A model of conceptual structure mapping

MISSING IMAGE

Material Information

Title:
A model of conceptual structure mapping
Physical Description:
v, 132 leaves : ill. ; 29 cm.
Language:
English
Creator:
Belyayeva, Dina, 1967-
Publication Date:

Subjects

Subjects / Keywords:
Linguistics thesis, Ph. D   ( lcsh )
Dissertations, Academic -- Linguistics -- UF   ( lcsh )
Genre:
bibliography   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1997.
Bibliography:
Includes bibliographical references (leaves 124-131).
Statement of Responsibility:
by Dina Belyayeva.
General Note:
Typescript.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 028002193
oclc - 37823823
System ID:
AA00017623:00001


This item is only available as the following downloads:


Full Text










A MODEL OF CONCEPTUAL STRUCTURE MAPPING


By

DINA BELYAYEVA














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































Copyright 1997

by

Dina Belyayeva














ACKNOWLEDGMENTS

I would like thank Ira Fischler for many helpful discussions and constructive

suggestions. Special thanks are extended to William J. Sullivan, whose academic

guidance and friendship was instrumental in completion of this research study.

Furthermore, I thank J. P. Chalarca and Laura Cunningham for help in running

Experiment 4.














TABLE OF CONTENTS

page


ACKNOWLEDGMENTS ........................................ .............................. ii

A B STR A C T ....................................................................... ......................................... v

CHAPTERS

1 IN TR O D U C TIO N ............................. ..... ..............................................................1

2 BILINGUAL MEMORY RESEARCH....................... .........................2
Developmental Changes in Bilingual Representation ................................................ 12
Concept Acquisition and Representation................................. .......................... 14
Language-Specific Organization and Bilingual Representation........................ ..15

3 THE MODEL OF CONCEPTUAL STRUCTURE MAPPING................................16
The Structural Assum ption.................................................... ..........................17
The Developmental Assumption......................... ...... ......................26
Conceptual Structure Mapping ..........................................................28

4 EXPERIMENT 1: MODIFIED REPLICATION OF THE SHOLL ET AL.
(1995) STUDY .......................... ...................... ..............................31
M ethod .........................................................................................................................35
R esults............................................................................... .....................................38
D discussion ............................. .............................................................................44

5 EXPERIMENT 2: REPLICATION OF THE SHOLL ET AL. (1995) STUDY.........49
M ethod .................................................................................... .. ............................. 50
R esu lts............................................................................... .....................................52
D iscu ssion ................................................................................ ..............................57

6 RELATION ASSESSMENT: A TEST OF THE PROPOSED MODEL OF
CONCEPTUAL STRUCTURE MAPPING.................................................. 62
Experiment 3 .......................................................64
E xperim ent 4 ........................................................................ ..................................73
E xperim ent 5 ......................................................................... ................................. 79

7 GENERAL DISCUSSION .................. .............. ...................................91





iv


8 IMPLICATIONS FOR RESEARCH IN BILINGUALISM................................99

9 CONCLUSION...................... ..................................................................... 103

A PPEN D IX ............................ .................................................................... ................104

REFERENCES ... ................. ...................................................................124

BIOGRAPHICAL SKETCH .............................................................132













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

A MODEL OF CONCEPTUAL STRUCTURE MAPPING

By

Dina Belyayeva

August, 1997


Chairman: William J. Sullivan
Major Department: Linguistics

The present dissertation proposes a model of bilingual memory. The notion of

conceptual structure is introduced to denote a manner in which conceptual knowledge is

lexically represented in a language. The major claim here is that bilingual lexical and

conceptual processing is accomplished via conceptual structure mapping. The model

assumes that production patterns of a bilingual speaker are largely affected by the degree

of (in)compatibility between conceptual structures of two languages. A series of

experiments were designed and conducted in order to obtain supporting empirical

evidence. A discussion of the results along with a review of prior experimental work

show evidence for the position advanced here. In addition, it is argued that the model

proposed here can correct inadequacies observed in other models of bilingual memory

and that it can be extended to account for other issues of language production and

development.













CHAPTER 1
INTRODUCTION

Bilingualism has emerged as a discrete field. It developed out of studies in

cognitive psychology and linguistics investigating the relationship between language and

cognition. The growing body of knowledge on bilingualism has influenced the evolution

of models designed to capture the intricate symbiosis of lexical and conceptual

representations in the memories of bilingual speakers. Despite the increasing descriptive

accuracy achieved by some models, a number of problems remain unresolved. Transfer,

code-switching, and developmental issues are among them. The resolution to the problem

may lie in a different approach to representation. This work proposes a model of

conceptual structure mapping which will be reviewed as a dynamic approach to

representational issues. The model provides an account for all existing empirical evidence

and will accommodate the remaining problems, as well.












I In the literature the term 'bilingual' is applied in a very broad sense. It has been used to
refer to balanced/fluent bilingual speakers as well as to novice second language learners.
The term is also used to refer to individuals who speak more than one language.













CHAPTER 2
BILINGUAL MEMORY RESEARCH

Bilingual research literature offers a number of different approaches to the issues

of bilingual production and development. The largest group of studies on bilingualism

focuses on the issues of storage and representation. The central argument of the models

addressing storage issues is whether conceptual information is represented in language-

specific or amodal way. In particular, the studies have been trying to establish whether a

bilingual's two lexicons are represented in separate memory stores (the separate code

view) or whether they are represented in a single memory system (the common code

view). The extensive experimental research produced the data which was often

interpreted as supporting one view or the other (McCormack, 1977; Kirsner, Smith,

Lockhart, King, & Jain, 1984; Scarborough, Gerard, & Cortese 1984; Tzelgov & Eben-

Ezra 1992; Williams 1994). Durgonoglu and Roediger (1987) pointed out that the studies

often neglected the conditions under which language-dependent and language-

independent processes may influence the results. A series of experiments was conducted

to demonstrate that different types of tasks produce results that lead to different

conclusions. Specifically, data-driven tasks like word fragment completion and repetition

priming produce results that draw attention to the surface characteristics of the stimuli

and support the separate code view. Conversely, conceptually-driven tasks like free recall








and translation draw attention to the conceptual content of the stimulus and generally

produce results that support the common code view.

Heredia, Weldon and McLaughlin (1991) provided additional evidence that data-

driven tasks generate results supporting the separate code view whereas results obtained

in conceptually- driven tasks provide support for the common code view of bilingual

memory. Their first experiment replicated the findings ofDurgonoglu and Roediger

(1987). In the second experiment the generality of the previous findings was extended

even further. It was suggested that conceptual information might aid performance on a

data-driven task if it could be used to retrieve perceptual properties of the stimuli studied.

Therefore, the participants were given explicit instructions to use English fragments as

cues to generate Spanish translations after viewing monolingual and bilingual word lists.

Conceptual information, and as predicted the test instructions required the use of both

perceptual and conceptual information, the performance was reported to be comparable in

both within- and between-language conditions. The findings reported by Smith (1991)

provided additional evidence that it is possible to obtain cross-language facilitation in

traditional data-driven tasks. The absence of cross-language facilitation on word-fragment

completion tasks reported by Durgonoglu and Roediger (1987) and Watkins and

Peynircioglu (1983) was due to absence of semantic involvement at study, where Spanish

and English stimuli (in the former study) and Turkish and English stimuli (in the latter

study) were presented as random word lists. Smith (1991) used English and French

sentences that created associations with the stimuli presented as fragments during a word-

fragment completion task. The sentences did not use overt word forms, thus eliminating

the possibility of reliance on perceptual properties of the material studied. The results








demonstrated that sentence processing at study leads to semantic involvement at test and

hence to cross- language facilitation in a word-fragment completion task. As concluded in

Tobossi (1988), Smith (1991) and Hummel (1993), experiments using single-word

stimuli are more likely to demonstrate results supporting the independent view of

bilingual memory representation, because the study and test demand attention to surface

characteristics. On the contrary, use of contextually-embedded stimuli is more likely to

support the interdependent view, because contextual activation involves more abstract

levels of representation. Along these lines, it is important to note that availability of

contextual information should be regarded an important factor by the studies addressing

issues of lexical and conceptual access. Contextual embedding of lexical stimuli helps to

discriminate between different senses that may be activated upon presentation of a lexical

form, and to channel activation in a particular direction. The use of contextual

information is of particular importance for bilingual studies, since it provides a means to

distinguish between inter- and intralanguage organizational patterns.

The two competing views have been replaced by models that consider hierarchical

organization of conceptual and lexical representations in bilinguals. The dual-coding

theory (Paivio & Desrochers, 1980) was put forth to account for the discrepancies

between data-driven and conceptually-driven tasks in terms of variability of encoding

(Aredt & Gentile, 1986; Paivio, Clark, & Lambert, 1988; Vaid, 1988). Two systems

(verbal/symbol and conceptual/imagery) were proposed to be connected, yet independent.

On the basis of the findings that pictures were categorized faster than both L1 and L2

words, it was concluded that one system could be sometimes activated without the other.

That is, in picture categorization tasks conceptual representations could be accessed via








visual stimuli. Although the verbal system could be also activated, its activation was not a

necessary condition for conceptual access. The theory however has been criticized on the

grounds that it reduces conceptual representations to visual and verbal codes (Potter &

Kroll, 1987).

The hierarchical models proposed by Paradis (1981), Potter, So, Von Eckardt, and

Feldman (1984), and Snodgrass (1984, 1993) posit separate stores in which the conscious

representations of verbal and pictorial stimuli occur. Current research in the area of

bilingual memory has been centered around the two plausible models proposed by Potter,

So, van Eckardt, & Feldman (1984). The models differ in the connections they assume

between the conceptual and the lexical levels (see Figure 2-1). The word


Word Association Model


Concept Mediation Model


Figure 2-1. Word association and concept mediation models as proposed by Potter,
So, van Eckardt, & Feldman (1984).








association model postulates that the word meanings in the second language (L2) are

accessed via the bilingual's dominant language (LI), which is directly connected to the

conceptual store. The concept-mediation model assumes a direct connection between the

L2 lexical store and the conceptual store. The comparison of translation to picture naming

in L2 was viewed as a critical test for the models proposed. According to the word

association model, translation should take less time than picture naming in L2. That is, in

translation tasks, bilinguals need only associate the words stored in two different lexical

systems, whereas in picture naming tasks bilinguals first access conceptual storage,

mediate the concept with a lexical representation in LI and only then associate the L1

lexical item with a translation equivalent in L2. According to the concept mediation

model, both tasks require access to a conceptual level; therefore translation from L1 to L2

should take approximately the same amount of time as picture naming. 1

The experimental studies demonstrated a dissociation in performance of less and

more fluent bilinguals (Potter, So, von Eckhardt, & Feldman 1984, Kroll & Curley 1988,

Chen & Leung 1989). There was an asymmetry in translation latencies with longer times

taken to translate from L1 to L2, and significantly shorter translation times in the opposite

direction. As in picture naming studies, the differences between L2 and L output were

reported to be smaller for more fluent bilinguals than for less fluent bilinguals. Hence,

models were proposed to capture successive developmental stages in bilingual lexical

organization. The word association model was proposed to describe processing patterns

of novice bilinguals, who presumably mediate between the conceptual store and the L2



1 The latter prediction may be true if pictures cannot directly elicit their names.








lexical store via LI lexical representations. Conversely, the concept mediation model

supposedly reflects the ability of expert bilinguals to access concepts from either lexical

store directly. Some findings, however, were not consistent with the proposed dichotomy.

Significant asymmetries were reported even for very fluent bilinguals (Keatly, Spinks, &

Gelder 1990), whereas according to the models, fluent bilinguals should exhibit less

asymmetry in their production. De Groot and Nas (1991) reported equality of associative-

priming effects within and between languages, even though the overall response time

latencies obtained for L2 primes were larger than the latencies obtained for LI primes.

To accommodate the contradictory evidence, it was suggested that the relationship

between the word association and concept mediation stages in a bilingual speaker's

development is overlapping rather than strictly dichotomous (Kroll, 1993). The two

models have been amalgamated in the mixed model (Potter, Kroll, Yachzel, Carpenter, &

Sherman, 1986) that posits one common conceptual and a number of separate lexical

stores. The lexical stores are related indirectly via the common conceptual store and

directly at the lexical level of representation. It has been demonstrated that such a

configuration successfully accounts for the results which reveal both word association

and concept mediation processing patterns at the same level of bilingual proficiency.

The mixed model was later developed into the revised hierarchical model (Kroll

1993, Kroll & Stewart 1994, Dufour & Kroll 1995), which was proposed to deal with a

wider range of bilingual data. Specifically, it was designed to accommodate the

asymmetric patterns associated with bi-directional mediation between a bilingual's

lexical stores. The asymmetric patterns of the results have been associated with a

representational configuration which assumes that languages are connected via the








conceptual level when the direction of mediation proceeds from LI to L2, and that only

the lexical level is employed in the opposite direction (see Figure 2-2).


Figure 2-2. The revised hierarchical model as outlined in Kroll (1993), Kroll and
Stewart (1994), and Dufour and Kroll (1995). The thick solid lines stand for very
strong connections, and the dashed lines stand for weaker connections.



Kroll and Stewart (1994) employed transfer-appropriate processing logic to test

the representation proposed in the revised hierarchical model. According to the principle

of transfer-appropriate processing, memory performance is a function of the overlap

between study and test operations. In other words, performance on the test depends on the

overlap between the types of processing required by the encoding condition and the test

(Morris, Bransford, & Franks, 1977; Roediger & Blaxton, 1987). Since mediations from

L1 to L2 and from L2 to L1 were hypothesized to engage different activation patterns,







translations from L1 to L2 and from L2 to L1 were used to test the hypothesis.

Categorized and randomized word lists in L1 and L2 were used as stimuli. Categorized

lists were expected to draw attention to conceptual properties of the stimulus material,

and consequently, create interference in the tasks that require conceptual activation

(picture naming and translation from L1 to L2). On the other hand lexical-level

processing was predicted to have no effect on the tasks that do not require access to the

conceptual level of representation (word naming and translation from L2 to L1). L -to-L2

and L2-to-L1 translations of randomized and categorized lists were compared to word

and picture naming in L and L2. The findings supported the prediction that the two

directions of translation engage different interlanguage connections. Category

interference in bilingual translation was reported only when translation was performed in

the direction from L1 to L2. The results obtained from the incidental recall task also

agreed with the principle of transfer-appropriate processing. That is, the direction of

translation that was hypothesized to require concept mediation (Ll>L2) had a category

interference effect in production but a category advantage in recall. The direction of

translation that was hypothesized to be lexically mediated (L2 >L 1) was insensitive to the

effects of semantic context in production and also in recall.

The directional asymmetry was also demonstrated in semantic priming

experiments. Priming effects reported were significantly greater when primes were

presented in LI and targets were presented in L2 than when primes were presented in L2

and targets were presented in L1 (Neely, Keefe, & Ross 1989, Altarriba 1992, Kroll &

Sholl 1992). It was suggested that similar to Ll-to-L2 translation L1 priming of L2








targets initiates the interlanguage connection in the direction from LI to L2 and therefore

should result in greater semantic activation than L2 priming of L1 targets.

