ravin explains that unless any autonomous causative argument exercising the action of
cutting is specified the verb is not considered to entail the use of an instrument.
speaking, the act of putting substance into some sort of container. They are observed,
however, to behave contrastively in the following ways:
(56) a. fill the glass with water
b. *fill water into the glass
(57) a. *pour the glass with water
b. pour water into the glass (Gropen et al. 1991:155)
One explanation offered by Gropen et al. about the contrastive behaviors is thatpour
specifies the manner in which a substance is transferred, whilefill does not necessarily
entail such a specific manner involved in the act (Gropen et al. 1991:160-1).
Building on the concept of specification, Levin and Rappaport Hovav (1994,
1995) explain the differing syntactic behavior of break and cut regarding the ergative
alternation. On their view, cut lexically specifies the use of a sharp instrument in its LCS
as shown in (53) above and its use should be implicitly, if not explicitly, presumed at all
subsequent levels of derivation. On the other hand, since break does not specify the use
of an instrument in the LCS, the manner in which the event of breaking occurs remains
unknown unless specified by some other means such as adjuncts. As mentioned in
Section 3.4.2 above, Levin and Rappaport Hovav (1995) attribute the lack of
specification of a causer or a causing event for break to the fact that break can take a
range of subjects such as instruments and natural forces.
On the same principle, Levin and Rappaport Hovav (1992b:131) discuss the
distinction between wipe verbs and clear verbs. That is, wipe verbs such as erase, mop,
rinse, scour, vacuum, and wipe, which lexicalize a manner or an instrument component,
do not alternate with intransitive uses, whereas clear verbs such as clear, clean, and
empty, which do not lexicalize a manner or an instrument component, do occur in
intransitive construction. Similarly, from the perspective of manner-instrument Hale and
Keyser (1993:90, 1997:54) and Kiparsky (1997:494) account for the unalternatability of
smear-class verbs, which contrast with alternating splash-class verbs.
Given the view of specification, I argue that in considering the contrast between
break and cut the concept of agentivity is too broad and ambiguous for us to capture.
Instead, my proposal is that the concept of specification is more straightforward in
differentiating the two verbs semantically.20 I will demonstrate later in this chapter that
the lexical specification of means explains the altematability of many change-of-state
verbs not only in English but also in Japanese. The generalization to be drawn from the
arguments provided so far might be that lexical specification of an instrument or a means
plays a key role in determining the syntactic realization of the argument structure
associated with an event verb.
3.6 Conditions for Ergative Alternation: Japanese
In the previous sections, we have observed that verbs that participate in the
ergative alternation in English are semantically characterizable to a large extent. The
question to be addressed in the remainder of this chapter is whether the semantic analysis
20 The notion of semantic specification described above does not explain all alternating
phenomena. There are still several verbs which appear to be evidence against our
account. The verb kill, which has raised a problem with the analysis of agentivity, still
resists our analysis of semantic specification (Brousseau and Ritter 1991:56-67). That is,
the verb does not specify a means or an instrument lexically whereby the act of killing is
performed. For instance, when we say Tom killed the senator, it is simply impossible to
determine how Tom carried out the act unless provided with additional information
contextually. In this respect, kill is comparable to break, but it fails to undergo the
ergative alternation (see Section 220.127.116.11 above). I would argue that the verb kill is a
matter of lexical idiosyncrasy with regards to ergative alternation.
provided for the English ergative alternation may also account for the alternatability of
verbs in Japanese.
3.6.1 Change of State
Like in English, many verbs that participate in the ergative alternation in Japanese
involve change of state (Miyajima 1972, Nishio 1978, 1982, Hayatsu 1987, 1995,
Jacobsen 1992, Mitsui 1992, Mitsunobu 1992). For instance, it is observed that
Jacobsen's (1992) Class I verbs contain several verbs of physical change of state, which
correspond to Levin's (1993a) break class verbs.21
(58) break verbs (Levin 1993a) Class I (Jacobsen 1992)
break war- ware-
tear (off) yabur- yabure-
smash kudak- kudake-
Furthermore, a number of Japanese de-adjectival verb pairs such as hukamar- hukame-
'deepen' and katamar-/katame- 'harden,' which usually express a gradual change of
state, belong to Jacobsen' s Class III. In short, one may state that change of state is the
key semantic property of Japanese ergative pairs.
Hayatsu (1995) emphatically illustrates this point by comparing two semantically
similar verbs kawakas- and hos-. According to Hayatsu, both verbs are transitive and
generally understood to mean 'to dry.' While the English dry ergativizes, only kawakas-
does so in Japanese.
21 Teramura (1982:271ff) does not consider the verbs of this group ergative pairs. Instead,
he refers to them as zihatutai inchoativeve voice'), arguing that the intransitive members
of the verbs of this group are derived from the transitive bases mediated by the suffix -e
(e.g., ware- <- war- 'break').
(59) a. Tom dried his clothes in the sun.
b. His clothes dried in the sun.
(60) a. Taroo-wa huku-o kawakas-ita. (< kawakas + ta)
Taro-TOP clothes-ACC dry-PAST
'Taro dried his clothes'
b. Huku-ga kawai-ta. (< kawak- + ta)
'The clothes dried'
(61) a. Taroo-wa huku-o hos-ita.
Taro-TOP clothes-ACC hang to dry-PAST
'Taro hung his clothes out to dry'
b. *Huku-ga hosat-ta. (< hosar- + ta)
clothes-NOM hang out to dry-PAST
'The clothes hung out to dry'
Hayatsu points out that hos- lacks its intransitive counterpart since it does not inherently
entail the resultant state of dryness, whereas kawakas- does (1995:179-180). This will be
illustrated more clearly by the following examples:
(62) a. Taroo-wa huku-o hosi-te kawakas-ita.
Taro-TOP clothes-ACC hang-and let dry-PAST
'Taro hung his clothes and let them dry'
b. *Taroo-TOP huku-o kawakasi-te hos-ita.
Taro-TOP clothes-ACC letdry-and hang out-PAST
'Taro let his clothes dry and hung them out'
Thus, it will be more appropriate for hos- to be translated into 'hang out (to dry).' In
short, the differing syntactic behaviors demonstrated by kawakas- and hos- in (60) and
(61) seem to point to the significance of change of state in determining the altematability
of verbs in Japanese. Following the contrast between kawakas- and hos-, Hayatsu
(1995:182) lists Japanese verbs that alternate in transitivity, noting that many of them
involve certain change of state or position.
(63) ore- breaking "
kire- "get cut off, severed"
tubure- "get crushed"
koware- breaking "
ware- breaking "
nobi- "get extended"
hirogar- spreadingn out"
nie- boiling "
koge- "get scorched"
nure- "get wet"
yogore-" get dirty"
somar- "get dyed"
kimar- "get decided"
sadamar- "get decided"
kie- "go out"
horobi- "go to ruin"
tae- "die out"
ori- "get off'
sagar- "get lower"
tore- "get taken, harvested"
nuke- "come out"
hazure- "come off'
hage- peelingn off'
hanare- "move away from"
hodoke- "come untied"
moge- "come off'
tuk- "adhere to"
kir- "cut, sever"
hiroge- spreaderr out"
nuras- "make wef'
nagas- "wash away"
oros- "let off'
tor- "take, harvest"
nuk- "pull out"
hazus- "take off'
hag- "peeltr off'
hanas- "separate from"
mog- "pluck off'
ire- "put in"