Additional evidence for the representational organization proposed in the revised

hierarchical model was provided by Sholl, Sankaranarayanan, and Kroll (1995). The

study they conducted examined the effects of prior picture naming on translation

latencies. According to the transfer-appropriate processing principle and the

representational organization proposed in the model, prior picture naming should affect

translation latencies only in the L -to-L2 direction, because only this direction was

proposed to require concept mediation and, therefore, only this direction could share the

same underlying semantic processes as picture naming. The results confirmed the

prediction. The significant facilitation in translation from L1 to L2 following picture

naming in both L1 and L2 and the absence of transfer in the opposite direction were

interpreted as evidence for the dissociation between the types of processes engaged, given

the direction of mediation between L and L2 bilinguals' lexical stores.

Since the retrieval of picture names did not affect translation times in the L2-to-

L1 direction, it was also proposed that the locus of transfer was not at the level of

retrieving the lexical form. The latter inference, however, is not fully supported by the

experimental data. The magnitude of facilitation in L -to-L2 translation was significantly

greater for the stimuli previously named in L2 than for the stimuli named in L1. The

observed difference cannot be explained in terms of greater conceptual activation during

L2 picture naming, because picture stimuli should produce equal conceptual facilitation,

regardless of the target language in the naming task. According to the principle of

transfer-appropriate processing and the representational organization of the revised








hierarchical model, both L and L2 picture naming should produce equal facilitation in

the task that requires activation at the conceptual level of processing, i.e. in translation

from L1 to L2. The data, however, suggests that some lexical activation is possible in the

direction from LI to L2. To equalize contradictory evidence, one could suggest that

shorter translation times following L2 picture naming could result from long-lag

repetition effects. In other words, accelerated production of L2 lexical forms in L -to-L2

translation tasks could be influenced by prior articulation of these forms in the L2 naming

task. This proposal does not contradict the mixture assumption of processing (Cabeza,

1995). The mixture assumption says that conceptual and perceptual processing should be

viewed in terms of two distinct continue that do not necessarily trade off against each

other (Roediger & McDermott, 1993). Wheeldon and Monsel (1992), suggested that

long-lasting facilitation could be a results of repeated exercise of the mapping from

meaning to phonological form. As demonstrated in the experiments conducted by

Monsell, Matthews, and Miller (1992), prior production of word forms in the naming task

and in response to a definition effected long-lasting repetition priming in the picture

naming task. This account of differential facilitation in Ll-to-L2 translation following L1

and L2 picture naming is compatible with the main assumptions of both the revised

hierarchical model and the transfer-appropriate processing principle, but it does not

explain why such an effect was not present in the L2-to-L direction. The discrepancies

found between the data and some essential theoretical assumptions cannot be resolved

within the representational organization of the revised hierarchical model.








Developmental Changes in Bilingual Representation


The studies that compare the performance of less and more fluent bilinguals in

picture naming, word translation, and lexical decision tasks reported a dissociation in the

response patterns of the less and more fluent bilingual speakers (Chen & Leung, 1989;

Kroll & Curley 1988; Altarriba, 1992). Less fluent speakers exhibited significant

asymmetries in response time latencies. Conversely, response time asymmetries obtained

from more fluent bilinguals were not significant. Similarly, more fluent bilinguals

demonstrated comparable priming effects in both directions (De Groot & Nas, 1991).

Dufour and Kroll (1995) proposed that there is a developmental shift in the way

bilinguals mediate between their two languages. As bilinguals become more proficient,

they develop more conceptual nodes that can be activated by both L1 and L2 lexical

items, in which case they gradually switch to conceptual mediation in both directions (see

Figure 1-2). The developmental proposal was expected to demonstrate a degree of

empirical adequacy, since it was developed from translation latency data obtained from

less and more fluent bilinguals in several previous studies (Potter et al. 1984, Kroll and

Curley 1988, Chen and Leung 1989). However, the results reported by Kroll and Curley

(1988) show that a switch to conceptual mediation may occur quite early, long before

novice bilinguals achieve production fluency.

Additional evidence for early conceptual involvement in novice bilinguals was

provided by Griffin and Harley (1996). They examined effects of the direction of

association on word list learning. Four groups of novice English-French bilinguals were

given a list of words which two groups studied in the L -to-L2 direction and another two








groups studied in the L2-to-L1 direction. The groups performance was later tested in

translation tasks from L1 to L2 (a production task) and from L2 to L1 (a comprehension

task). Conditions at encoding and test were partially counterbalanced. It was found that

matching encoding and retrieval conditions and using LI as a target language

(comprehension) were particularly favorable for a better performance at test, whereas

direction of learning (i.e. direction of association) had no significant effect on group

performance. A comparison between the two directions of learning over the

comprehension and production tasks showed that word association learning from L1 to

L2 was less disadvantaged by the demands made on it under the reverse condition (i.e.

translation from L2 to L1) than was the L2-to-L1 learning condition. In other words, it

was demonstrated that in word-association learning, a more stable performance at test

resulted when learning proceeded in the direction associated with conceptual processing.

According to the revised hierarchical model word association in the direction

from L2 to LI is achieved at the lexical level of processing and should provide a stronger

interlanguage connection for novice bilinguals than the direction from L1 to L2 which is

considered to be established at the conceptual level of processing. Furthermore, it was

suggested that L2-to-L1 lexical connections are stronger than those from LI to L2,

because "second language words are frequently taught by associating them to first

language (e.g., casa means 'house') but not vice versa" (Kroll 1993, p. 70). The findings

of Griffin and Harley (1996) were not compatible with the above assumptions. It was

demonstrated that direction of association was not a crucial factor affecting bilingual

performance at test as compared to matching of encoding and retrieval conditions.

Additionally, it was shown that successful retrieval of a lexical form largely depends on








how well a given lexical representation is established in the memory of a bilingual

speaker. The latter evidence suggests that the issue of directionality is relevant as long as

there is a significant developmental asymmetry in LI and L2 representations. It also

suggests that the dichotomy between lexical and conceptual processing in the bilingual

memory may not be as strict as defined in the reviewed models of bilingual lexical and

conceptual representation.

Concept Acquisition and Representation


The bilingual research paradigm has concentrated on representational issues

without regard to the issues of lexical and conceptual development. A developmental

approach may provide insights to the problem of the bilingual representation, because it

considers representation as the relations between conceptual and lexical information

within a functional context. The findings from word and category learning studies suggest

that concept acquisition cannot be considered without social-pragmatic and functional

contexts (Tomasello 1996). Social-pragmatic contexts are closely associated with

activities and routines such as schemas and scripts, which provide a framework for

emerging functional and thematic categories. As reported in object-sorting studies

(Fivush, 1987), 14- and 20-month-old infants tend to select objects sequentially in the

same script category (kitchenware vs bathroom accessories) from the total set of objects

available to them at a greater than chance level probability. Although the script-based

categories are assumed to be functional, many of the categories develop into traditional

taxonomic categories (e.g. furniture, clothing). Findings from semantic priming studies

showed that categorical organization constitutes an important factor in lexical and








conceptual access (Collins & Quillian 1969, Meyer & Schvanveldt 1971, Collins &

Loftus 1975).

Language-Specific Organization and Bilingual Representation


Variations in socio-cultural routines affect the relative structure of the knowledge-

based constructs that determine lexical-conceptual activation patterns. Mappings of

lexical-conceptual relations onto knowledge-based constructs are lexicalized in the ways

that may not be shared across languages. The following often-cited example from Talmy

(1985) demonstrates different lexicalization patterns of conceptual primitives such as

MOTION, PATH, FIGURE, GROUND and MANNER in English (1) and Spanish (2):

(1) The bottle floated into the cave.

(2) La botella entro flotando en a la cueva.

English was said to conflate MOTION and MANNER into a single lemma "float",

whereas in Spanish speakers use separate lemmas to convey these conceptual primitives.

However, in Spanish the lemma "entro", which denotes MOTION, also makes reference

to PATH and GROUND. The implication is that for Spanish-English bilinguals Spanish

lexicalization patterns are associated with a lower activation threshold than contrasting

English lexicalization patterns, unless the contrasts are highlighted by relevant contextual

information. Since social- pragmatic contexts specify encoding and retrieval conditions,

they may be used as a point of reference for matching lexical and conceptual

representations in lexical access. Hence the relation between lexical-conceptual mappings

and knowledge-based constructs provides a valuable theoretical component that has not

been considered by the models of bilingual lexical representation and storage.













CHAPTER 3
THE MODEL OF CONCEPTUAL STRUCTURE MAPPING

The purpose of the present study is to propose a model of bilingual memory that

considers traditional issues of lexical and conceptual development like concept formation

and word learning as central to the issues of storage and representation. The model claims

that the representational organization of bilingual memory arises from relations between

conceptual and lexical information in given socio-pragmatic and functional contexts.

The proposed model introduces the notion of conceptual structure as its major

functional component. Conceptual structure can be succinctly defined as a set of

knowledge-based constructs such as frames, domains, and schemas that accommodate the

conceptual knowledge and lexical items of a language in a way which reflects language-

specific lexicalization patterns. The configuration of a given conceptual structure is

determined by a number of contributing factors, e.g. the perceptual abilities of a language

learner, the socio-pragmatic context of concept-acquisition episodes, and the available

linguistic means. Stable cooccurrence of these factors determines a central tendency in

the formation of a conceptual structure. This central tendency ensures that the speakers of

a given language community possess highly compatible conceptual structures. Instability

within the structure is associated with variability of the aforementioned contributing

factors and with different degrees of entrenchment of particular structural elements within

the memory of an individual speaker. The degree of entrenchment of a particular part of








the structure may be specified in two ways. It may be specified in terms of the overall

frequency of activation which accumulates during the lifetime of the structure or in terms

of contextual saliency of recent activation episodes.

Conceptual structure constitutes a heuristic device that explicates observed

activation patterns in terms of a relational network organization of conceptual and lexical

material. The notion of conceptual structure and its functions are explained in the

structural and developmental assumptions elaborated below.

The Structural Assumption


The key components of the structural assumption draw heavily upon the theories

and empirical findings of cognitive semantics and cognitive psychology. The present

description of the model is not fully formalized. The formal mechanisms assumed include

a relational network storage system with spreading activation. These are compatible with

several theories, e.g. Lamb's stratificantional theory and Langacker's Cognitive

Grammar2, but the model I use herein is not fully specified.

One component of the structural assumption encompasses the same structural

principles as the ones developed in Langacker's relational-network (RN) model.3 The

RN model was proposed to capture the relationships that hold between various senses of





I Lamb 1966, 1971.
2 A more comprehensive review of Cognitive Grammar can be found in Langacker 1987
and Rudzka-Ostyn 1988.
3 Langacker's model is neither the first nor the most highly elaborated one available, but
it suffices as a starting point for the preliminary studies herein.








lexical items and to provide a theoretical construct for the prototype model of

categorization. 4

The RN model maintains that the senses of lexical items (= relational

connections) in a network are determined by two types of relationships: schematicity and

extension. Schematicity represents the degree to which a particular sense can be defined:

a finer-grained instantiation or specialization, and a more approximate representation or

abstraction. An extension is a relationship in which some basic specifications are

suspended or ignored to form a new meaning.

In the example provided by Langacker (1988), a relational network of the lexical

item ring was modeled to demonstrate relationships of schematicity and extension. Thus,

the meaning 'a circular piece ofjewelry' can be said to represent a specialization of a

more abstract sense 'a circular object'. The relationship between the meaning 'arena' and

the abstraction 'a circular object' represents an extension, in which strict geometric

specifications are suspended and only functional specifications remain relevant (see

Figure 3-1).

Langacker's relational networks were introduced primarily to address issues of

category membership and to account for prototype effects. Although no specific

configuration was required by the facts, the RN approach addressed only functional-

similarity relation between senses in a network. A more varied approach to access and

activation issues that permits any analogical connection between nodes in relational



4 Prototype Theory and the notion of prototype in its current interpretation have been
developed in a series of experiments and papers by Rosch (1973, 1974, 1975a, 1975b,
1978).













ARENA |
ULARI ------ -


Figure 3-1. A relational network of the English lexical item ring proposed by
Langacker (1988). Prototypical meaning is given in bold. Dashed lines point to sense
that are extensions of the corresponding abstractions and specializations.



networks was not considered. Consequently, the networks were given a rigid structural

organization, which can not account for the instability in activation patterns observed in

different contexts. Various phenomena associated with instability in graded categories

have been largely explored in the works of Barsalou (1983,1985, 1987). Barsalou

maintains that all categories, formal, linguistic and goal derived, possess a graded

structure. Graded structure reflects a degree of activation during access. Graded structure

can vary substantially with changes in linguistic context and in point of view; that is,

particular concepts may become more or less accessible in various contexts.

Unlike the relational network model, the frame-based organization proposed by

Barsalou (1992) employs an extensive knowledge base to account for effects arising from








the instability of graded structures. Barsalou argues that it is implausible to maintain that

the wide range of graded structures observed are all stored in long-term memory. He also

proposed that people can construct a wide range of concepts and relations for the same

category in working memory. Figure 3-2 provides an example of the frame for the

concept 'vacation'. The frame provides a wide knowledge base that contributes to

concept formation. Depending on the context, personal goals, or point of view, people

incorporate different information from long-term memory into the current concept that

they construct for this category.


Figure 2-2. Example of a frame for 'vacation' proposed by Barsalou (1992).



Barsalou's approach to categorization has been criticized for its rather

unconstrained nature (McCauley 1987). Indeed, it is unlikely that such highly structured








body of knowledge as language would be without any stable organization and would

depend largely on temporary constructs composed in working memory.

The present study employs both an RN and a frame-based approach to produce a

theoretical construct that could constitute a stable, yet dynamic representational

organization. Such a representation is needed to account for lexical-conceptual activation

patterns that are both persistent and unstable. A conceptual structure notion is proposed to

accommodate this need. Organizationally, a conceptual structure is best rendered as a

network of relational networks. Figure 3-3 captures a part of the conceptual structure of

the English language which may be activated upon a presentation of the lexical form ring.

The multi-tiered network depicted incorporates both approaches: a network of interrelated

senses and a frame-based organization. Tiers constitute frames of reference in which only

one particular sense in the relational network can be fully specified. The connections

between senses of lexical item resemble relations proposed in the RN model, whereas

relations between senses within a given tier/frame of reference are similar to the ones

proposed within a frame-based approach.

The organization of tiers given in Figure 3-3 is arbitrary and is dictated partly by

limitations of the two dimensional format. The organization of tiers reflects activation

patterns which are largely influenced by such factors as socio-pragmatic context,

frequency of activation, and degree of entrenchment. These factors are also likely to

affect conceptual structure formation for each speaker. The common abstraction 'circular

entity' is schematically represented between tiers so as to provide a connection between

various frames of reference. The common abstract representation constitutes a

conceptualization of a relation between an entity/concept and the environment of its












































Figure 3-3. A schematic representation of a part of the English conceptual structure
associated with the lexical form ring.



application. This relation is not a perceiver-independent property of an entity taken alone,

but rather a conceptualization which has human information processing consequences.

That is, it is psychologically defined. Similar ideas about the conceptualization

mechanism were presented in terms of affordances by Gibson (1966,1979) and further








developed in the approach of psychological essentialism as discussed by Medin (1989).

These approaches to conceptualization, however, do not have clearly defined theoretical

constructs which can capture conceptualization mechanisms pertinent to lexical

processing.

Langacker's (1987) theory of Cognitive Grammar offers categorizing

relationships of schematicity and extension that can gives rise to a number of meanings

when applied to different contextual frames. As depicted in Figure 3-3, the abstract

common representation 'circular entity' can be conceptualized as a piece of jewelry worn

around a finger or as a circular mark, depending on the degree of schematicity required

within a given frame of reference. Contexts that require the suspension of certain basic

specifications constitute the basis for extensions. For example, the frame of reference for

the meaning 'arena', emphasizes particular functional properties and suspends limiting

geometric specifications. The basis for metaphorical extensions is the application of

abstrations to contextual frames different from those they were conventionalized in

(Langacker, 1987; Lakoff, 1987). The most common example of a metaphorical

extension is the representation of temporal relations in spatial terms. For example,

"behind" has a meaning "in the past" when used metaphorically. While differences

between frames of reference have been considered to be the major source of polysemy

(Fillmore & Atkins, 1992).

The structural assumption provides a lexical-conceptual basis for the mechanisms

which govern lexical processing. Thus, activation patterns between tiers are determined

by (1) similarity of other senses to an abstraction, and (2) similarity between frames of

reference. Similarity relations between senses can be established at the basic level, where








judgments primarily involve perception and interpretation of our immediate

environments. Many relations between concepts are so closely tied to looks and

affordances that they seem at first to be perceptually given (Keil, 1987; Mervis, 1987;

Markman, 1987). In other words, people adopt an essentialist heuristics; that is, the

hypothesis that things that look alike tend to share deeper properties (Medin &

Wattenmaker, 1987). Similarity between frames of reference can be established on the

basis of common functional properties that can be realized within appropriate contexts. In

Figure 3-3, such similarity is captured by greater number of connecting lines between

tiers.

Similarity between abstractions and specializations facilitates activation spreading

between the senses (represented in bold lines in Figure 3-3), whereas differences between

contextual frames inhibit such activation. The greater the functional difference between

frames, the longer it takes to establish a relation between the senses of a given item in the

network. For example, the abstraction of a circular entity establishes a connection

between the specializations 'piece of jewelry worn around finger', 'any circular piece of

jewelry', 'circular mark', and 'arena'. Similarity between the domains of application of

the first two senses permit an immediate activation of these two nodes in the network.

The contextual frames which are used to denote a piece of jewelry and a circular mark are

less similar; however, the relationship between the two may be established on the basis of

perceptual salience that is reflected in their closer connection to the common abstraction

'circular entity'. Therefore, it should take a somewhat longer time to activate the meaning

'circular mark' following activation of the meaning 'a piece ofjewelry'. It requires an

even longer time to access the meaning 'arena', because the contextual frame which








constitutes the domain for this meaning bears no resemblance to the other frames of

reference.

The proposed representational organization can accommodate effects resulting

from instability within the graded structure of a lexical category influenced by linguistic

contexts, socio-pragmatic situations, and current goals. These and other linguistic, socio-

pragmatic and environmental factors have been demonstrated to be major sources of

instability affecting category structure and prototype effects (Barsalou, 1983, 1987;

Lakoff, 1987). Thus, retrieval patterns of non-prototypical senses, which are generally

characterized by longer response latencies and lower frequency of recall, may be greatly

affected by the activation of relevant contextual frames. For instance, the activation of

such concepts as 'boxing' and 'tree trunk' may provide a speeded access to less

prototypical senses in the RN representation of the lexical item ring.

The notion of conceptual structure provides a means of accounting for the

instability of graded structure in a lexical category as a result of context effects. The

salience of a particular contextual frame affects the prototype status of the senses in a

network and, subsequently, defines the activation pattern that follows the presentation of

a lexical form in a meaningful context. Context availability is a factor in focusing the

activation pattern that follows the presentation of a lexical form. The focus of the

activation might even be reduced to a single tier. Hence, the model of conceptual

structure mapping predicts that contextual embedding of a lexical form will facilitate a

more direct access to the relevant node in the network. Facilitated access can be

demonstrated empirically by reduced response latencies and a greater proportion of

responses associated with a particular target sense in a given relational network.








The Developmental Assumption


The notion of conceptual structure allows us to consider lexical-conceptual

organization as a result of the lexical and conceptual development of an individual

speaker. The processing patterns underlying concept formation and language acquisition

are of particular importance, since they determine the possible activation patterns in

lexical access. The development of a conceptual structure can be depicted in terms of the

growth of individual relational networks and inter-network connections following natural

routes of lexical and conceptual development.

Conceptual development is one of the facets of human cognitive ability to

compare two events, noting discrepancies and categorizing similarities. This ability is

fundamental for word and category learning. The use of contrastive contexts has been

considered a major means of word learning (Tomasello 1996). A similar principle

contributes to developments of relational networks of the acquired lexical items:

contrastive context provide information about different senses of the same lexical item.

Understanding those contrasts enables a language learner to form a relational network for

a given lexical item. The developmental assumption maintains that bilingual speakers

possess a number of conceptual structures that may be at different stages of development,

as a function of the individual history of the acquisition of the speaker's two languages.

Depending on the extent of the primary conceptual development of an individual speaker,

conceptual structures may develop simultaneously or sequentially. Thus, bilingual

children develop two conceptual structures simultaneously by associating them with

particular lexicalization patterns provided by separate concept-acquisition episodes.








Simultaneous formation does not condition parallel development of L and L2 conceptual

structures (henceforth CSI and CS2) in early bilinguals.

Lexical and conceptual representations of early bilinguals are particularly context-

dependent. Context dependence provides conditions for the complementary development

of conceptual structures with some degree of overlap. The non-overlapping areas are

associated with impeded ability to relate concepts expressed in different languages. The

complementary development of conceptual structures provides a condition for transfer-

type errors and code-switching.

Unlike bilingual children, who are in the process of developing both of their

conceptual structures, adult L2 learners already have a well-establish conceptual system

associated with CS1. In this case they have a solid conceptual system to rely on in the

course of CS2 development. Greater reliance on CSI in L2 learning has been

demonstrated by the findings that adult learners are less inclined to use formulaic

expressions. Instead of learning conventionalized L2 idiomatic expressions, they apply

L extensions to L2 representation, producing transfer-type errors (Wong-Fillmore 1976,

Yorio 1989). CS2 development in adult bilinguals can be captured in terms of mappings

of L2 lexical items on CSI representations. As L2 learners extend their L2 corpus within

the context that provide them with contrastive evidence about L2 lexicalization patterns,

CS2 begins to develop. Isolated L2 representations that were initially mapped onto

convergent CS nodes, begin to evolve their own relational networks that ultimately lead

to CS2 formation.








Conceptual Structure Mapping


Similarity in the socio-pragmatic context of concept formation and a common

pool of linguistic means ensures a similarity between (not absolute identity of) conceptual

structures formed in the same speech community. Conceptual structures developed in

different speech communities may differ (1) in the degree of schematization permitted

between senses in a network, (2) in the sets of contextual frames that provide bases for

the network senses, or (3) in profiling patterns5.

The first condition can be illustrated by the following example. The concepts

'ladder' and 'staircase' are not differentiated in the Russian language. They share a

common abstraction: a means for upward movement. Specifications that pertain to

functional differences between a ladder and a staircase can be suspended. Hence, the

concepts relate different parts of the network to a single lexical item lestnica. Which

concept is activated when lestnica is heard depends on the functional context, which is

very narrowly defined. On the other hand, in English, the functional difference between

the two concepts is central and cannot be suspended. As a result, two lexical items are

used to communicate these two concepts.

A relational network of the Russian lexical itemprobka can be used to illustrate

the second condition. The prototypical sense of this lexical item can be conceptualized in

English as 'plug, stopper, cork', whereas in Russian it is a single indivisible



5 Cognitive Grammar posits hierarchies of domains to provide the basis for various
concepts. The parts of the domains that a linguistic unit invokes are called the base. The
notion of a profile is used to indicate that some facet of the base is raised to a prominent
level (Langacker 1988, pp. 53, 59).








conceptualization of a device that keeps liquid in a container. The relational network of

the Russian word probka is characterized by a considerably larger set of applicable

contextual frames. For instance, the common abstraction of the network senses can be

extended to denote a traffic jam. Although English borrows terms from the same semantic

field to denote congested driving conditions (e.g. bottleneck), the event is conceptualized

differently in the two languages. While in English it is conceptualized as a condition

associated with a too narrow passage, in Russian it is conceptualized a result of some

obstruction in a passage of whatever size.

Linguistic predications in LI and L2 may differ in profiling patterns that are

imposed on a base. For example, the Russian word ruka denotes a concept that conflates

the concepts activated by the English words 'hand' and 'arm'. The contrasting profiling

patterns of the English words are evident from the inappropriate pictures evoked by

sentences (2) and (4). However, the profiling pattern of the Russian word ruka renders the

Russian equivalents of all four sentences appropriate.

(1) She has a child in her arms.

(2) She has a child in her hands.

(3) She has a pen in her hand.

(4) She has a pen in her arm.

Note, however, that there are only two Russian sentences involved: U nee rebenok v

rukax translates both (1) and (2) and Unee rucka v ruke translates both (3) and(4). In

short, some transfer-type errors may result from the bilingual's failure to recognize

contrasts in one-to-many profiling patterns of L1 and L2.








As shown, the conceptual structures of two different languages may have

networks which are highly compatible, less compatible, or even incompatible.

Compatible areas are associated with perceptually salient domains of human cognition.

Thus, the network representation of concrete nouns has been demonstrated to be highly

compatible in many languages. The results obtained in semantic priming (Jin & Fischler

1987, de Groot& Nas 1991) demonstrates that a greater cross-linguistic priming effect is

characteristic of concrete words, whereas greater language-specific variations are

characteristic of abstract words. Incongruities resulting from mapping incompatible areas

in two conceptual structures account for performance deficits in bilingual production, e.g.

longer response times and lexical transfer.

The proposed model contends that conceptual structure mapping provides a

functional account of the lexical-conceptual representation of the bilingual memory. The

model does not posit a strict dichotomy of lexical and conceptual level processing.

Instead it uses the notion of a structured relational network to capture the intricate

relations between lexical and conceptual information. The mapping metaphor represents a

link between the two conceptual structures of a bilingual and helps to explain the

activation patterns observed in bilingual production.













CHAPTER 4
EXPERIMENT 1: MODIFIED REPLICATION
OF THE SHOLL ET AL. (1995) STUDY

The revised hierarchical model of bilingual memory proposed by Kroll and

Stewart (1994) maintains that the language-specific lexical representations of a bilingual

are linked at both lexical and conceptual levels. Specifically which level is activated

depends on the direction of interlanguage connections. If mediation between languages

proceeds in the direction from L1 to L2, the languages connect at the conceptual level,

where amodal conceptual representations of lexical items are activated. If bilinguals

mediate in the direction from L2 to L1, connections between L2 and L1 words are

established at the lexical level. Such connections provide direct lexical links between L1

and L2 stores and do not require activation of amodal conceptual representations.

Sholl, Sankaranarayanan, and Kroll (1995) used transfer-appropriate logic to

provide support for the configuration of connections outlined in the revised hierarchical

model. According to the principle of transfer-appropriate processing, memory

performance is a function of the overlap between study and test operations. The revised

hierarchical model predicts that conceptual activation should affect bilingual performance

when the direction of mediation between bilinguals' two languages proceeds from L1 to

L2, but not in the opposite direction. Effects of prior picture naming on translation times

from L1 to L2 and from L2 to LI were examined to provide evidence supporting the

types of connections proposed in the revised hierarchical model of bilingual memory. The








results demonstrated that the magnitude of facilitation following picture naming in both

L1 and L2 was greater when the translation was performed from LI to L2 than in the

opposite direction. The reported effects were interpreted as resulting from transfer that is

possible between the tasks that have similar processing requirements. Translation and

picture naming were thoroughly investigated in a number of studies (Potter, So, von

Eckhardt & Feldman 1984, Glaser 1992, Snodgras 1993). Both tasks were assumed to

share lexical and conceptual processing requirements. As follows from the revised

hierarchical model, processing requirements in translation differ, depending on the

direction of translation. Since it was proposed that translation from LI to L2 is a

conceptual-level task and translation from L2 to LI is a lexical-level task, the facilitation

observed in L1-to-L2 translation following picture naming was interpreted as evidence of

transfer that is possible between conceptually-driven tasks. On the other hand, the

absence of any significant effects on L2-to-L1 translation latencies was taken as an

indication of a different type of processing, which was assumed to be lexical. In addition,

it was suggested that the locus of transfer was not at the level of retrieving the lexical

form, because production of picture names had no effect on L2-to-L 1 translation

latencies.

One purpose of the present experiment was to obtain additional evidence

regarding the locus of transfer. Since picture naming provides both conceptual and lexical

activation, the observed transfer effects might have resulted from both. It is necessary to

determine exactly what effects) each of these causes have by itself. We must therefore

dissociate lexical and conceptual factors at the encoding stage of the experiment. For this

purpose lexical decision and object identification tasks were substituted for picture








naming. In a lexical decision task one should identify whether a string of letters

constitutes a word. Presenting translation stimuli in a lexical decision task was supposed

to ensure prior activation of the target lexical representations. An object identification

task was assumed to provide the same amount of conceptual activation as picture naming

(Kroll & Potter, 1984) without the explicit production of a lexical form. In the object

identification task reported in the Kroll and Potter (1984) study participants were required

to identify whether a drawing represents a real object. However, it is possible that in such

a task conceptual representations of easily identifiable objects may not be sufficiently

activated. To ensure sufficient activation, an object sorting task was used. In this task all

stimuli were unambiguous pictures of common objects. A participant was instructed to

decide whether an object could be found in her or his home. Pictures were sorted into

three piles according to the type of a decision provided ("yes," "no," and "don't know").

The principles employed in an object sorting task are very similar to the ones used

in categorization. Traditionally, categorization was assumed to be a task that could suit

this purpose (Dufour & Kroll 1995). Unlike picture naming, categorization does not

require the explicit production of a lexical form that is associated with a particular

concept. However, categorization can not be considered a purely conceptual task,

inasmuch as it establishes a lexical link at a comprehension level. In order to perform this

task participants should activate lexical representations of category names. The task may

create language bias by using a particular language to name categories. Although it may

be argued that providing category names in both languages could constrain the exclusive

access to a single lexical store, parallel presentation of Ll and L2 category names could

trigger activation of lexical-level connections. In other words, bilinguals may be








encouraged to perform an implicit matching of L1 and L2 lexical representations of a

target concept. Nevertheless, language bias may be due to cross-linguistic differences in

categorization patterns. This bias is closely linked to culturally- or experientially-bound

instances of language use. For example, native speakers of American English tend to put

corn in a vegetable category, since one of its most common functions fits the profile of a

vegetable, while Russian native speakers would be more inclined to categorize corn as a

grain. In a case when a target stimulus may represent different categories in a bilingual's

two languages, the availability of one but not the other category name can condition the

access to a particular lexical store. Careful selection of experimental stimuli and category

names may help to eliminate language bias, yet it would not eliminate implicit activation

of lexical-level connections. Hence, an object sorting task could be viewed as a better

task to provide sufficient conceptual activation with less lexical interference.

The other purpose of the present experiment was to examine whether transfer was

the only factor affecting translation accuracy of the stimuli previously named as pictures.

According to the transfer-appropriate logic employed in the Sholl et al. (1995)

experiment, translation responses that were facilitated by prior picture naming should be

exactly the same as naming responses. Although all experimental stimuli were concrete

nouns of high imageability, many of them were polysemous lexemes that could trigger a

number of translation responses in both languages. The model proposed here maintains

that activation of a particular sense associated with a stimulus lexical item depends on a

number of factors such as (1) a bilingual's familiarity with a target sense, (2) its prototype

status in a given relational network, (3) differences in relational organization of

equivalent lexical items in LI and L2. The model predicts that transfer from the study








task may not be the only factor affecting bilinguals' performance at test. The choice of a

lexical form in translation may be affected by each of the aforementioned factors as well.

For instance, if a Russian-English bilingual is not familiar with a target sense (e.g. 'screw

nut') of the English stimulus lexical item nut, the non-target Russian form oreh 'kernel'

will be produced in response. Therefore, it is critical to distinguish between translations

that are the same as the names activated by primes (henceforth, primed translations) and

translations that fail to produce target/primed lexemes, yet could be considered correct

responses (henceforth, unprimed translations). Since such a distinction was not made in

the original study, it is not clear whether the reported effects could be a result of transfer

or could be triggered by other processes. Separate analyses of both overall and primed

translation data are required before any substantial claims could be made.

Method


Participants

Twenty-eight Russian-English bilinguals participated in the experiment. All

participants were native speakers of Russian. The age of the participants ranged from 18

to 42. On average, these participants had studied English for 11.2 years, and had been in

an English speaking environment for 4.7 years. At the beginning of the experimental

session, participants were requested to fill out a questionnaire about their language

learning experiences and were also asked to evaluate their L2 proficiency. Participants

noted their dominance in Russian. They did not consider themselves as balanced

bilinguals. They rated themselves at an advanced level for most aspects of L2 production








and at a near-native level in some areas of L2 production and most areas of L2

comprehension. No compensation was offered for participation.

Materials

The base materials were 170 words selected from the Snodgrass and Vanderwart

(1980) picture-word norms and 20 non-words generated on the basis of 10 English and 10

Russian words. There were two priming conditions and two translation conditions.

Picture primes (40 target and 10 filler pictures) were presented in the object sorting task,

and word primes (80 target words: 40 Russian and 40 English, and 20 non-word filler

items) were used in the lexical decision task. Eighty 80 Russian and 80 English words

were used in each translation condition. Out of 80 translation stimuli, 20 words were

previously shown as pictures, 20 words were seen in L1 (Russian), 20 words were

presented in L2 (English) and 20 words were new/unprimed stimuli. All translation

stimuli were partially counterbalanced across four priming conditions (picture primes, L

primes, L2 primes, new stimuli/unprimed) and two language conditions (translation from

L1 to L2 and from L2 to L1). The experimental stimuli and the conditions in which they

were presented are listed in Appendix (Tables A-1 and A-2).

Apparatus

The experiment was conducted on an IBM compatible notebook computer with a

24 cm color active-matrix display. The computer ran a program written in Quickbasic.

The instructions and the stimuli were presented in white 0.5 x 0.5 cm letters on black

background. Picture stimuli were presented on flash cards. All responses were oral.

Responses were audio recorded so they could be later checked for accuracy.








Procedure

Since priming conditions were blocked in the original study (naming in LI and

L2), priming conditions were also blocked in this experiment. Participants were presented

with a block of picture primes and a block of word primes before translation. Blocks of

picture and word primes were counterbalanced across participants. To ensure proper

activation of picture prime stimuli participants had to decide whether an item presented

in a picture could be found in their home. They were requested to sort pictures into three

piles according to the type of the response they could provide: "yes," "no," "don't know".

In a lexical decision task participants were asked to decide whether a stimulus

item constitute an English word, a Russian word or a non-word. Word primes were

presented on a computer screen. At the beginning of every trial a fixation point appeared

on the screen. It was visible for 1 second and was immediately followed by stimulus

presentation. Participants had to press color-coded keys as they made a decision: red for

Russian words, green for English words and yellow for non-words. The yellow key was

positioned in the center of the keyboard. Left and right assignment of the red and green

key was counterbalanced across participants.

As in the original study, translations from L1 to L2 and from L2 to LI were

performed in separate blocks. The order of production language was counterbalanced

across participants. Translation stimuli were presented on a computer screen. At the

beginning of every trial a fixation point appeared on the screen. It was visible for 1

second and was immediately followed by stimulus presentation. Translation latencies

were measured for every stimulus by a timer which was set to go off the moment a








stimulus word appeared on the screen. Participants were instructed to press a green key at

an onset of translation production. In case they could not translate, they were asked to

press a yellow key, which enabled them to proceed to the next stimuli. There was a one

second break before the next trial. Both lexical decision and translation tasks were

preceded by a short practice session to ensure a consistent performance at test.

Participants were tested individually. The experimental session lasted from 20 to 30

minutes.

Results


The data obtained in the translation task were trimmed according to the same

principles as in the Sholl et al. (1995) study; that is, latencies with values greater than 2.5

standard deviations above a given participant's mean response time were excluded from

the analysis. Latencies from incorrectly translated items were not included in the set. In

addition, an item analysis of translation data was performed. Three stimulus items that

received less then 10 % of responses were removed from the analysis because of the lack

of data.

Correct responses provided to primed translation stimuli did not always denote the

same concepts that were activated by picture and word primes. Although all experimental

stimuli were concrete nouns of high imageability, many of them had multiple meanings.

Chances of providing unprimed translations were equally high in both directions. For

example, the English word nail may be translated in Russian as nogot' and gvozd' to

denote a horny growth at the end of fingers and toes, and a pointed piece of metal,

respectively. In a like manner, Russian word lestnica means both 'ladder' and 'staircase'.








The data obtained in this experiment was analyzed on the overall number of

correct translations and on the number of primed translations (i.e. translations of the

stimuli that produced primed lexical forms).

Overall Translation Data

Figure 4-1 presents mean translation latencies as a function of the direction of

translation (from L1 to L2 or from L2 to L1) and the priming condition (i.e. whether

translation stimuli were primed by pictures, L1 word, L2 word or were new/unprimed).

Translation latencies for the new items demonstrated greater asymmetry than the

asymmetry reported in the Sholl et al. (1995) study and its replication. On average,

translation times from L1 to L2 were 433 ms longer than translation times from L2 to L1

(1,933 ms and 1,500 ms respectively).






| 2.2
8 2 -. Picture BLI1 L2 ONew











2L1>L2 L2>L1
1.8
1.6





0.4
0.2


Ll>L2 L2>L I


Figure 4-1. Overall translation data: Mean translation latencies as a function of the
direction of translation and the priming condition.








A 2 x 4 ANOVA (two directions of translation: L >L2, L2>L1, and four priming

conditions: translation stimuli primed by pictures, LI and L2 words, and new/unprimed)

was performed on mean translation times using participants as random factors. As in the

original study, the analysis revealed a significant main effect of the direction of

translation, F(1,216) = 11.3,p<.01. However, the main effect of the priming condition

and interaction between the direction of translation and the priming condition were not

significant (p>.5).

Comparison of translation latencies in both directions showed that LI primes

produced differential effects. Unlike other translation stimuli, the concepts that were

previously primed by L1 lexical items did not demonstrate translation asymmetry. A one-

way ANOVA with priming condition as a within subject variable were performed on

translation latencies for each direction. The effect of the priming condition on translation

latencies was not significant in both directions (p>.5). Additional paired sample t-tests of

translation latencies showed that the effect of priming approached significance only in the

direction from L2 to L1. Participants were 317 ms slower when translating stimuli primed

by L1 words as compared to the stimuli primed by L2 words (t=2.06,p=.05). Effects of

other conditions were not significant (p>.05). Contrary to the results reported in the

original study, the L2-to-L direction was more sensitive to different priming conditions.

Accuracy data are given in Table 4-1 as a function of the direction of translation

and the priming condition. As in the original study, participants were more accurate when

translating from L2 to L1 than in the opposite direction. A 2 x 4 ANOVA performed on

the accuracy data using participants as random factors demonstrated a significant main

effect of the direction of translation, F (1,216) = 23.23, p<.01. In addition, the analysis








revealed a significant interaction between the direction of translation and the priming

condition, F(3,216) = 9.09,p<.01. The effect of the priming condition was not significant

(p>.5). A one-way ANOVA with the priming condition as a within-subject variable was

performed on accuracy data for each direction. The effect of priming was significant in

the direction from LI to L2, F(3,108) = 5.82, p<.01, and was marginally significant in the

opposite direction F(3, 108) = 3.93, p<.05. Additional paired sample t-tests revealed that

both lexical (L2 word) and conceptual (picture) primes produced equal effects in the

direction from LI to L2 (t
when translating stimuli primed by pictures than unprimed stimuli or stimuli primed by

L2 words (t-2.68,p=.01 and t=3.23,p=.01 respectively).


Table 4-1. Proportion of correct translations in different priming conditions.

Priming condition:
Picture L1 word L2 word Unprimed
Ll-to-L2 .81 .70 .80 .70
STD .10 .14 .12 .15
L2-to-L1 .78 .86 .83 .84
STD .9 .11 .09 .10



Primed Translations

Latencies of unprimed translations were removed from the set of the overall

translation data. Responses that did not match L2 picture names or L2 lexical forms

presented at study were removed from L1-to-L2 translation data, and, conversely,

responses that did not match LI picture names and L1 word primes were removed from

L2-to-L 1 translation data. Translation data of new items remained the same as in the








analysis of the overall translation data. On average 2.25 trials were omitted in L1-to-L2

translations, and 2.21 trials were omitted in L2-to-L1 translations in a data set obtained

from a single participant. A list of unprimed translations is provided in Appendix (Table

A-3).


Table 4-2. A comparison of mean translation latencies in seconds for the overall and the
primed translation data. Primed translation latencies are given in parentheses if different
from overall translation latencies.

Priming condition:
Picture L1 word L2 word Unprimed
Ll-to-L2 1.75 (1.74) 1.79 1.76 (1.78) 1.93
STD .59 (.58) .58 .62 (.69) .67
L2-to-L1 1.51 (1.50) 1.77(1.78) 1.46 1.50
STD .37 (.38) .67 (.69) .47 .44



Table 4-2 presents a comparison of translation latencies for the overall and the

primed translation data sets. As reported in the replication of Sholl et al. (1995),

comparison of the overall and primed translation latencies revealed only a negligible

difference. A 2 x 4 ANOVA performed on mean translation latencies using participants

as random factors showed the effects as reported in the analysis of the overall translation

data. There was a significant main effect of direction of translation, F(1,216) = 11.3,

p<.01. The main effect of priming and the interaction between the direction of translation

and the priming condition were not significant (p>.5). A one-way ANOVA with priming

condition as a within-subject variable did not show a significant effect of priming in both

directions (p>.5).











overall a primed





0
0.9





o 0.6
0o
0.5



L1 >L2 L2 >L1




Figure 4-2. A comparison of accuracy values for the proportion of all correct
translations and the proportion of primed translations



Figure 4-2 presents a comparison of the overall and primed accuracy data. Data

patterns observed in Figure 4-2 suggest that higher accuracy values reported in the

analysis of the overall translation data for the stimuli primed by pictures and L2 words in

Ll-to-L2 translation and L1 words in L2-to-L1 translation were not a result of priming. A

2 x 4 ANOVA was performed on the accuracy data with the direction of translation and

the priming condition as within-subject variables. As in the analysis of the overall

translation data, there was a main effect of the direction of translation, F(1,216) = 21.57,

p<.01. Interaction between the direction of translation and the transfer condition was also

significant, F(3,216) = 6.11, p<01. Separate one-way ANOVA with the transfer

condition and the direction of translation as within-subject variables were performed on

accuracy of the primed translation data. Contrary to the previous analysis, priming effects








were no longer significant in the direction from L1 to L2, p>.5, whereas in the opposite

direction priming effects became significant, F(3,108) = 6.45, p<.O1.

Discussion


According to the revised hierarchical model, translation from LI to L2 is a

conceptually-driven task and should benefit from prior conceptual activation of

translation stimuli. Therefore, translation latencies of the stimuli which were previously

presented in the object sorting task should demonstrate a significant reduction of

translation latencies as compared to the stimuli presented as LI and L2 words or the

stimuli which did not receive any prior activation. The experimental data reported here do

not support this contention. Neither the effect of priming nor the interaction between the

priming condition and the direction of translation were significant.

Latencies obtained in this experiment demonstrated a significant translation

asymmetry. Translation times for the new items in Ll-to-L2 translation were 433 ms

longer than in the opposite direction. This asymmetry was even greater than the

asymmetry reported in the original study, which was 149 ms. Therefore, it would be

reasonable to conclude that bilingual speakers employed in this experiment were

dominant in their first language, as was a group of bilinguals in the original study.

The results obtained in the original study demonstrated that English-Spanish

bilinguals were concept mediators only in the direction from L1 to L2. On the basis of

extensive experimental evidence Dufour and Kroll (1995) proposed that more fluent

bilinguals may gradually switch to concept mediation in both directions. According to

their proposal, as fluency increases, bilinguals switch to conceptual mediation in both








directions. In such a case translation asymmetry should be significantly reduced and

transfer between picture naming and translation should be effective in both directions.

According to Sholl et al. (1995), transfer effects are an indication of conceptual level

processing during translation. In the Sholl et al. (1995) study transfer between picture

naming and translation effected a reversal of the standard translation asymmetry. Unlike

the translation latencies of new words that were longer in Ll-to-L2 translation than in the

opposite direction, L1-to-L2 translation latencies of the concepts that were previously

named in L2 were shorter than their L2-to-L 1 translation latencies. Such a reversal effect

may be significantly reduced in the case of bi-directional conceptual activation during

translation resulting from insufficient translation asymmetry. As a result, effects of

experimental variables become less transparent in the data of more fluent bilinguals. The

latency data which exhibits greater translation asymmetry would be more transparent for

transfer effects than the data that does not exhibit such asymmetry. Notwithstanding, this

study failed to obtain transfer effects in the latency data.

The lack of a transfer effect in the Ll-to-L2 direction in this data suggests that the

locus of priming may not be at the conceptual level, as proposed in the original study.

Moreover, L2-to-L translation latencies demonstrated sensitivity to different types of

lexical stimuli in this direction: there was a significant difference between translation

latencies produced in responses to the stimuli which were previously presented as LI and

L2 words. Responses to LI-primed words were significantly slower than response to L2-

primed words (t=2.06, p=.05). Although sensitivity to differences in lexical stimuli in the

direction from L2 to LI agrees with the assumption that that translation in this direction

is performed at the lexical level of processing, it contradicts the conjecture that the locus








of transfer is not at the level of retrieving the lexical form. The leading assumption of the

revised hierarchical model (i.e. that translation from L2 to L1 is essentially a lexical-level

task) also became problematic for the principle of transfer-appropriate processing.

Following the logic of transfer-appropriate processing, prior lexical activation should

facilitate performance of the task that requires lexical processing. As shown in the

analysis of L -to-L2 latency data, such facilitation was not obtained. Prior processing of

L2 lexical stimuli resulted in a 39 ms facilitation of response times, which was not

significant. Contrary to the transfer-appropriate logic, prior exposure to L1 stimuli

produced a negative effect: the time to translate the stimuli previously seen as L1 words

was 270 ms longer than the time to translate new words. This negative effect remained

when latencies ofunprimed translations were removed from the data set. Since all

translation stimuli were counterbalanced across the priming conditions, the effect was not

due to stimuli selection. Although the reliability of this effect is rather questionable, the

mere likelihood of this episode questions validity of the assumption that was tested in this

experiment.

Additional evidence regarding the locus of transfer comes from analyses of the

accuracy data. Accuracy was analyzed on the basis of all correct responses (overall

accuracy) and on the basis of responses which produced exactly the same lexical forms as

were supposed to be activated by primes (primed accuracy). The effects reported in the

analysis of the overall accuracy conform to predictions that follow from the revised

hierarchical model. That is, in the direction from LI to L2, bilinguals were more accurate

in their translations of the stimuli that were previously presented as pictures and L2

words, whereas in the opposite direction neither of the priming conditions were








significantly different. Higher response accuracy to the stimuli primed by pictures and L2

words in L1-to-L2 translation could be interpreted as evidence of transfer that occurs in

the direction from Ll to L2. However, an itemized analysis of the accuracy data revealed

that not all correct translations could result from prior presentation of a prime. For

example, in the Russian language, the word lestnica is used to denote two concepts that

have distinct lexical representations in English ('ladder' and 'staircase'). In translation

from LI to L2 the stimulus lestnica was unprimed in one subset of stimuli and was

primed by a picture of a ladder, by the Russian word lestnica and by the English word

ladder in other subsets. In L1-to-L2 translation both English words were produced,

regardless of the priming condition. In the subsets where the concept of ladder was

activated either by a picture or by the English word, "stairs" or "staircase" were

considered unprimed responses.

Unprimed responses were removed from the analysis of the primed translations.

The analysis revealed that actual effects of priming were the reverse of what was

expected: priming effects were significant only in the direction from L2 to LI, and as

with the latency data, the effect of transfer was negative in this direction. The combined

evidence from the overall and primed accuracy analyses indicate that higher accuracy

ratio of responses to primed translation stimuli did not result from transfer, since the

proportion of target lexical items produced in response to primed translation stimuli were

lower than the proportion of correct translations provided in response to unprimed

translation stimuli (see Figure 4-2).

The results suggest that both lexical and conceptual priming may not be

considered a reliable factor affecting translation accuracy. Translation accuracy is a








complex notion that is affected by a bilingual's familiarity with patterns of lexical-

conceptual organization in the target language. A set of lexical items activated by a

particular concept may be different in a bilingual's two languages. A bilingual may not

have a sufficient knowledge of L2 lexical-conceptual organization in order to account for

the differences. Otherwise, production demands may not require them to discriminate

between sets of senses which can be activated by a stimulus lexical item in both

languages. Since the translation of isolated stimuli does not require such discrimination,

this type of a task may not provide an adequate account of the processes engaged in the

activation and retrieval of lexical information in both languages. The use of contextual

information along with single word stimuli can help to channel activation in a particular

direction, in which case experimental results would not be uncontrollably confounded by

other factors. For example, it was shown that use of meaningful contexts at study leads to

semantic involvement at test even on such traditional data-driven task as word-fragment

completion (Smith, 1991). Thus, memory performance has been demonstrated to be not

only the function of the overlap between encoding and retrieval conditions, but also the

function of the interaction between task demands and stimulus processing requirements.

Since in the Sholl et al. (1995) study the major emphasis was put on the former, the

results may not be a comprehensive reflection of bilingual memory performance.













CHAPTER 5
EXPERIMENT 2: REPLICATION OF
THE SHOLL ET AL. (1995) STUDY

In the Sholl et al. (1995) experiment picture naming was used at the encoding

condition to provide conceptual activation, and translation was used at the test. The

results were predicted to demonstrate a transfer between the tasks that share similar

processing requirements. Following the representational organization proposed in the

revised hierarchical model, picture naming was expected to affect translation times of

previously named concepts only in the direction from LI to L2, because only this

direction was proposed to require conceptual activation. The results reported by Sholl et

al. (1995) confirmed the prediction. Significant reduction in translation times from L1 to

L2 and the absence of thereof in the opposite direction was found to be in agreement with

the principle of transfer-appropriate processing. Shorter latencies observed in Ll-to-L2

translation of the previously named concepts were interpreted as an indication of transfer

which is possible between tasks that require conceptual processing. In addition, the lack

of transfer in L2-to-L translation following name production in picture naming task was

interpreted as evidence suggesting that the locus of transfer is not at the level of lexical

form retrieval. The combined evidence was taken as support for the configuration

proposed in the revised hierarchical model.

The purpose of this experiment was to replicate the effects reported by Sholl et al.

(1995). It is not clear from the analysis reported by Sholl et al. (1995) whether








translations of previously named concepts were matched up to the responses that were

produced in the naming task. The analysis of the responses obtained in the Experiment 1

demonstrated that some correct responses produced unprimed translations. As mentioned

in Chapter 3, the present model maintains that transfer is not the only factor affecting

bilingual performance at test. Therefore an itemized analysis of the translation data is

needed to assess the extent of transfer effect on the primed responses.

Method


Participants

Fifteen Russian-English bilinguals participated in the experiment. All participants

were native speakers of Russian who had lived in an English-speaking environment for at

least 2.5 years (5.2 on average) and had studied English as a foreign language for at least

2 years (3.8 on average) before coming to the US. At the beginning of each experimental

session, participants were requested to fill out a questionnaire about their language

learning experiences and were also asked to evaluate their L2 fluency level. Participants

did not consider themselves to be balanced bilinguals. They rated themselves at an

advanced level for most aspects of L2 production and at a near-native level in some areas

of L2 production and most areas of L2 comprehension. Participants were not given any

compensation for participation.

Materials

The materials were selected according to the same principles described in Sholl et

al. (1995). There were 80 pictures selected from the Snodgrass and Vanderwart (1980)

norms. Picture names formed a variety of superordinate categories in both English and








Russian. Pictures and their word names were divided into four versions and were partially

counterbalanced across conditions and across participants. Half of the materials were

presented at study; that is, 20 pictures were used in the LI naming task and 20 other

pictures were used in the L2 naming task. An additional 20 pictures were added to the set

as fillers to be used in picture naming: half of them were used in the L1 naming block and

half of them were used in the L2 naming block. The remaining 40 picture names were

presented as new stimulus items in the translation task: half of them were presented in

translation from L1 to L2 and half of them were used in translation from L2 to LI. Except

for deliberate repetition of the target picture names in the translation task, no concepts

were repeated for a given participant. There were two blocks of translation stimuli. Each

block consisted of 10 picture names that were named in LI, 10 picture names that were

named in L2, and 20 words that were presented as new items. Stimulus materials and the

experimental conditions in which they were used are provided in Appendix (Tables A-4

and A-5).

Apparatus

The experiment was conducted on an IBM compatible notebook computer with a

24 cm color active-matrix display. The experiment ran a computer program written in

Quickbasic. The instructions and the stimuli were presented in white 0.5x0.5 cm letters

on black background. All responses were oral. Responses were audiotaped so they could

be later checked for accuracy.








Procedure

Participants were presented with two blocks of picture stimuli and were asked to

name each block of pictures in their first (Russian) and second (English) language. The

order of LI and L2 naming was counterbalanced across participants. Pictures were

presented on flash cards. Participants were requested to name them as quickly as they

could. Naming latencies were taken for the whole set of pictures. The naming task was

immediately followed by translations from L1 to L2 and from L2 to L1, which were

performed in separate blocks. The order of language production was counterbalanced

across participants. Translation stimuli were presented on a computer screen. At the

beginning of every trial a fixation point appeared on the screen. It was visible for 1

second and was immediately followed by a stimulus presentation. Translation latencies

were measured for every stimulus by a timer which was set to go off the moment a

stimulus word appeared on the screen. Participants were instructed to press a green key at

an onset of translation production. In case they could not translate, they were asked to

press a yellow key, which enabled them to proceed to the next stimulus. There was a one-

second break before the next trial. Each translation block was preceded by a short practice

session to ensure a consistent performance at test. Participants were tested individually.

The experimental session lasted 10 minutes.

Results


Picture Naming

Mean naming latencies for pictures named in LI and L2 were calculated. A one-

way ANOVA was performed on naming latencies using participants as random factors.








Picture naming language was a within-subjects variable. Bilinguals were more fluent

when naming pictures in LI (1,444 ms) than in L2 (1,849 ms), the observed difference

was significant, F(1,28) = 7.11,p<.05. A one-way ANOVA performed on accuracy data

showed that bilinguals were significantly more accurate when naming pictures in L1

(99.33% correct) than in L2 (90.22% correct), F(1,28) = 18.82, p<.01. Analysis of the

naming errors revealed that most errors in L1 and L2 were essentially tip-of-the-tongue

phenomena. Participants commented that they were familiar with both L1 and L2 names,

but they could not access them at the time. The analysis of the naming data showed that

although participants were slightly dominant in L1, they were equally fluent in both

languages.

Overall Translation

The data obtained in the translation task was trimmed according to the same

principle as in the Sholl et al. (1995) study; that is, latencies with values greater than 2.5

standard deviations above a given participant's mean response time were removed from

the set. In addition, latencies from incorrectly translated items were also removed.

Figure 5-1 presents mean translation latencies as a function of the direction of

translation (from L1 to L2 or from L2 to LI) and the transfer condition (i.e. whether

translation stimuli were previously named as pictures in L1, L2 or whether the stimuli

were new). As in Sholl et al. (1995) translation latencies for the new items demonstrated

some asymmetry. On average, translation times from L to L2 were 86 ms longer than

translation times from L2 to L1 (1,541 ms and 1,455 ms, respectively).








A 2 (Direction of Translation: from LI to L2 and from L2 to L1) x 3 (Transfer

Condition: concepts were previously named in L1, in L2 and presented as new) ANOVA

was performed on mean translation latencies using participants as random factors.






o

1 2
1.8 Ml.I EL2 DNet,



1- --

o 0.8
0.4

0.2
0
L1>L2 L2>L1


Figure 5-1. Overall translation data: Mean translation latencies as a function of
direction of translation and transfer condition. L1 = first language; L2 = second
language.



Direction of translation and transfer condition were considered within-subject variables.

The analysis did not show significant main effects of the direction of translation or the

transfer condition (p>.5). Neither did it reveal a significant interaction between the

variables (F
A comparison of translation times in both directions showed that only the words

that were previously named in L2 were characterized by shorter translation latencies

when compared to new words. A one-way ANOVA with transfer condition as a within-

subject variable was performed on translation latencies for each direction. The effect of








the transfer condition on translation latencies was not significant in either direction (F
Additional paired sample t-tests of translation latencies did not reveal any significant

effects (p>.1). The latency data obtained in this experiment did not replicate effects

reported by Sholl et al. (1995).


Table 5-1. Proportion of correct responses for presented (primed) and new unprimedd)
stimuli. Proportion of correct primed translations are given in parenthesis if different
from the overall accuracy values (i.e. all correct translations are considered).

Priming condition:
L1 naming L2 naming Unprimed
Ll-to-L2 translation .79 .87 (.75) .79
STD .14 .11 (.13) .12
L2-to-L1 translation .92 (.85) .79 .85
STD .10(.14) .16 .08



Accuracy data are given in Table 5-1 as a function of the direction of translation

and the transfer condition. As in the Sholl et al. (1995) study, participants were more

accurate when translating from L2 to LI than in the opposite direction. However,

contrary to the effects reported by Sholl et al. (1995), an ANOVA performed on accuracy

data demonstrated a significant interaction between the direction of translation and the

transfer condition, F(2,78) = 5.43,p<.01. The effects of the direction of translation and

the transfer condition were not significant (p>.5). A one-way ANOVA with the transfer

condition as a within-subject variable was performed on accuracy data for each direction.

The effect of the transfer condition was not significant in either direction (F
Additional paired sample t-tests revealed that the effect of prior naming only approached

significance in the L1-to-L2 direction. Participants were more accurate when translating








the concepts that were previously named in LI than in L2 (t-2.69,p=.01). The difference

between the new condition and each of the old conditions (words named in LI and L2)

was not significant (p>.05).

Primed Translation

The latencies of unprimed translations were removed from the set of the overall

translation data. Responses that did not match L2 picture names produced in the study

were removed from L -to-L2 translation data, and, conversely, responses that did not

match LI picture names were removed from L2-to-L1 translation data. Translation data

of new items remained the same as in the analysis of the overall translation data. On

average 1.46 trials were omitted in Ll-to-L2 translations, and 1 trial was omitted in L2-

to-L1 translations in a data set obtained from a single participant. A list of unprimed

translations is provided in Appendix (Table A-6). Table 5-2 presents a comparison of

translation latencies for the overall and the primed translation data sets.


Table 5-2. Mean translation latencies in seconds as a function of the direction of
translation and the transfer condition (L 1 naming, L2 naming, unprimed). Primed
translation data are given in parenthesis if different from the overall translation data.

Priming condition:
L 1 naming L2 naming Unprimed
Ll-to-L2 translation 1.55 1.33 (1.34) 1.54
STD .34 .33 (.35) .35
L2-to-LI translation 1.44 (1.42) 1.41 1.45
STD .35 (.36) .30 .37



As in the analysis of the overall translation data a 2 x 3 ANOVA was performed

on mean translation latencies using participants as random factors. Direction of








translation and transfer condition were considered as within-subject variables. As in the

analysis of the overall translation data, there were no main effects of experimental

variables. Neither direction of translation nor transfer condition were significant (F <1).

There was no significant interaction between the variables (F<1). An additional one-way

ANOVA with transfer condition as a within-subject variable did not show any significant

effects (F<1).

The primed accuracy data are given in Table 5-1 in parenthesis. A 2 x 3 ANOVA

was performed on the accuracy data with the direction of translation and the transfer

condition as within-subject variables. As in the analysis of the overall translation data, the

effects of the direction of translation and the transfer condition were not significant

(p>.5). However, contrary to the previous analysis, interaction between the direction of

translation and the transfer condition was no longer significant: p>.5. Separate one-way

ANOVA with the transfer condition and the direction of translation as within-subject

variables were performed on accuracy of the primed translation data. Neither of the

analyses revealed significant effects (F<1). Additional paired sample t-tests did not show

any significant effects (p>.5).

Discussion


The results of the experiment did not replicate effects reported by Sholl et al.

(1995). Since experimental designs were identical in the original study and in its replica,

the effects could have been influenced by differences between the two groups of

participants. Analyses of naming data demonstrated a clear difference in L2 proficiency

levels between bilinguals employed in both studies. L and L2 naming latencies obtained








from English-Spanish bilinguals in the original study demonstrated a significant

difference. Absence of such a difference in the naming data of Russian-English bilinguals

suggests that Russian-English bilinguals were more fluent in their second language than

the English-Spanish group. The smaller difference between L and L2 naming fluency

could be the reason why translation latencies of the Russian-English group did not

demonstrate significant effects of experimental variables. As was mentioned in the

discussion of the previous replication, the latency data which does not exhibit significant

translation asymmetry would be less transparent for transfer effects. In this case latency

data may not be considered a reliable measure of transfer that was predicted to occur

between tasks sharing similar processing requirements.

An alternative means of establishing whether there was a transfer between picture

naming and translation is to consider accuracy data. A comparison of accuracy data

confirmed the assumption that Russian-English bilinguals were more proficient in their

second language than bilinguals from the original study. Naming responses produced by

Russian-English bilinguals were correct at 90.2%, whereas only 65.8% of names

produces by English-Spanish bilinguals were correct. Nonetheless the accuracy data

obtained from the Russian-English group demonstrated a significant effect of the

direction of translation. In other words, Russian-English bilinguals were significantly

more accurate when naming pictures in their first language than doing the same task in

the second language. Although participants commented that in most cases their omissions

could be viewed as tip-of-the-tongue phenomena, the failure to access L2 names could

also be interpreted as an evidence of bilinguals' dominance in their first language.

Consequently, the accuracy data can be considered a reliable measure of transfer between








picture naming and translation. The model, however, does not make explicit predictions

about accuracy. The model was initially proposed to accommodate latency data, and it is

not clear whether the same predictions could be extended to a different domain, i.e.

accuracy data.

In the original experiment latency and accuracy data were affected differently by

experimental variables. An analysis of latency data showed that only the interaction

between the variables was significant, whereas the accuracy data analysis reported only

main effects of the experimental variables. The analysis of this experiment's naming data

demonstrates that accuracy was more sensitive to differences in L1 and L2 proficiency

than latency was. Therefore, the logic that is used with latency data may not apply to

accuracy.

Analyses of accuracy data were conducted separately for the overall number of

correct translations and for accuracy of primed translations. The overall accuracy data

was characterized by a significant interaction between the direction of the translation and

the transfer condition. The proportion of correct translations was higher for the words that

were previously named in the target language. Such an interaction could be considered an

effect of transfer that occurs in both directions. This interpretation conforms to the

configuration outlined in the revised hierarchical model. However, an itemized analysis

of accuracy data revealed that not all correct translations of previously named concepts

were the same as responses produced in picture naming tasks. Therefore, effects reported

in the analysis of accuracy data conducted for the overall number of correct translations

may not necessarily be a reflection of transfer due to prior picture naming. A separate

analysis of primed translations did not reveal any effect of experimental variables on








accuracy data. The reported cases of correct unprimed translation poses a serious problem

for the revised hierarchical model. The model does not have a means to account for them

mainly because it does not have a component which can handle the finer-grain issues of

representation such as polysemy and homophony.

According to the model, translation responses should be primed by prior

conceptual activation only when bilinguals employ conceptual level while translating.

Therefore accuracy of less fluent/unbalanced bilinguals should be affected in L -to-L2

translation, whereas translation accuracy of more fluent speakers should demonstrate

transfer effects in both directions, or as suggested in Chapter 4, may not show transfer

effects in either direction due to the lack of translation asymmetry. The latter was not

supported by the data reported in Sholl et al. (1995). The analysis of accuracy data of

unbalanced bilinguals demonstrated main effects of the transfer condition in both

directions, although it was maintained that bilingual participants were concept mediators

only in the direction from L1 to L2. (The latter was drawn on the basis of latency data).

However, an unequivocal interpretation of such results is possible only if it is clearly

specified whether all correct translations were the same as naming responses produced at

the encoding condition of the experiment. It is equally difficult to draw definite

predictions about accuracy on the basis of the configuration outlined in the revised

hierarchical model. For instance, in L -to-L2 translation, concept activation by naming

picture in LI does not ensure the accurate connection between a concept node and an L2

word. If a concept is named in L2, the connection between a concept and its L2

representation is activated. However, the configuration of connections proposed in the

revised hierarchical model does not imply that a connection between a concept and an LI








word will also be established during L2 naming, in which case, accurate translation is

possible only when L1 stimulus will activate exactly the same representation as the one

activated in L2 naming. For example, activation of a connection between the concept

'stool' and its LI lexical representation taburet did not always ensure an accurate L1-to-

L2 translation (i.e. stool). Responses that were considered as inaccurate were omissions

and productions of the basic level term chair. On the other hand, an activation of the

concept 'moon' in L2 naming did not always help to establish a connection between the

concept and its LI representation mes 'ac. As a result, a different concept ('month') was

activated upon presentation of the Ll stimulus mec ac, and a response provided in L1-to-

L2 translation was different from the name produced in L2 naming.

Analyses of the overall and the primed accuracy data revealed that the data cannot

be used to support the revised hierarchical model. The effect of transfer reported in the

analysis of the overall number of correct responses and the absence of thereof in the

analysis of primed data suggests that there are some factors of bilingual performance that

were not identified on the basis of the configuration of lexical-conceptual connections

promoted by the revised hierarchical model. Effects reported in the accuracy data

analyses remain unaccountable mainly because the model does not possess a mechanism

that can (1) explain why translation stimuli may not activate the same concepts as the

ones activated during picture naming, and (2) predict cases in which prior picture naming

fails to prime target translations. One solution to this problem is to define a framework

that will recognize issues of language-specific lexical-conceptual organization as central

to the issue of lexical access.













CHAPTER 6
RELATION ASSESSMENT: A TEST OF THE PROPOSED MODEL OF
CONCEPTUAL STRUCTURE MAPPING

Earlier proposed models of bilingual memory were designed to accommodate

experimental evidence arising from specific task demands on single-word stimuli.

Latency data obtained from data-driven and conceptually-driven task was taken as

evidence of a dissociation between lexical and conceptual processing of word stimuli.

However, as demonstrated in the reported replications of Sholl et al. (1995), latency alone

can not provide unambiguous evidence of bilingual processing patterns. The present

model of conceptual structure mapping was proposed to account for the shortcomings of

earlier models. It addresses the stimulus processing requirements associated with

language-specific differences in the relational network organization of lexical stimuli. It

also addresses instability of prototype effects resulting from graded structure of lexical

categories.

The model and its predictions were tested in a relation assessment experiment. In

this experiment word pairs to be judged for their semantic relatedness were presented in

contextually-embedded (primed) and isolated unprimedd) conditions. The relation

assessment task was selected because it can demonstrate prototype effects and degrees of

activation spreading arising in the relational organization of the target word stimuli.

Prototype effects are generally associated with two things: the senses of a stimulus that

exhibit higher perceptual salience or the senses that are most often associated with the








given lexical forms. Hence, a word pair with closely related prototypical senses should

yield shorter latencies. Among the causes of this is the immediate availability of the

target senses upon the activation of the stimulus's relational organization. Conversely, a

stimulus pair with related nonprototypical senses is more likely to generate a greater

number of negative judgments or longer latencies in the isolated condition. Contextual

embedding affects the graded structure of stimulus lexical categories by restricting

prototype effects to domains associated with the activated contextual frames. The

activation of a relevant contextual frame eliminates the need for an extensive search in

the relational network of a stimulus lexical item. Consequently, it also provides quicker

access to the target conceptual representation. Thus shorter response latencies were

expected in the primed condition of the experiment.

According to the proposed model the data should demonstrate specific patterns

associated with the additional processing demands placed on bilingual speakers. The

model maintains that differences in response patterns of bilinguals are conditioned by the

activation threshold in the dominant conceptual structure. Therefore a bilingual's

performance should provide evidence for the relational organization in the bilingual's

dominant conceptual structure. Two predictions were made. The evidence should be

particularly salient in the isolated condition. However, since prototype effects associated

with a particular organization in the dominant conceptual structure are presumably

affected by contextually embedding the lexical stimuli, the data obtained from bilingual

and monolingual speakers in the primed condition should reveal comparable patterns of

activation.








Experiment 3


Method


Participants

Four groups of monolingual and bilingual speakers participated in the experiment.

The participants were students and visiting scholars at the University of Florida. The

fluency of bilingual speakers was at the level that fulfills the English language

requirement at the University of Florida, which corresponds to a score of 550 or higher on

TOEFL (Test of English as a Foreign Language). The monolingual participants, who

were taken from a pool of general psychology students, received one experimental credit

for their participation in the study; the bilingual participants did not receive any

compensation. A group of nineteen native speakers of American English not fluent in any

second language and a group of fifteen native Russian speakers fluent in English were

tested in the unprimed condition. A group of twelve American English monolingual

speakers and a group of thirteen native Russian speakers fluent in English were tested in

the primed condition.

Materials

The corpus of stimuli in the unprimed condition included 51 target and 49 filler

English word pairs. The target word pairs comprised three groups. These groups were

categorized as closely related, distantly related, and unrelated, with 17 pairs in each

group. Closely related word pairs were selected in such a way that the prototypical

meanings of the two lexical items were closely related (e.g. 'fortune' 'wealth'). In








distantly related pairs the prototypical meaning of one lexical item was closely related to

a non-prototypical meaning of the other lexical item in the pair (e.g. 'limit' 'ceiling'). In

unrelated word pairs none of the senses in the networks of the stimulus lexical items were

considered to be related (e.g. 'juice' 'sock'). The relatedness categorization was based

on relatedness ratings of three native speakers of English and were found to be in

agreement with primary meanings of the standard variety of American English, as

reflected in the New Webster's Dictionary.

The corpus of the stimuli in the primed condition consisted of 51 target and 14

filler English word pairs, which were preceded by 65 sentences. The target word pairs

were the same as in the unprimed condition. The sentences used as primes were intended

to activate contextual frames, highlighting target conceptual representations. For

example, the word pair 'limit' -'ceiling' was preceded by the sentence "The

administration has introduced new ceilings on the value of preferential contracts for

minorities." The sentences were selected from authentic English texts available through

LEXIS/NEXIS online news service. Stimulus materials used in this and other relation

assessment experiments are provided in Appendix (Table A-7).

AnMaratus

The experiment was conducted on an IBM compatible notebook computer with a

24 cm color active matrix display. The experiment ran a computer program written in

Quickbasic. The instructions and the stimuli were presented in white 0.5 x 0.5 cm letters

on blue background.








Procedure

The experiment conformed to a 3 x 2 x 2 mixed factorial design. The independent

variable manipulated within participants had three values corresponding to the distance

between the prototypical meanings of the words presented in a pair. The word pairs were

grouped into three sets: closely related, distantly related and unrelated. Binary language

(monolingual vs bilingual) and priming unprimedd vs primed) conditions were distributed

between four groups of participants. Reaction times in milliseconds and percent of

positive judgments in the relatedness decision were assessed as dependent variables.

In the unprimed condition, the words constituting a pair were presented

consecutively. Stimulus and interstimulus intervals were set automatically at 1 second.

The timer was set to go off the moment the second word appeared on the screen.

Participants were instructed to make judgments of the semantic relatedness between the

words. They had to respond as soon as they read and understood the second word in a

pair. Response keys marked by "yes" and "no" caps were also color coded, green and red

respectively. The "yes" and "no" response keys were located at opposite ends of the

computer keyboard to ensure accuracy of responses. Left and right assignment of "yes"

and "no" keys was counterbalanced across participants. Participants were instructed to

press a white key in the center of the keyboard if they did not know the wordss. Each

trial was followed by an automatically set break that allowed the participant to self-pace

the experiment. In the primed condition every trial was preceded by a sentence or two.

The sentences comprised less than four lines on a computer screen. In order to ensure that

participants attended to the sentence stimuli, each sentence was followed by a








comprehension question with four possible answers. Responses were made by pressing

the number of a chosen answer. The computer recorded only the correctness of the

response to the comprehension question. Both experimental sessions were preceded by a

training session which allowed participants to familiarize themselves with the task.

Participants were tested individually. The experiment lasted from 10 to 20 minutes in the

unprimed condition and from 40 to 50 minutes in the primed condition.

Results and Discussion


The data from participants who had mean reaction times higher than 3 seconds for

monolingual speakers or 5 seconds for bilingual speakers and a standard deviation higher

than 2 were removed from the set as outlying. To ensure comparable levels of second

language proficiency, the data from bilingual speakers who provided "don't know"

responses to more than 10 % of the stimulus pairs and who scored lower than 80 % on the

comprehension test in the primed condition, were also removed from the set. The

remaining data from 17 monolingual and 12 bilingual speakers in the unprimed condition

and 8 monolingual and 8 bilingual speakers in the primed condition were analyzed.

Analyses of variance (ANOVA) were performed on mean response times and percentage

of positive judgments using participants as random factors.

Relatedness judgments

Figure 6-1 presents the percentage of positive judgments as a function of network

distance and priming for both bilingual and monolingual speakers. The results

demonstrate a highly significant main effect of network distance, F(2,82)=65.377,

p<.001. Although the effect of priming was only marginally significant (F(1,41)=6.519,








p<.05), the interaction between network distance and priming was significant,

F(2,82)=5.170, p<.008. Priming made some distant meanings more salient and, as a

result, the words in the category of distantly related word pairs were judged as related

approximately 20% more often in the primed condition than in the unprimed condition by

both monolingual and bilingual groups of speakers.


unrelated


distant


Figure 6-1. Mean percentage of positive judgments as a function of network distance
for bilingual and monolingual speakers in unprimed and primed conditions.



There was no main effect of the language variable (F<1), indicating that the

response types provided by monolingual and bilingual speakers were equivalent. Separate

ANOVA were performed on bilingual and monolingual data to discover possible

differential effects of distance and priming on different groups of speakers. The analysis

of the bilingual data revealed main effects of distance and priming, F(2,54)=86.587,

p<.01, and F(1,54)=7.668, p<.001 respectively, and a significant interaction between


---e-primed monolingual
- --primed bilingual
- O -unprimed monolingual
- unprimed bilingual








distance and priming (F(2,54)=5.594, p<.01). The analysis of the monolingual data

demonstrate only main effect of distance (F(2,69)=24.706, p<.01). Although the main

effect of language was not significant, the separate analyses demonstrated that relatedness

judgments provided by bilinguals were greatly affected by the prior presentation of target

lexical items in meaningful contexts in the primed condition. For example, the percent of

positive judgments provided by monolingual speakers for the prototypical and

nonprototypical meaning in the word pair 'ceiling'-'limit' increased from 75 to 83

following presentation of the prime sentence "The administration has introduced new

ceilings on the value of preferential contracts for minorities," while the percent of

positive judgments provided by bilingual speakers increased from 27 to 67 percent.

The activation of a particular meaning relation within a precisely defined

contextual frame diminishes the negative lexical transfer that results from straightforward

mapping of incompatible representations in the conceptual structures of a bilingual.

Relatedness judgments obtained in the experiment were shown to be greatly influenced

by the availability of contextual cues. Paired-sample t-tests conducted on the data

produced in response to distantly related word pairs demonstrated a significant difference

in responses produced by bilingual and monolingual speakers in the unprimed condition,

t=2.02, p=.05, and a remarkable agreement between the two groups in the primed

condition, p>. .

The processing differences demonstrated between the group of English

monolingual speakers and the group of Russian-English bilingual speakers may have two

interpretations. They may result either from mapping the two conceptual structures or

from the general processing limitations of bilinguals. The proposed model suggests that








conceptual structure mapping should demonstrate processing patterns which reflect

language-specific relational organization. The model maintains that the effects of lexical-

conceptual organization in the dominant language of bilinguals are most conspicuous

when the target lexical items are presented in isolation. As predicted by the model,

bilinguals demonstrated results different from the result produced by monolinguals when

the target lexical items were not accompanied by any information about their relational

organization within the target conceptual structure.

Response latencies

Figure 6-2 presents the mean reaction times as a function of network distance for

bilingual and monolingual speakers in unprimed and primed conditions. As predicted, the





2.2
---primed monolingual
0 2 primed bilingual
0 -unprimed- monolingual
1.8 -- 0- -unprimed- bilingual
*. ............. .. ..
S 1.6
S 1.4

1.2


unrelated distant close




Figure 6-2. Mean response latencies (in seconds) as a function of network distance
for bilingual and monolingual speakers in unprimed and primed conditions.








F(1,41)=7.282, p<.001. Response time latencies of bilingual speakers were

approximately 300 ms longer than response time latencies of monolingual speakers in

both conditions. The main effect of the language variable supports the assumption that

longer latencies produced by bilingual speakers are conditioned by the processing

demands placed on bilingual speakers. Processing L2 lexical material requires operation

in the conceptual structure characterized by a higher activation threshold. It also requires

the suppression of the dominant conceptual structure, which may access a relational

network incompatible with the target conceptual organization. As a result, bilinguals

require additional time to process L2 stimuli.

The response times also demonstrated a significant main effect of network

distance, F(2,82)=5.635, p<.005.On the average, the participants required 90 ms longer to

respond to distantly related word pairs than to closely related word pairs, and they

required 80 ms longer to respond to unrelated word pairs than to distantly related word

pairs. The results support the prediction that the prototypical meanings of the lexical

items are accessed faster than nonprototypical meanings. The paired sample t-test

demonstrated statistical significance in response times between closely and distantly

related word pairs (t=2.872, p=.06) and between closely related and unrelated word pairs

(t=3.322, p=.02). The latency difference between distantly related and unrelated word

pairs was not statistically significant. This fact indicates that speakers abandon the search

when the accessed nonprototypical meanings in the networks of the two lexical items fail

to match. Priming reduced the response latencies of bilinguals and monolinguals by

approximately 150 and 180 ms, respectively. However, the effect of priming was not

statistically significant, p>.1. A ANOVA performed on the response latencies of








monolingual speakers demonstrated a main effect of priming, F(1,69)=7.355,p<.01. As

predicted, the latencies of monolingual speakers reflected prototype effects in the

unprimed condition by providing significantly longer latencies in response to distantly

related word pairs. The prototype effects were significantly reduced following

presentation of a priming sentence which activated the contextual frame of a target

nonprototypical sense in a distantly related word pair. As a result, nonprototypical

meanings were accessed faster in the primed than in the unprimed condition. Analysis of

response times produced by bilingual speakers did not show any significant effects. The

effect of priming on the response latencies of bilingual speakers was not significant. It

could be obscured by additional activation within L1 conceptual structure.

The analyses of response times did not reveal significant interactions between the

variables and conditions of the experiment: two-way interactions between language and

priming, network distance and language, and network distance and priming were not

significant (F
priming only approached significance at F(2,82)=2.193,p<.118. The absence of

significant interactions may be due to the ceiling effect resulting from generally large

response latencies produced by both monolingual and bilingual speakers. One possible

way to eliminate the ceiling effect is to reduce the time of stimuli presentation and the

duration of the interstimulus interval. In this experiment the time elapsed from the onset

of the first word to the onset of the second word in a pair was 2 seconds. This is a

significant amount of time to conduct extensive search in the semantic relational network

of the first stimulus word. Shorter presentation times will require participants to attend to

the next stimulus before they are able to consider a number of possible meanings








associated with the given word form. In this case response latencies produced by

bilingual and monolingual speakers in primed and unprimed conditions may manifest

significantly greater differences than are reported with the present design.

Experiment 4


This experiment was conducted to test the hypothesis that the results obtained

from bilingual speakers in the unprimed condition are influenced by the dominant

conceptual structure. The judgments provided for closely related word pairs were

expected to exhibit similar patterns across different language groups for two reasons.

First, there is a conspicuous relation between the prototypical meanings of the words in

the pairs. There is also the similarity in relational organization pertaining to perceptual

salience. Similarly, unrelated word pairs were predicted to demonstrate small differences

because of the obvious incompatibility in meaning. The differences in judgments

provided for distantly related word pairs were expected to exhibit the patterns conditioned

by the relational organization of the conceptual structures of bilinguals' dominant

languages. The semantic relations between prototypical and nonprototypical senses in

distantly related pairs of the target language may not be present in the relational

organization of the equivalent lexical items in bilinguals' dominant languages.

Consequently, bilinguals with different dominant conceptual structures were expected to

demonstrate distinct relatedness judgments.








Method


Participants

Thirteen native Spanish speakers fluent in English and twelve English

monolingual speakers were taken from the pool of general psychology students at the

University of Florida. The students received one experimental credit for their

participation in the study.

Apparatus and Materials

Apparatus and materials were the same as in the unprimed condition of

Experiment 3.

Procedure

The procedure was identical to the procedure in the unprimed condition of

Experiment 3. The experiment lasted from 10 to 15 minutes.

Results and Discussion


The results obtained from the groups of Spanish-English bilinguals and English

monolinguals were analyzed together with the results obtained from Russian-English

bilinguals in the unprimed condition of Experiment 3.

A 3x3 mixed factorial design was used in the analysis the data. As in Experiment

3, the within-subject independent variable had three values corresponding to the distance

between the prototypical meanings of the words in a pair, i.e. closely related, distantly

related, or unrelated. The language variable (dominant language) was distributed between

the three groups of English, Russian and Spanish native speakers. Reaction times in








milliseconds and the percent of positive judgments in relatedness decisions were

assessed as dependent variables.

The data were trimmed according to the criteria adopted in Experiment 3. The

remaining data from 8 monolingual English speakers, 10 Spanish English and 12

Russian-English bilingual speakers were analyzed. ANOVA was performed on mean

reaction times and percentage of positive judgments using participants as random factors.

Relatedness Judgments

Figure 6-3 demonstrates the percentage of positive judgments as a function of

network distance for the three groups of speakers. The results demonstrated a highly

significant main effect of the network distance, F(2,81)=88.850, p<.001. Although the


S- English
-U- Russian
-- Spanish

unrelated distant close


Figure 6-3. Mean percentage of positive judgments on word meaning relatedness as
a function of organizational differences in conceptual structures of bilingual and
monolingual speakers. (English = English monolinguals, Russian = Russian-English
bilinguals, Spanish = Spanish-English bilinguals)








analysis of positive judgments on word relatedness demonstrated the effect of language as

non-significant, paired sample t-tests demonstrated that the speakers of the three

languages provided different judgments about word relatedness in pairs. Thus, the tests

demonstrated statistical significance in judgments provided by English monolinguals and

Russian-English bilinguals for unrelated and distantly related word pairs (t=1.936,p-.08,

and t=-2.587,p=.02, respectively). The responses of Russian-English and Spanish-

English speakers revealed marginally significant difference in their judgments for only

distantly related word pairs (t=1.622, p=. 1). The judgments of distantly related words

provided by English monolinguals and Spanish-English bilinguals only approached

significance at t=-1.00, p=. 1.

The absence of a significant difference in the judgments of English monolinguals

and Spanish-English bilinguals as compared to the difference observed in the judgments

of English monolinguals and Russian-English bilinguals can be a result of greater

similarity in the relational organization of the stimulus English lexical items and their

Spanish equivalents. The greater differences in responses of English monolinguals and

Russian-English bilinguals can also be attributed to significantly less extensive

experience with the L2 conceptual structure. Unlike Russian-English bilinguals, Spanish-

English monolinguals attended high schools in the US and considered themselves to be

equally fluent in both languages. One may argue that the latter may serve as counter-

evidence for the hypothesis being tested; that is, the differences in judgments of different

language groups reflects the proficiency level of bilingual speakers. However, both

bilingual groups produced comparable response latencies (Figure 5-4), which were

significantly longer than the latencies produced by monolingual speakers. The combined








evidence suggests that proficiency is only one of the factors affecting bilingual

performance. Another factor is distinct relational organization within the conceptual

structures of a bilingual's two languages. As predicted, the major differences were

demonstrated in the relatedness judgments of the distantly related word pairs. These

differences and the marginally significant interaction between language and distance

variables (F(4,81)=2.382,p<.5) support the assumption that language-specific relational

organization influences the performance of bilingual speakers and therefore should not be

disregarded by models of bilingual memory.

Response latencies

Figure 6-4 presents the mean response times as a function of the semantic network

distance for bilingual and monolingual speakers. As in Experiment 3, the response times

demonstrated a main effect of language in the between-subject condition, F(2,81)=

12.440, p<.01. The reported significance of the language variable was due to a great

difference between the response times of bilingual and monolingual speakers. The

response latencies of bilingual speakers were approximately 650 ms longer than the

response latencies of monolingual speakers. The paired sample t-tests between the

response latencies of Russian-English and Spanish-English bilinguals were performed to

investigate whether language-specific organization causes differences in the response

latencies of bilingual speakers. The tests did not reveal significant differences in response

times of the two bilingual groups. This may be due to an overall increase in response

times associated with second language processing. Unlike monolingual speakers,

bilingual speakers require additional processing time to recognize the differences between








the semantic structures of the two languages and to suppress the dominant language in

cases where its structure does not coincide with the structure of the language being used.


unrelated


distant


Figure 6-4. Mean response latencies (in seconds) as a function of organizational
differences in conceptual structures of bilingual and monolingual speakers. (English
= English monolinguals, Russian = Russian-English bilinguals, Spanish = Spanish-
English bilinguals)



The effect of network distance was not significant (p>.5). The paired sample t-

test between the values of the semantic network distance variable demonstrated statistical

significance only in the response times of Spanish-English bilinguals between closely and

distantly related word pairs (t- 2.235, p=.01). The latency differences between other

conditions were not statistically significant. The analyses of response times did not reveal

significant interactions between language and distance variables. As suggested earlier, the

fairly long stimulus and inter-stimulus intervals (totaling 2 seconds) might have been why


--English
- Russian
-*- Spanish








some effects were not significant. Shorter stimulus and interstimulus intervals may not

provide enough time for participants to explore the semantic networks of the stimulus

lexical items. Furthermore, bilingual speakers may not have enough time to recognize

differences in the lexical-conceptual organization of L1 and L2 conceptual structures.

Therefore, experimental results may exhibit greater language-specific differences in both

relatedness judgments and latency data produced by monolingual and bilingual speakers.

Experiment 5


This experiment was conducted to examine whether shorter stimulus and inter-

stimulus intervals would reveal greater language-specific differences in the data of

bilingual and monolingual speakers. In addition, in the primed condition an intervening

comprehension task was removed to obtain more salient priming effects.

Method


Participants

Twenty seven monolingual speakers of American English and 38 Russian-English

bilinguals participated in the experiment. All Russian-English bilinguals were native

speakers of Russian who had lived in an English-speaking environment for at least 3.5

years (6.2 on average) and had studied English as a foreign language for at least 2 years

(5.8 on average) before coming to the US. The age of the participants ranged from 8 to

39. At the beginning of each experimental session, bilingual participants were requested

to fill out a questionnaire about their language learning experiences and were also asked

to evaluate their L2 proficiency. Participants noted their dominance in Russian. They did








not consider themselves as balanced bilinguals. They rated their second language

proficiency at an advanced level for most aspects of L2 production and at a near-native

level in some areas of L2 production and most areas of L2 comprehension. No

compensation was offered for participation.

Materials

Two hundred word pairs were generated by the experimenter on the basis of the

stimulus materials used in the pervious relation assessment experiments. Sixteen native

speakers of American English who did not participate in the experimental session were

asked to rate the meaning relatedness of the words in a pair. Ratings were performed on a

seven-point scale. The maximum value (7) was used to denote the greatest relatedness,

and the minimum value (1) was used to indicate the absence of such a relationship.

Stimulus materials were selected on the basis of rating results. Forty word pairs rated 5 or

higher with standard deviation of less than 1.5 were selected as the group of closely

related words. Forty word pairs rated 2 or lower with a standard deviation of less than 1.5

were selected as the group of unrelated words. A group of distantly related words

comprised forty word pairs with a mean rating between 2.5 and 4.5 and with standard

deviation higher than 1.5. In each group, there were thirty target and ten filler word pairs.

A priming sentence was provided for each stimulus pair. The sentences were

selected from authentic English texts available through the LEXIS/NEXIS online news

service. The sentences comprised less than four lines on a computer screen. The sentences

were not rated according to their potential to prime particular relationships between the

words in a pair. The stimulus materials were divided into two sets with an equal number








of unrelated, closely and distantly related words in each set. Unrelated, closely and

distantly related words appeared in random order. Both sets were presented to all

participants. The sets were counterbalanced across priming conditions in such a way that

half of the participants responded to set 1 in the unprimed condition and to set 2 in the

primed condition, while the other half of the participants responded to set 2 in the

unprimed condition and set 1 in the primed condition.

Apparatus

The apparatus was the same as in Experiment 3.

Procedure

The experimental design was the same as in Experiment 3. The stimuli were

presented in two blocks. In the first block the stimuli were given in the unprimed

condition, and in the second block the stimuli appeared in the primed condition. There

were 60 trials in each condition.

In the unprimed condition a fixation point was shown on the screen at the

beginning of every trial. It was visible for 1 second and was immediately followed by the

stimulus. The first word in a pair was visible for 340 ms and was followed by the second

word after a 60 ms inter-stimulus interval. The timer was set to go off the moment the

second word appeared on the screen. Participants were instructed to press color-coded

keys as they made a decision: green for 'yes' responses, red for 'no' responses, and white

if they did not know the wordss. The white key was positioned in the center of the

keyboard. Left and right assignment of the red and green keys was counter balanced








across participants. There was a one-second break before the next trial. On average, it

took about five minutes to complete the task in the unprimed block.

In the primed condition every trial was preceded by a sentence or two. The

sentences comprised less than four lines on a computer screen. Participants were

instructed to read a sentence at a comfortable speed. They could proceed to an

experimental trial by pressing the space bar. The experimental task and the procedure

were the same as in the unprimed condition.

Both conditions were preceded by a short practice session to ensure consistent

performance in the test. Participants were tested individually. The experimental session

lasted from 25 to 30minutes.

Results and Discussion


Latencies with values greater than 2.5 standard deviations above a given

participant's mean response time were excluded from the analysis. Data sets of the

participant who did not respond to more than 20 % of trials (including those that were

removed when the latency data was trimmed) were not included in the analysis. The

remaining data from 36 bilingual and 26 monolingual speakers were analyzed. Half of the

data were obtained in response to stimulus set 1 and the other half was produced in

response to stimulus set 2. Since the purpose of the experiment was to demonstrate

language-specific and context effects on word stimulus processing, ANOVA were

performed on the same set of stimuli using participants as random factors. As a result,

separate analyses were performed for each stimulus set.









Stimulus Set 1


Figure 6-5 presents the proportion of positive responses on word meaning

relatedness as a function of organizational differences in conceptual structures of

bilingual and monolingual speakers. A 3 x 2 x 2 ANOVA was performed on proportion

of positive responses using participants as random factors. As predicted, the analysis


1.00
4 0.90
g 0.80
0
- 0.70
0 0.60
0.50
g 0.40
o 0.30
S0.20
0.10
0.00


Unrelated


Distant


Close


Figure 6-5. Stimulus set 1: Proportion of positive responses on word meaning
relatedness as a function of organizational differences in conceptual structures of
bilingual and monolingual speakers.



demonstrated significant main effects of distance and language (F(2,174)=303.82, p<.01

and F(1,174)=7.40, p<.01, respectively). The effect of priming only approached

significance, F(1,174)=4.20, p<. ;). Two-way interactions between distance and

language, distance and priming, and language and priming, and a three-way interaction


----------------_^' ---






Bilingual Unprimed
/- -- Bilingual Primed
0 Monolingual Unprimed
-- M-- Monolingual Primed








between distance, language and priming were also significant (F(2,174)=9.29, p<.01;

F(2,174)=10.89,p<.01; F(1,174)=622.04,p<.01; and F(2,174)=7.19, p<.01;

respectively). As in Experiment 3, the responses of bilingual and monolingual speakers

were equally affected by priming. The positive judgments provided for distantly and

closely related words were significantly facilitated by the prior activation of a relevant

contextual frame. Facilitation was greater in responses to distantly related words. On

average, there was an 18 % increase in positive judgments provided for distantly related

words as compared to a 7 % increase in positive judgments provided for closely related

words. The proportion of positive responses to unrelated words was not affected by

priming. The findings provide additional evidence for a network configuration as adopted

by the proposed model. Since the meaning relation in closely related word pairs exists

between prototypical senses, the relationship can be easily established, even when the

lexical items are presented in isolation. For distantly related words, the relation exists

between non-prototypical meanings and may not always be established without sufficient

activation of a relevant contextual frame. Therefore, contextual priming should be

especially beneficial for distantly related word pairs. On the other hand, contextual

priming should not affect decisions provided in response to unrelated words. Presumably

this results from the fact that the senses in the relational networks of such lexical items

cannot be used to establish a relation in meaning.

The results also demonstrated greater language-specific differences. Although

only English stimuli were presented, the responses of Russian-English bilinguals suggest

a substantial activation of LI lexical-conceptual organization. The relations between L2

words might be obscured because of a distant or nonexistent relation between equivalent








L2 lexical items. For example, the meaning relations between closely related words in

pairs such as 'watch' 'guard', 'sharp' 'clever', 'drain' 'empty', are not as easily

established between the Russian equivalents of the individual lexical items. Distant

meaning relations between words in pairs such as 'shoulder' 'edge', 'bug' -'defect',

'bed' 'foundation', 'foot' 'base' do not exist between the Russian equivalents of these

lexical items. As a result, the proportion of positive responses in the bilingual data was

significantly lower than the proportion of positive responses in the monolingual data.

Furthermore, the magnitude of contextual effects was different in the data of the two

language groups. The latter was reflected in a highly significant interaction between

language and priming. Positive responses of monolingual speakers to distantly related

words demonstrated a 21 % increase in the primed condition as compared to a 16 %

increase reported in the data of Russian-English bilinguals.

Figure 6-6 presents mean response latencies of bilingual and monolingual

speakers as a function of priming and network distance. A 3 x 2 x 2 ANOVA performed

on mean response latencies using participants as random factors demonstrated that effects

of distance, language and priming were not significant (p>.5). Although two-way

interactions between distance and language, and distance and priming were not

significant (F<1), an interaction between language and priming was significant,

F(1,174)=9.56, p<.01. A three-way interaction between distance, language, and priming

was not significant (p>.5). As predicted, the latency data was also affected by shorter

stimulus and inter-stimulus intervals. As compared to the latency data reported in

Experiments 3 and 4, there was no significant difference in the latencies of monolingual

and bilingual speakers. It may be argued that the latency effect could be a function of










2.40
4 -Bilingual Unprimed
g 2.20 Bilingual Primed
S 2.00 O -Monolingual Unprimed
S .0 -Monolingual Primed
1.80
1.60
1.40
C
S 1.20
1.00
Unrelated Distant Close





Figure 6-6. Stimulus set 1: Mean response latencies of bilingual and monolingual
speakers as a function of priming and network distance.



bilingual fluency; that is, more fluent bilinguals may produce latencies comparable to the

latencies of monolingual speakers. Nonetheless, fluency cannot be the only factor

affecting the latencies of bilingual participants. As shown in Experiment 4, regardless of

the difference in L2 fluency, Russian-English and Spanish-English bilinguals produced

equivalent latencies. Therefore, it is reasonable to assume that the shorter response

latencies were triggered by shorter stimulus and interstimulus intervals. Since shorter

presentation times did not provide an opportunity for a thorough network search,

responses of bilingual and monolingual speakers demonstrated more salient language-

specific and contextual effects as compared to the data reported in Experiments 3 and 4,

where longer response latencies were associated with less conspicuous language-specific

and contextual effects. Furthermore, contrary to the findings reported in Experiments 3








and 4, response latencies of both bilingual and monolingual speakers were slightly longer

in the primed condition than in the unprimed condition. A slight increase in processing

times in the primed condition might be due to a shorter break between the presentation

of a sentence prime and a stimulus word pair. Since an intervening comprehension task

was not used in this experiment, the relation between a prime and a stimulus word pair

became more apparent. Some effects reported in the latency data might reflect a strategy

adopted by the participants. Once the participants noticed that context could help to

establish the meaning relation between word stimuli, they were compelled to discover a

meaning relation in all pairs. The longer latencies produced by monolingual speakers in

the unprimed condition in response to unrelated stimuli could be due to such strategic

processing.

Stimulus Set 2

Figure 6-7 presents the proportion of positive responses on word meaning

relatedness as a function of organizational differences in the conceptual structures of

bilingual and monolingual speakers. A 3 x 2 x 2 ANOVA was performed on the

proportion of positive responses using participants as random factors. The analysis

demonstrated significant main effects of distance and language (F(2,174)=364.51, p<.01

and F(l,174)=17.93,p<.01, respectively). The effect of priming was only marginally

significant, F(1,174)=5.50, p<.05;). Two-way interactions between distance and

language, distance and priming, and language and priming, and a three-way interaction

between distance, language and priming were also significant (F(2,174)=19.58,p<.O1;

F(2,174)=25.79,p<.01; F(1,174)=762.68,p<.01; and F(2,174)=16.83,p<.01;








respectively). Contrary to the data obtained for set 1, the responses of monolingual

speakers to set 2 demonstrated a negative priming effect. That is, the proportion of

positive responses in the unprimed condition was greater than in the primed condition.

According to the proposed model, such an effect could be due to the activation of

incompatible contextual frames. In other words, target relations between word stimuli

were obscured by the contexts provided.


1.00
S0.90
0.80
0.70
S 0.60
0.50
S0.40
0 0.30
. 0.20
S0.10
0.00


Unrelated


Distant


Close


Figure 6-7. Stimulus set 2: Proportion of positive responses on word meaning
relatedness as a function of organizational differences in conceptual structures of
bilingual and monolingual speakers.



A post hoc rating of sentence stimuli was performed by fifteen native speakers of

American English in order to uncover any incompatibility between the contextual frames

activated by priming sentences and the target meaning relations. Only the sentences that

were used to prime meaning relations between closely and distantly related words were


*.O






--Bilingual Unprimed
-Bilingual Primed
S-- Monolingual Unprime
o0 Monolingual Primed








rated. The sentences were rated on a seven-point scale according to the degree of

facilitation they could provide. The maximum value (7) was used to indicate maximum

facilitation, and the minimum value (1) was used to denote the lack of any facilitation. On

average, set 1 and set 2 were rated at 5.14 and 4.58, respectively. A t-test performed on

the two means demonstrated a significant difference between the values assigned to the

sentences in the two sets, t = 2.669, p<.02. As follows from the model and the analysis of

priming sentences, the negative effect of priming reported for stimulus set 2 was due to

the greater number of sentences that activated contextual frames incompatible with the

target meaning relations.

Unlike the monolingual group, Russian-English bilinguals demonstrated a

positive priming effect. Nonetheless, the responses of bilingual and monolingual speakers

demonstrated greater compatibility in the primed condition than in the unprimed

condition. This fact supports the prediction that the data obtained from bilingual and

monolingual speakers in the primed condition should reveal comparable patterns of

activation.

Figure 6-8 presents mean response latencies as a function of priming and network

distance. A 3 x 2 x 2 ANOVA performed on mean response latencies using participants

as random factors demonstrated that only the interaction between language and priming

was significant, F(l,174)=10.02, p<.01. The effects of other variables and their

interactions were not significant (p>.5). As with the relatedness judgments, the latency

data of bilingual and monolingual speakers demonstrated greater compatibility in the

primed condition. In the unprimed condition, the processing patterns of bilingual and

monolingual speakers were influenced by dominant conceptual structures. Conversely, in








the primed condition the processing patterns were constrained by activated contextual

frames. This fact provides additional support for the assumption that embedding the

lexical stimuli in a context reduces the prototype effects associated with a particular

organization within the dominant conceptual structure.


- 1.80
1.70
1.60
C, 1.50
S 1.40
8 1.30
O 1.20
1.10
1.00


Unrelated


Distant


Close


Figure 6-8. Stimulus set 2: Mean response latencies as a function of priming and
network distance.



The relation assessment experiments reported here were conducted to evaluate the

relational network approach proposed for the model of conceptual structure mapping.

Relational networks were demonstrated to be sensitive to the factors causing instability in

the graded structure of lexical categories. The relational network organization of the

bilingual lexicon constitutes a dynamic structure able to accommodate language-specific

and language-independent patterns of bilingual production.


U'
-* a Bilingual Unprimed
--- Bilingual Primed
- Monolingual Unprimed O
-8- Monolingual Primed













CHAPTER 7
GENERAL DISCUSSION

The model of conceptual structure mapping was proposed to provide an alternate

approach to bilingual lexical and conceptual representation. In particular, it was designed

to account for the activation patterns in the memory of bilingual speakers which can not

be accounted for by the models that maintain a strict dichotomy between lexical and

conceptual levels of processing. The proposed model contends that the patterns of lexical

and conceptual activation in bilinguals are largely determined by an interaction between

language specific lexical-conceptual representations. Such interaction is a function of a

number of structural and developmental factors. The role of these factors is explicated in

the structural and developmental assumptions of the model.

The structural assumption is elaborated on the basis of the conceptual structure

notion that offers a means of accounting for the prototype effects that arise in the graded

structures of lexical categories in various socio-pragmatic and functional contexts. The

developmental assumption offers a means of accounting for the differences observed

between the performances of more and less fluent bilinguals. A bilingual has conceptual

structures associated with L1 and L2, which I call CSI and CS2. The mapping metaphor

describes CS2 development as a process of mapping L2 lexical items onto converging

conceptual representations in CS1. CS2 development is a result of the development and

expansion of relational networks and inter- network connections. Relational networks of








individual lexical items evolve when a lexical item is consistently used in different

contextual frames. Areas in CS and CS2 that are mutually incompatible have been

demonstrated to be the major reason for production deficits such as negative transfer and

longer response latencies.

Since the notion of conceptual structure is central to the problems of lexical-

conceptual activation, an experimental data analysis without a detailed account of the

bilinguals' response types is incomplete. As demonstrated in the experiments reported

here, a quantitative analysis of latency and accuracy data do not provide adequate

evidence of the processes that govern lexical and conceptual access in the bilingual

memory. Thus, in the first replication of the Sholl et al. (1995) study, separate analyses of

the overall and primed responses in the translation data provide discriminative results.

Since the analysis of the primed accuracy did not reveal any effects of priming, the

effects reported in the overall data analysis can not be interpreted as transfer from picture

naming to translation. Similarly, an analysis of the primed accuracy data in the modified

replication of the Sholl et al. (1995) study demonstrates that the inhibitory effect in Ll-to-

L2 translation following the presentation of L1 word primes was not a result of transfer.

The combined evidence from the two experiments can not be accounted for by previous

models. This inability results from the fact that they do not consider the development of

relational network organization in a bilingual's two languages as central to the issues of

lexical and conceptual processing. Such models do not possess the theoretical constructs

that account for unprimed responses and predict the conditions of their occurrence.

According to the structural assumption offered in the proposed model of

conceptual structure mapping, lexical items are related to many frames and may activate a








whole network of senses. Inasmuch as bilinguals' conceptual structures exhibit

differences in their lexical-conceptual organizations, there is a possibility of obtaining a

number of responses to a single translation stimulus. As a result, bilinguals may provide

responses different from the primed (or target) lexical forms. Conditions for the

production of unprimed responses can be elaborated on the basis of both the structural

and the developmental assumptions. The structural assumption provides a functional

construct (i.e. a frame) to demonstrate that the development of a relational network

organization is a result of lexical items' use in contrastive socio-pragmatic and functional

contexts. The activation threshold of a particular sense in the relational network of a

lexical item depends on (1) the degree of a speaker's familiarity with the given usage, (2)

the stability of the graded structure of the given lexical category, (3) representational

differences in the relational network, or differences in weights assigned to different

frames in the graded structure of the lexical category in CS1 and CS2. The relation

between the specified factors is implicational. That is, if a speaker is not familiar with a

particular sense, other factors do not apply, and if the graded structure of the acquired

lexical category is highly stable, language-specific differences may not produce a

significant effect. For example, the English word nail may be translated in Russian as

nogot' and gvozd' to denote a horny growth at the end of fingers and toes, and a pointed

piece of metal, respectively. If Russian-English bilinguals were familiar with the former

sense, but not with the later sense, the Russian lexical representation gvozd' ('pointed

piece of metal') would not be activated at all. Since the latter sense is not associated with

highly stable prototype effects in the relational network of the English lexical item nail

(i.e. its activation depends on availability of a relevant contextual frame), its activation








threshold in the Russian conceptual structure can be attenuated by differences in the

relational organization of English and Russian lexical categories.

The proposed implicational relation between developmental and structural factors

may be criticized on the grounds that it may lead to highly individualized descriptions,

whereas the objective of any theoretical analysis is to capture a general trend. However, it

is important to note that early stages of conceptual structure development are inherently

idiosyncratic, because language acquisition episodes reflect personal language acquisition

experiences of an individual. Thus, a CS2 of a novice L2 learner constitute a set of

isolated lexical representations that are mapped onto converging nodes in CS1. Since

only one particular sense in a given relational network is acquired, novice bilinguals may

demonstrate greater variability in their responses to single word stimuli. Growing

experience with L2 provides a learner with a number of contrastive contexts that enable a

novice bilingual to develop relational networks and inter-network connections for the

representation of a word. As CS2 evolves, a bilingual's familiarity with a variety of

senses in any given relational network will increase. Hence, bilingual production will be

less susceptible to individualized developmental factors and more open to prevalent

structural constraints, such as graded structure stability and language-specific differences

in representational organizations of CS 1 and CS2.

One of the major issues addressed by the model proposed here is that the patterns

of bilingual lexical and conceptual processing are not reduced to a simple dichotomy. The

effects of lexical and conceptual factors were examined in the modified replication of the

revised hierarchical model (Sholl et al. 1995) study. Contrary to the predictions that

follow from this model, there was no significant facilitation of responses to the stimuli




Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EBFQF968U_NEXPV0 INGEST_TIME 2013-10-09T22:46:07Z PACKAGE AA00017623_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES