- Permanent Link:
- The role of phonology in reading-the verification hypothesis and Korean learners of English as a second language
- Copyright Date:
- Subjects / Keywords:
- Automated word recognition ( jstor )
English language learners ( jstor )
Foreign language learning ( jstor )
Modeling ( jstor )
Phonemes ( jstor )
Phonemics ( jstor )
Phonology ( jstor )
Trucks ( jstor )
Word frequency analysis ( jstor )
Words ( jstor )
- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- Copyright the author. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
- Embargo Date:
- Resource Identifier:
- 51740175 ( OCLC )
This item is only available as the following downloads:
THE ROLE OF PHONOLOG Y IN READING â€“ THE VERIFICATION HYPOTHESIS AND KOREA N LEARNERS OF ENGLIS H AS A SECOND LANGUAGE By DANIEL HUFNAGLE A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLM ENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2002
This thesis is dedicated to Mom, Dad, and Hyeran.
iii ACKNOWLEDGEMENTS I would like to thank everyone who has helped me get to this p oint. Of course, it takes more than a student to write a thesis, but without these people, I could not have grown and matured as a scholar or a person. My advisor, Theresa Antes, deserves the first and most vigorous nod of appreciation. She read my dr afts many times, gave sage like suggestions, and extended more patience to me than I deserved. Just as important, she assured me that I was doing the right thing, keeping my confidence level high when I was sure that I had made a disastrous mistake or lag ged too far behind schedule. Many thanks are also extended to the other member of my committee, Ratree Wayland, who helped with many of the technical aspects of this study. Without her, I would have had no sound files, no computer program, and thus no thesis. Despite her enormous workload, she took the time, something she did not have to spare, to work with me. A special expression of appreciation goes to Ann Wehmeyer, who listened to me ramble for an entire semester in her Writing Systems class about reading and phonology and yet still encouraged me. She also helped a lot in the initial stages of this study and read the thesis for the defense. My colleagues, professors, and friends in the Program in Linguistics all deserve some praise for their help and friendship in my three long years as a
iv masterâ€™s student. I would especially like to mention Douglas Adams, Phil Monahan, Hee nam Park, and Sanghee Yeon for their friendship and, occasionally, guidance. Each of them has helped me to see in new ways. I give thanks to Hyeran Bae, my love and light, for keeping my spirits up and pushing me to go farther than I ever thought possible. I would also like to thank her (in addition to Hyungjin Bae and Sunghee An) for delivering multitudes of participants t o me. For turning me towards the wonder of linguistics, I would like to thank Jamie Womack, who lent a few linguistics books to me when I had no idea where to turn academically. I was hooked from the moment I picked up the first book. Finally, I tha nk my family, who brought me to this world and yet encouraged me to reach for greater ones. During my eight years at the University of Florida, they have seen me through five different majors and never doubted that I would end up doing something that I lo ve.
v TABLE OF CONTENTS page ACKNOWLEDGEMENTS iii ABSTRACT vii CHAPTER 1 INTRODUCTION ..................................................................... 1 2 PSYCHOLINGUISTIC MODELS OF WORD RECOGNITION .................................................. 4 Introduction ...................................................... ........................ 4 Visual Model ............................................................................ 4 Dual Route Model .................................................................... 8 Phonological Model ................................... .............................. 11 Verification Hypothesis ............................................................ 13 Nature of Phonological Representation ................................... 19 Nature of Word Recognitio n in Relation to Reading .............................................................. 21 Summary .................... ........................................................... 25 3 INTEGRATION KOREAN LEARNERS OF ENGLISH AS A SECOND LANGUAGE ............ .............. 26 Introduction .............................................................................. 26 Korean Consonant Phonemes ................................................. 26 Korean Versus English Phonology ................. .......................... 27 Predictions from the Model ....................................................... 2 9 Conclusion ................................................................................. 31 4 EXPERIMENTAL METHODOLOGY ............ ............................. 32 Goals and Variables...... .. 32 Participants .......................................................... 33 Procedure .................................................................................. 34 Hypotheses. 39
vi 5 RESULTS AND DISCUSSION ............................................... ... 40 Treatment of Data ........................................................................ 40 Descriptive Statistics .................................................................... 42 Lexical Decision Task .............................................. .................... 45 Phonemic Distinction Task ........................ ............................... ... .. 50 General Discussion and Conclusion ........... ............................... .. 51 APPENDIX A INFORMED CONSENT .................................... ........................... 55 B PHONEMIC DISTINCTION WORD LIST WITH POSSIBLE ANSWER CHOICES .................................... 57 C BACKGROUND QUESTIONNAIRE .......................................... 61 D LEXICAL DECISION TASK STIMULUS LIST BY PHONEME AND CONDITION .......................................... 63 REFERENCES .............................................................................. 65 BIOGRAPHICAL SKETCH... 68
vii Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillm ent of the Requirements for the Degree of Master of Arts THE ROLE OF PHONOLOG Y IN READING â€“ THE VERIFICATION HYPOTHESIS AND KOREA N LEARNERS OF ENGLIS H AS A SECOND LANGUAGE By Daniel Hufnagle May 2002 Chair: Dr. Theresa A. Antes Major Department: Lin guistics This study examines the role of first language phonology in second language reading in Korean learners of English. There is a discussion of various psycholinguistic reading models, with the conclusion that recognition of single words is carried out by a primarily phonological process which includes a verification mechanism. Then, implications for second language learners are discussed, with a focus on how Korean learners of English deal with the differences between Korean and English phonologic al systems. The experiment that follows included a phonemic distinction task as well as a lexical decision task. Both tasks utilized words with phonemes that are problematic for Korean learners of English: /f/, /p/, /d/, /t/, /r/, and /l/, as well as co ntrol words. The lexical decision task used three conditions: normal words,
viii visually similar nonwords, and pseudohomophones (i.e., nonwords, which, when processed by Korean phonology, are homophonous with normal words). It was found that words containin g certain phonemes were more difficult to process (i.e., had longer reaction times and higher error rates). In addition, the error rates of words in the lexical decision task were correlated with error rates in the phonemic distinction task by phoneme. I t was also found that pseudohomophonous nonwords were more difficult to process than visually similar nonwords. These findings lend support to the notion that reading is a phonologically based phenomenon. It was also found that pseudohomophonous nonwords took an average of 140ms longer than their word counterparts to process, which lends support to a verification mechanism added to the basic phonological decoding scheme.
1 CHAPTER 1 INTRODUCTION Researchers from such diverse fields as linguistics, p sychology, education (both first and second language), and neurology have long been interested in how readers obtain meaning from print (Bernhardt 1991, Smith 1994). Because of this wide base, many of the theories of reading that have been developed are i ncompatible or overlook relevant findings in other areas. On the one hand, there is the classic top down versus bottom up debate that has plagued educational research and which concerns primarily the incorporation of background knowledge in the reading pr ocess (Bernhardt 1991). Psychological models, on the other hand, vary from being based entirely on visual features to being reliant entirely on phonological principles (Frost 1998). By facing some of these differences head on in a very specific area of reading, it may be possible to narrow the gap between various models of reading. This type of analysis may help answer the so called big question: Is there a single theory of reading that can work across languages and writing systems (a cognition centered approach), or do we require several models to accommodate variations across languages and scripts (a language specific approach)? In other words, is there a single mental process that can account for reading in all languages, or do readers of different w riting systems (or types of writing systems) use fundamentally different processes?
2 One of the areas of reading research that is still full of these types of questions is word recognition. In fact, there are several recent articles that, after more tha n thirty years of research, attempt to spell out more precisely the basic definitions and axioms inherent in different models (see, e.g., Stone and Van Orden 1994, Frost 1998). After such a long time, a casual observer might think that these basic notions should be unambiguously defined. Clearly, the continued need for this type of research demonstrates that the area is stuck in a quagmire. However, with the help of the insights available from a variety of disciplines, I believe it is possible to constru ct a theory of basic word recognition that is a cognitively based, strongly phonological model of reading. It is important to note from the outset that the model supported here implies reading and recognizing single words outside of the context of a sent ence. It does not address contextual effects such as semantic priming because these effects would make the model vastly more complicated. In addition, the model advocated here should be able to plug in to the models of reading that look at reading more b roadly. Contextual effects are touched on briefly later, in the section that describes the nature of the phonological representation, and how this model connects to broader reading models is discussed in the concluding section. This paper takes a step tow ards establishing a phonological reading model by describing some models of word recognition and arguing for a phonological model of reading with a verification module. Then, a psycholinguistic experiment that tested the predictions of this model regardin g
3 the types of problems that Korean learners of English might have when reading English is described and analyzed in terms of all models presented.
4 CHAPTER 2 PSYCHOLINGUISTIC MODELS OF WORD RECOGNITION Introduction Before it is possible to synthe size the models of reading, there must be a description of the various models that currently exist. The models will be discussed in terms of their major features and the assumptions behind them. A look at some studies within each model will provide both evidence for and criticism of the various proposals. This section examines reading from a psycholinguistic perspective, including purely visual, dual route, and purely phonological models. The evidence points towards a phonological theory of written word recognition. The question that follows is how phonology is accessed in the minds of readers. It will be argued that readers assemble a phonological representation from print by combining graphemes (+++ /grf/) rather than accessing the ph onological forms of words as wholes ( /grf/). Visual Model Strong visual models of word recognition assume a sequence of activation such as the following: Print Meaning ( Phonological representation) Figure 1 The Vis ual Model
5 For these models, the written representation of a word can directly activate the meaning, without recourse to sound. 1 It must be noted that no model reviewed here totally ruled out a sound based route to the activation of meaning. Baluch and Besner (1991) state the consensus view of those who adhere to visual activation when they say that sound based routes are used in only two cases: when beginners are learning sound based (specifically alphabetic and syllabic) scripts and when experienced re aders are reading low frequency words. Consequently, strong visual models do not prevent phonology from having a role, but that role is generally a small one that is essentially absent from fluent reading in any orthographic system. 2 Baluch and Besner (1991) used an oral reading task in Persian to determine whether or not native speakers used a visual word recognition scheme. Persian has the convenient feature of being simultaneously phonologically opaque in some cases (a full phonological representati on is not accessible simply by looking at a word) and transparent in others (a full representation is available form the graphic form). This is because some vowels are required to be written with vowel letters, while others are optionally represented usin g diacritics, as in Arabic or Hebrew. Participants were asked to read target words 1 The optionality of the activation of the phonological representation in the above schematic is due to the fact that phonological representation is present in some models that say that sound representations are automatically generated but postlexical, and n ot in others that claim that phonological activation is not automatic at all. The distinction is not relevant here. 2 In no case have I seen an explanation of how a beginning reader shifts from a phonological route to a visual one. Practically, explai ning this shift would mean explaining how a word becomes a so called sight word, or how automaticity works. Though not critical to the argument here, I feel this is an important point that should be explored further.
6 aloud, the important measure being the reaction time between presentation and the initiation of vocalization. Their findings are commonly represented throughout the lite rature of visual access proponents: targets with higher word frequencies yield shorter reaction times. In other words, given any two words, the one that is more frequently accessed in a personâ€™s lexicon will be accessed faster by that person than the one that is less frequent. The word frequency effect is the hallmark of the visual model. The argument of visual model enthusiasts regarding word frequency effects proceeds as follows: if words are assembled phonologically bit by bit, then how can it be that words that are more common are retrieved faster? Surely phonological assembly cannot simply move faster if the printed material that is being examined is a common word, given that the mind does not know that the string is a more common word until it has finished processing the printed form. The word frequency effect has been shown in many language types, from orthographically transparent Persian (Baluch and Besner 1991) to more opaque 3 orthographies such as French (Peereman et al. 1998). This is one o f a few distinct advantages that visual models seem to have over phonological ones. Another advantage is the word superiority effect, which represents the fact that words are recognized faster than non words (Baluch and Besner 1991, Frost 1998). A final advantage is simplicity of the reading system along the lines of Occamâ€™s Razor: obtaining lexical access directly from visual form is certainly less 3 Transparent orthographies are those in which the phonological representation of a printed form is readily available. Opaque orthographies, on the other hand, are those in which this relationship is much more complex.
7 circuitous than obtaining a lexical entry from a phonological form, which must in turn be reached via a pr inted form. However, these advantages may not be as persuasive as they initially appear. In an excellent analysis of the word recognition debate, Frost (1998) points out that these are the only aspects of reading that purely visual accounts can predict . The fact that reading ability has been found to correlate much more with linguistic variables (phonological awareness or language disorders, for example) rather than pure visual acuity leads Frost to conclude that purely visual accounts are not tenable (Frost 1998). Other phenomena that occur in word recognition studies, including homophony, regularity, and consistency effects, are not predictable from visual accounts of word recognition and in fact undermine any purely visual accounts of word recogni tion (Frost 1998). Homophony effects include confusion of homophonic words ( their, there and theyâ€™re , for example). Visual accounts have no way to account for this confusion when, fully aware of the distinction among the words, skilled readers do not a lways notice mistakes in sentences such as, â€œThe book is over their.â€ Regularity (or feedforward consistency, depending on the researcher) effects occur when a spelling body maps to a single phonological representation (spelling sound), where words that have regular spelling bodies are recognized faster than those that have irregular spelling bodies (Stone et al. 1997). In other words, < le#> only maps to /l/ and nothing else (e.g. battle , riddle ), as opposed to < int#>, which maps to several different phonological
8 representations ( hint = /hInt/, pint = /pa j nt/). Visual models can not account for any effects relating to sound, unless the effect is found in unskilled readers or low frequency words (Frost 1998). Now, thinking in terms of sound spelli ng, consistency (or feedback consistency) effects occur when a phonological representation maps to a single spelling body (sound spelling), where words with consistent phonological representations are recognized faster (Stone et al. 1997). For example, /nd/ maps to < and#> and nothing else (e.g. , , ), while /ow/ maps to several orthographic forms (, , , ). In sum, strong visual models of word recognition allow for simplicity of the reading circuit as well a s a cogent explanation for the word frequency and word superiority effects. However, these models fail to predict many features that are found in reading. The following section deals with dual route models, which were the next step in the evolution of ps ycholinguistic models of word recognition. Dual Route Model Classical dual route models are an attempt to address some of the problems of visual models, while keeping the basic circuit intact: Print Meaning ( Phonological form) Phonological fo rm (by assembly) Meaning Figure 2 The Dual Route Model Van Orden explained that some researchers have used the term dual route to refer to an even split in the two pathways, while others presume primacy of the visual route (198 7). In the latter case, the phonological pathway either starts after the visual route, or is simply slower and more unreliable when compared to
9 the visual route. Here we will use Frostâ€™s (1998) definition of dual route models, which talks of dual routes distinctly giving preference to the visual mode. Frostâ€™s definition is useful not necessarily because it supports the point of view expressed in this study better (though it does); rather, it is used because this is how most of the dual route studies sinc e Van Ordenâ€™s 1987 study have used it (see e.g. Paap and Noel 1991, Luo 1996). The dual route model is often metaphorically represented as a horse race, with horses (one visual, one phonological) racing down a track to a finish line that represents lexi cal activation (see, e.g., Paap and Noel 1991). In such a race, the â€œhorsesâ€ do not interact, neither helping nor hindering one other. The outcome depends solely on the speed of the fastest route. As a model, having dual processes definitely takes car e of some of the problems listed above, namely, word frequency effects, as well as the homophony effects in low frequency words. However, not everyone is satisfied with the premises of the dual route model (e.g. Frost 1998, Lesch and Pollatsek 1998, Tan a nd Perfetti 1998). Each of these studies agrees that a dual route model can account for the aforementioned problems, but it still does not account for every effect. One problem is the fact that phonological effects are known to be very fast and take pl ace even in deep scripts such as Chinese (Tan and Perfetti 1998) and English (Lesch and Pollatsek 1998), with examples of each described in depth in the sections that follow. The dual route model proposes a horse race, but it is a race that the phonologic al horse is rarely supposed to win. In fact, it often does.
10 Despite this, the phonological route that is envisioned by dual route models is axiomatically necessarily slower than the lexical route in this model. This is due to two factors. First, visu ally based word frequency effects necessarily rely on a phonological processing mechanism that is slower than the visual one, as the effect would never be able to surface if phonological processing were faster than the visual pathways that recognize the pr inted form of the word (Van Orden 1987). Second, there is the widely acknowledged word superiority effect that was mentioned above (Baluch and Besner 1991, Frost 1998). However, the more conceptual theoretical problem is not with the phenomena that the dual route model cannot represent; rather it is the fact that writing represents language, and for virtually everyone, interactions with language are far more often verbal than written. Frost (1998) argues that a model that accesses language without usin g the vast resources that people already have in place for aural processing would be less than optimal. Note that this is not an argument that written language is simply visually encoded speech; rather, it is an argument that our minds already have an eff icient system in place for accessing the lexicon, and it does not make much sense to abandon the efficiency and speed of that system in favor of a purely visual system simply because the route is less circuitous. Thus far, we have seen models that rely on visual pathways as well as visual plus phonological pathways. What can a phonologically dominant model offer us? The next section offers some insight.
11 Phonological Model The most basic strong phonological models of word recognition could be represe nted as follows: Print Phonological representation Meaning Figure 3 The Phonological Model The most obvious disadvantage of this model when compared to the visual model is the introduction of an intermediate step. Why would it be necessary to posit this intermediate level? The research abounds with phonological phenomena that provide evidence for some sort of a phonological component to reading that is prelexical. One example is the phonological interference effect (Luo 199 6). In Luoâ€™s experiment, participants were to judge which of a pair of words (for example lion and bare ) was related in meaning to a third word ( wolf ). The idea was that the phonological form of bare would alternately generate the meaning for bear , thus causing interference in the decision making process. Reaction times and error data of the judgments were compared to visually similar controls (in our example, bean ) and it was found that homophonous words had consistently longer reaction times and/or hig her error rates than visually similar controls, supporting prelexical activation of phonology (Luo 1996). Other experiments using similar semantic relatedness tasks in Chinese one character (Perfetti and Tan 1998) and two character (Tan and Perfetti 199 9a) words have yielded similar results for Chinese. Semantic relatedness experiments are not the only ones that have given support to phonological models of reading. Backward masking experiments have been used to demonstrate prelexical phonological act ivation in English (Tan and
12 Perfetti 1999b). The researchers presented target words, which participants had to write down, for a very brief time (either 28 ms or 42 ms), followed by one of several masks for 28 ms: graphically similar masks, homophonic mas ks, semantically similar masks, control word masks, or pattern masks. 4 At the shorter 28ms target presentation, when word masks tended to cause slower reaction times than control pattern masks, graphically similar masks inhibited reaction times less than homophonic masks, which in turn hurt less than other types. On the other hand, at longer target presentations, when certain masks were associated with faster reaction times than controls, homophonic masks facilitated recognition twice as much as graphical ly similar masks. This data suggests that given enough time to pick up some basic visual information (i.e. enough time to pick out a phonological form), readers can use phonological information to help with word recognition (Tan and Perfetti 1999b). Th is effect has also been found in Chinese (Tan et al. 1996). Given this much evidence, what is wrong with a strong phonological account of reading? As mentioned above, a strong phonological account that has no reference to written forms cannot account f or commonly found frequency effects or word superiority effects (Baluch and Besner 1991). In addition, if readers proceed from graphic form to phonological form to meaning, how can anyone who uses an orthography without one to one phoneme to grapheme 4 Pattern masks provide a visual pattern (such as XXXXX) whose sole purp ose is to fill the visual sensory store with nonsense information. If the target word simply disappeared, nothing would replace the visual information left in the sensory store, leaving it intact for some time. Thus, without a mask, the effective present ation time (time on screen + residual sensory store time) is not known. On the other hand, a mask provides a means to equalize the effective presentation time of a stimulus and its real presentation time.
13 corr espondence ever know if something is spelled incorrectly as long as the letter string is homophonous with the correct spelling? For example, if we only process words in terms of their phonological forms, how can we ever distinguish < bare > and ? The answer lies in the verification hypothesis that was considered by Van Orden (1987) and the work that followed from it. Verification Hypothesis Van Orden (1987) argued that the addition of a word verification mechanism to the strictly phonological model shown above could account for all of the phenomena seen so far. Before discussing exactly how this takes place, it would be beneficial to review research about the time course of word activation, which will give us an explanation of what happens between the time readers see a word on a page and the time the word is retrieved from their internal lexicon. The various models generally contain the same elements: a printed form, a semantic form, and possibly a phonological form. The importance of the pieces varies within the models, and this importance is basically a reflection of the time courses that the models assume for the components of the models. Because it appears that the most extensive time course of word activation has been done in Chinese, 5 I wil l use it as a model. Using various primes and varying the time course of their appearance, Perfetti and Tan (1998) measured inhibition and facilitation of different primes in a Chinese character identification task. They used graphic, homophonic, and 5 Li Hai Tan and Charles Perfetti alone have autho red or co authored over a dozen articles on the time course of word recognition in Chinese. Regarding the time course of activation, there has also been some work done using electro physiological response patterns (ERPs) in English (Ziegler et al. 1999), and it appears that these data shows a similar time course of activation to the Chinese one shown by Tan and Perfetti.
14 s emantically related primes, with each prime type varying by semantic vagueness. 6 In the experiment, primes were shown on a computer screen and then followed by a target character, which remained on the screen until named. The data can be summarized as fol lows: Table 1 Inhibition and Facilitation (in ms) from Perfetti and Tan (1998) Prime Type \ Duration (ms) 43 57 85 115 Graphic 7 +80 8 60 75 15 Homophonic 7 0 +90 +95 +80 Sem. Related (Precise) 0 0 +80 +100 Sem. Related (Vague) 0 0 0 +55 From Table 1, we can see that graphically related primes helped naming at very short (43 ms) durations, and inhibited thereafter. Homophonic primes helped in all but the shortest duration, and had no effect when shown in the short duration. Fi nally, semantic primes helped when shown for the two longest durations (85 and 115 ms). It appears that the path to Chinese character recognition progresses as follows: graphic activation, followed very shortly by sound, and finally followed by meaning. This follows the strong phonological model very well. One question that comes to mind from the information presented in Table 1 is why the graphic primes had an inhibitory effect at presentation times longer 6 The degree of semantic vagueness is not entirely critical here, but a brief explanation can aid understanding of the data. Vagueness can be thought of in terms of the broadness of the semantic field of a word. For example, car is a relatively semantically precise word (even comparing train and trolley cars to the ones we drive), while vessel is semantically vague (think about blood, d rinking, and cargo vessels). 7 For graphic and homophonic primes, there was no difference between semantically vague and precise primes, so this data was combined in the table. 8 Cell data is the number of milliseconds that the prime type facilitated in naming. In other words, +50 represents a 50 ms shorter reaction time in naming the target character. All facilitation/inhibition numbers are my approximations (+/ 5 ms) taken from graphs provided in Perfetti and Tan (1998).
15 than 43 ms. Again, the phonological model he lps to explain. In my view, this model clarifies the situation because the graphically similar primes were real Chinese characters. The basic graphic information in these characters would have initiated the activation of their phonological forms after the initial time that was required for participants to gather essential visual information. These phonological forms, which were different from the phonological forms of the targets, would have interfered with the retrieval of the target charactersâ€™ phonolog ical forms. Another, related, question is why the semantic primes did not also have an inhibitory effect. The answer likely lies in the nature of the decision making process. Figure 4, seen below, shows exactly how each prime matched with their target s at progressive stages of the word recognition process. When this is compared to Table 1, above, we can see that except for graphic primes after 43 ms, each time there was some similarity, there was a faster naming reaction time. Perfetti and Tan (1998) posit that this faster naming time was due to partial lexical activation by the semantic prime. The nature of this activation and its relationship to the phonological model is discussed at the end of this chapter. Time 43 ms 57 ms 85 ms 115 ms Prime Type \ Similarity (+ = match, = no match) Graphic Graphic + Sound Graphic + Sound + Meaning Graphic + Sound + Meaning Graphic + + + + Homophonic + + + Related + Precise + + Related + Vague + Figure 4 Time Course of Activation in Reading
16 The next question to answer is why the graphic primes had an inhibitory effect. If we follow the phonological model, the basic mechanism is the print to sound process. Therefore, at 43 ms, when the graphic prime m atched the target, it helped. The others did not inhibit because, at 43 ms of exposure, they were totally unrelated. In other words, the mind had not yet considered the homophonous (or semantically related) primes to be related to the target because thei r phonological forms (or meanings) were not activated at this point. At 57 ms, the homophonic primes matched, so it helped. The graphic prime had activated a separate phonological form, but with the visual similarity, the two forms were actually competin g against one another. The semantically related primes were still altogether different (as the meaning had not been activated when the prime switched to the target), so there was neither interference nor assistance. As the meaning came to be available (a t 85 and 115 ms), the semantically related primes showed their influence, essentially partially activating a context for the lexical item before visually presenting it. That said, word superiority and frequency effects have yet to be accounted for. Thi s is where Van Ordenâ€™s (1987) verification hypothesis comes in. Van Orden (1987) proposed a variant on the strong phonological model that not only accounted for the homophone effect, but also explained why it happened. Before, it had been generally accep ted that homophone effects, if they really existed, 9 were simply errors or confusion on the part of readers (Van Orden 1987). Van Orden proposed a verification mechanism, whereby readers follow 9 Van Orden (1987) not es that some visual model proponents explained away homophony effects by claiming they were really word frequency effects or by saying that the only readers who exhibited such effects were unskilled readers.
17 the phonological route outlined above (i.e. Print Phonologi cal Form Meaning). Then the route goes through a feedback process, where phonological forms activate several possible meanings (the source of the homophone effect). Each of these meanings activates a spelling form against which the graphic letter string that is printed on the page is checked (Van Orden 1987). Therefore, by this explanation, reading need not be an entirely receptive skill; it can also have a major active component. Graphically, the model looks like this: Print Phonological representat ion Meaning Verification Figure 5 The Phonological Model with Verification Hypothesis This hypothesis explains homophony effects, such as longer recognition times for words that have homophones, very well. Words without homophones only have to matc h one phonological form to one spelling form; so only one instance of matching must take place, while homophonous pairs must make two (or more) comparisons. The verification hypothesis may also account for some of the lack of speed in poor and beginning readers. Since these readers are also often poor and beginning writers, they do not have the active spelling knowledge to help them verify the correctness of words. Explaining word superiority effects using the verification mechanism is not terribly d ifficult either. Non words whose phonological forms equal real words take longer to process because possible meanings are activated from a phonological representation that is in the lexicon, and the verification routine tries to match the spellings of the activated meanings to the printed form, and it simply cannot do it. This indecision causes longer reaction times or errors. For
18 example, a reader may see , and since this letter string has the same phonological form as word , there may be a delay i n deciding that is in fact not a word. On the other hand, if a non word is not homophonous to a real word, the mind may use time trying to find a word to match with it, as readers expect that letter strings will form words. This is true only whe n a letter string conforms to phonotactic constraints. For example, the letter string may cause hesitation because it â€œlooks likeâ€ a word. In contrast, would likely be rejected out of hand because it is unable to be phonologically processe d, at least in relatively fluent English speakers. The real question is how to account for word frequency effects. I believe that this effect can also be covered by the verification hypothesis. Recall that the frequency effect is exhibited when letter strings such as , a relatively common word, are processed faster than strings such as , an uncommon word. This is because more common words are closer to consciousness, and thus â€œeasier to findâ€, than less common ones. Since phonological pro cessing is automatic and prelexical, and more frequent words cannot magically let their phonological forms be known more quickly, perhaps the frequency effect is a result of the process of verification. It could be the case that made is processed faster t han maid because the verification process attempts to match the phonologically activated /ma j d/ of both and to the printed form of
19 the more common made before it attempts to match maid . 10 Since the verification process stops as soon as a perf ect match is found, it need not check for maid if made has already been identified. As a result, undergoes one cycle of verification, while first tries to match the spelling component of made because made is the more common of the two forms. From this section, we have seen that there is strong evidence for a phonological model of written word recognition that specifically requires prelexical phonological activation. However, the nature of this phonological activation has not been addressed. This nature is the topic of the next section. Nature of Phonological Representation There are two basic possibilities for phonological activation. The first is whole word, where a word form activates a phonological form as a whole (e.g. /f rEnd/ directly, without breaking up the spelling body into subword segments). The other possibility is that sub word units (in English, letters or multi letter graphemes) are processed to give a phonological representation. It is important to take a mome nt to realize that either proposal can account for the material presented so far, with the exception of the regularity effects noted above. While the issue is relevant in several areas beyond psycholinguistics (for example, the whole word versus phonics debate in teaching children how to read English), not much evidence exists on the subject. Many researchers have assumed a phonological representation that was the result of assembling graphemes, but this assumption should be supported. 10 I do not support the idea that this is a seria l process rather than a parallel process. It is possible that the process of â€œfindingâ€ made in the lexicon is faster than maid . The mechanism is the same as the one that helps the mind find team before tram .
20 In a series of experiments using a semantic relatedness task, Lesch and Pollatsek (1998) found some very interesting results supporting the idea that phonological representations are assembled. In their experiments, participants had to decide whether or not a pair of w ords was related. Stimuli were grouped into several categories: appropriate semantic associates (e.g. PILLOW BED), real homophones of associates (LETTER MALE, for LETTER MAIL), visually similar targets (PILLOW BEND), different targets (PILLOW HOOK), and f alse homophones (PILLOW BEAD). The false homophone is labeled as such because one of the possible readings of < ead#> is /Ed/, for example, br ead , h ead and inst ead . Under several sets of conditions, participants in their study made slower decisions regar ding whether or not homophonic and false homophonic stimuli were related, made more errors in those decisions, or both, while the visually similar stimuli did not differ from different stimuli in terms of error rates or decision latency (Lesch and Pollatse k 1998). In other words, the mind had a difficult time deciding whether or not LETTER MALE was a related pair. This observation shows the same phonological effects that we observed before. More interestingly, pairs such as PILLOW BEAD were also confused . The proposed cause of the confusion between stimuli such as PILLOW BEAD is that phonology is activated by the assembling of various graphemes and comparing the results with the printed form: Print Phonology Lexicon Verification /bid/ bead = BEAD BEAD /bEd/ bed BEAD Figure 6 Example of the Progression of Lexical Activation Using the Phonological Model with Verification Hypothesis
21 Given this progression, how did participants in the experiment make mistakes? If verificati on rules out the wrong response, what is the problem? The answer lies in the fact that the verification mechanism is not infallible. When people read, they naturally expect things to make sense together, so prelexical semantic activation may have played some part. If the process is not accomplished by phonological assembly, this should have affected the graphically similar condition as well to some extent, but did not. The problem may lie in the subconscious nature of the verification mechanism in cer tain situations. If readers see BEAD when they expect to see BED, and their phonological processor activates /bEd/ (in addition to /bid/), they may move on, confident that they have read BED. They may move on because they are not consciously aware that / bEd/ can be represented by a spelling body other than . It could also be that they simply have poor active spelling knowledge, but this is often not the case in skilled readers like the ones used in Lesch and Pollatsekâ€™s 1998 study. Another reason that BEAD might be read as BED under some circumstances could be that the letter string must pass some threshold of acceptability, and the phonological component of the string plays a significant part. This is the topic of the next section. Nature of Wor d Recognition in Relation to Reading Up to now, I have argued for a model that accounts for the reading of single words in which their graphic forms are assembled into their phonological counterparts. This phonological form is searched for in the lexicon , and when a
22 match is found, the spelling of the activated lexical entry is then matched with the graphic form that was read. The only problem with this model is that, except in experimental situations, people do not usually read single words. Words fit into contexts, and those contexts are the frames through which we see and interpret written words. This section attempts to explain how the (single) word recognition model argued for in this study fits into reading theory as a whole. Recall the PILLOW BEAD example in the previous section from Lesch and Pollatsek (1998). That BEAD was mistaken for BED is not only a result of the fact that < ead> may have the phonological form / Ed/, it was also because participants saw BEAD beside PILLOW. PILLOW helped to prime /bEd/ because it is semantically related to bed . When people make decisions, they do not arrive at an answer until they are confident enough that their answer is correct. It is common for people to make decisions based on incomplete data beca use complete data are often either impossible to find or would take too much time or effort to find. The latter reasons are common in reading because our purpose in reading is not to ensure that every word is spelled correctly, but rather to access the me aning of the written passage efficiently. In other words, rather than checking and rechecking to make absolutely sure that a written utterance says what people think it says, they accept it if it passes some minimum threshold of sensibility. In reading , this threshold can be reached by a variety of means. Stone and Van Orden (1994) and Perfetti and Tan (1998) laid out how this threshold is reached. Both of these studies suggest that there is something beyond the word
23 and the abilities of readers that contribute to processing in reading. Of course, researchers in education who have argued for top down approaches of reading have known this for decades (see e.g. Smith 1994). However, Stone and Van Orden (1994) and Perfetti and Tan (1998) explain specifi cally how the top down processes relate to lower level processes in cognitively theoretical terms. Both studies argue that visual, phonological, and semantic characteristics all contribute to the decision threshold. An example may help understanding. Imagine a situation where readers see the letter string by itself on a computer screen, and the image disappears. 11 If someone were to ask these readers if the word was in fact hammer , they will follow the phonological model laid out in the previo us sections. Their only means of answering correctly is to follow the model. In other words, readers must compare the visual and phonological forms of the string that appears on the screen to the visual and phonological forms of the words in the lexicon. Now, these imagined readers see a sentence, â€œSally hit the nail with a hammer,â€ on the screen. This time, the rest of the sentence primes the activation of hammer. In order to make a decision about whether the word hammer was on the screen, the amoun t of graphic and phonological information required from the word itself is less than in the single word example. When these imagined people search for hammer in their lexicon, they have more resources to find it if they can read the complete sentence. Be fore they even reach the letter string , 11 This example was given in one of my class es. It is an elaboration of an example given in Smith 1994.
24 readers will know that they are searching for a lexical entry with certain syntactic properties. They will also be aware of the semantic roles that the lexical entry is required to fulfill. In addition, th ey will have the extra textual knowledge to guess what Sally is likely to be hitting nails with. With this in mind, imagine that readers like to be 90 percent sure (this number is totally hypothetical) when making a lexical decision. In the single word example, they must scrutinize the information that is contained within the word itself. When searching their lexical databases, they will only have two criteria: graphic and phonological form. Thus, they will have to make the most of this information. In the sentence example, they will need to rely much less on either of those bits of information. When searching the database this time, they will search with all of the other information shown above, giving the mind a better chance of finding the correct lexical entry even if the graphic and phonological information is incomplete. This time, imagine that the sentence reads, â€œSally hit the nail with a hamster.â€ I read a passage that contained a sentence very similar to this several times, not noticing that was not hammer until it was pointed out to me. Because of the abundant sentential information leading me to the word hammer , the mismatch of graphic and phonological information to the actual word did not make me hesitate. By the time I ha d read the of (as /h/), I was convinced that it was going to be , so I did not process further. My decision threshold for hammer had been reached because the word that made the most sense with the given information was hammer .
25 T he final example was the most important to the discussion at hand. Though it is argued here that the model presented in this chapter is automatic and necessary for reading, it is not the case that the model must run its full course every time it is activa ted or that the full extent of the information gained from the model is used by every reader all of the time. On the contrary, I believe that most of the words that are read do not use the full power of the model, simply because that would not be efficien t. However, that does not mean the model is useless or does not exist. It must be remembered that extra word contextual knowledge is only one part of the decision process. Summary It has been argued that the only model that covers all of the phenomena that have been mentioned throughout the reading literature is the phonological model with the addition of the verification hypothesis. Not only does this model account for the various effects, it also fits nicely with the time course of word activation d iscussed above. How this model that accounts for single word reading meshes with higher level (discursive) reading models has also been discussed. The next question is how this model fits with readers of a second language.
26 CHAPTER 3 INTEGRATION â€“ KOREAN LEARNERS OF ENGLISH AS A SECOND LANGUAGE Introduction How can the verification hypothesis model work in a second language situation? In many cases, learners acquire second languages well after their first language has already been established. Co nsequently, first language phonology is already firmly entrenched in learners by the time they begin to learn second languages (Larsen Freeman and Long 1991). What implications do highly dissimilar phonological systems have for second language readers in a reading model that relies on phonology for part of its processing? Throughout this chapter, we will compare some aspects of Korean and English phonology and see what these differences predict for Korean learners of English as a second language. Korean Consonant Phonemes Perhaps the most obvious difference between Korean and English phonological systems is dissimilarity in the phonemic inventories. Korean has the interesting feature of having a three way distinction in stops that occur at the same poi nt of articulation (Sohn, 1994). Stops can have the features [+lax], [+tense], or [+aspirated] (see Figure 7, below). Note also that there are only a limited number of fricatives in the Korean phonological inventory.
27 Manner Labial Alveolar Palatal Vel ar Glottal Lax p t c k Tense p* t* c* k* Stop Aspirated p h t h c h k h Lax s h Fricative Tense s* Nasal m n ng Liquid r Figure 7 Korean Consonant Phoneme Inventory Korean Versus English Phonology In this section, some of the features of Korean and English phonology will be contrasted. Some things to examine include: consonant clusters, voicing distinctions, coda consonants, fricatives, and liquids. By looking at these features, we will find some very specific areas where the two phonological systems do not match up very well. The first contrast to examine is consonant clusters. Korean does not allow consonant clusters at the surface level, with the exception that there may be two consonants if they are in separate syll ables (Sohn 1999). For example, /ipta/, â€˜to wearâ€™, surfaces as [ipta], while /ancta/, â€˜to sitâ€™, gets the surface form [anta]. An explanation of which consonant undergoes elision is not needed here. It is only necessary to note that the fact that English often allows consonant clusters could be one of the problem areas for Korean learners of English. Another area of difference is voicing. Note that Figure 7 does not list a voicing distinction in Korean. Obviously, this is in contrast with English. Th is would not be so problematic if voicing did not play a role in Korean phonology. In Korean, [+lax] consonants that occur in an intervocalic position or between a voiced consonant (liquids and nasals) and a vowel will assimilate in voicing
28 (Sohn 1994). For example, /kapa ng 1 /, â€˜bagâ€™, becomes [kaba ng ]. Because this process is automatic, this can be another source of interference when learning English. Changes in coda consonants are one of the more interesting aspects of Korean phonology. Consonants that are in coda position become orally unreleased. This means that frication is lost, tenseness is never realized, and aspiration does not come to the surface. The most important feature that remains is place of articulation. So, a word like /kes/, 2 â€˜thingâ€™ , becomes [ket]; /is*ta/, â€˜to be, past tenseâ€™, surfaces as [itta]; and /puek h /, â€˜kitchenâ€™, becomes [puek]. A final problem area to be mentioned in this chapter is the Korean liquid, /r/. Most people easily notice the problems that Korean learners of Eng lish have distinguishing English /l/ and /r/. The problem stems from the fact that /r/ is realized as [l] in syllable final positions and in onsets that follow another /r/. This causes significant difficulties with perception and production in Korean lea rners of English. Taken together, it is easy to see how Korean learners of English can have so many problems with English phonology. The section that follows suggests ways that these differences between Korean and English phonological systems could inte rfere with reading processes in Korean learners of English. 1 I could not convert the IPA symbol for a velar nasal to electronic format, so it is presented here as . 2 As is somewhat common among researchers of Korean, lower case â€˜eâ€™ is used here to represent a back, mid, unrounded vowel
29 Predictions from the Model It should be clear from Figure 7 that both reading and writing may be challenging if learners use phonology to map phonemes to letters. If visual models of reading a re correct, however, there should be no such effects when students write: readers should be able to use visual pathways to call up word spellings. There should be no interference from phonology. However, if the phonological model is correct, then there s hould be phonological effects. That is, sounds that do not map well lead to graphemes that do not map very well either. If Korean learners of English follow the phonological model of reading that includes a verification mechanism that was proposed in C hapter Two, they should have problems in two places due to underdeveloped knowledge of grapheme to phoneme correspondence rules. First, they would have problems building English based phonological forms from letter strings (i.e. print phonological repre sentation). Second, they should have trouble with the verification mechanism (meaning verification). Building the initial phonological forms should be more difficult for learners than native speakers if learners are forced to use L1 phonology because t he â€œirregularâ€ grapheme to phoneme mappings of English seem to be everything from irregular to downright chaotic. The verification process is hindered because there are several possible active phonological forms, and each of those forms maps to several di fferent graphic forms. The brain has a lot of extra work with each candidate that arises. In addition, second language learners have many of the problems that beginning L1
30 readers have, including the mistakes that are caused by imperfect active spelling knowledge. What does all of this mean, practically? It is hypothesized that Korean learners of English as a second language will have problems (i.e. slower readings or errors) when reading English words with certain consonants. I have collected many a necdotal pieces of data from writing samples of my students (e.g. perfect perpect , deliver deliber , friend priend , just to name a few). The verification mechanism is essentially an instantiation of active spelling knowledge, and in my view, this act ive spelling knowledge is the same knowledge that readers use to write words. Because verification and writing are both active procedures, and both come from already activated lexical entries, I posit that they are essentially one and the same. Therefore , I put forward that those Korean learners of English who write PERPECT (instead of PERFECT) would read that same letter string as perfect in much the same way that native English speakers read BEAD as BED. This is because PERPECT and PERFECT are homophon ous when processed via Korean L1 phonology. Taking it a step farther, other Korean learners of English who do not make the mistake of writing PERPECT while writing may still read PERPECT as perfect . 3 This is because writing relies totally on active spell ing knowledge, while the active spelling knowledge that is inherent in the verification mechanism is less important. A word like PERPECT will be a false homophone for any Korean, much like the ones described in Lesch and Pollatsek (1998) in section 2.5 ab ove. Because 3 This is not to say that Korean learners of English often read material with these types of errors. The point here is to illustrate the process and show the mechanism, which relies largely on phonology.
31 there are so many possibilities for mapping from (English) print to (Korean) phonology, there should be many more false homophones. This makes reading a much more difficult task for Korean learners of English. Conclusion We have reviewed t he evidence for a phonological model of word recognition and discussed some of the potential implications for one set of second language readers. The chapters that follow describe an experiment that was designed to test the model laid out in Chapter Two. Essentially, this experiment follows the models of first language psycholinguistic experiments, though the participants are Korean learners of English instead of native English speakers. The false homophones will be of the type discussed above in this ch apter, where Korean and English phonology map poorly to one another.
32 CHAPTER 4 EXPERIMENTAL METHODOLOGY Goals and Variables The major goal of the experiment was to show how phonology interacts with reading. Several tasks were designed to show thi s interaction. Participantsâ€™ phonological knowledge of English had to be determined, both at an overall and phoneme specific level. In addition, a measure of participantsâ€™ ability to distinguish between words and nonwords in a variety of conditions was n eeded. When the study was set up, the tasks were designed to fit into a statistical scheme that would allow relationships phonology and reading to be discovered. The main variables that were considered were participantsâ€™ phonological knowledge of Englis h (using the phonemic distinction task, described below) and ability to recognize written English words, both in terms of speed and accuracy (using the lexical decision task, described below). In addition, participantsâ€™ estimation of word frequency (as de termined by the subjective word frequency task) was measured, though it could not be used in the primary analysis for reasons that will be described below. Phonological knowledge and word frequency were independent variables, and word recognition ability was the dependent variable.
33 Participants Participants in this study were adult natives of South Korea currently living in the Gainesville community. 32 out of 35 participants (16 female, 16 male) gave usable data for this study. Three others had some p roblems with their data or backgrounds that precluded analysis of the data. The data for one of these participants was unusable because her baby was not cooperating with our need for silence during the experiment. Since it was crucial for the participant s to have total concentration on the tasks in the experiment, I removed her data from the analysis. Another participant spent nearly half of her (pre critical period) life in Venezuela, so her data mimicked what I would expect of a native Spanish speaker rather than a native Korean speaker and thus was not suitable for this analysis. The final participant who was removed from the study admitted to me that she was a speech pathology major and explained that she was not linguistically nave, given extensive training in transcription. This last participant was the only person tested who came reasonably close to guessing the exact nature of the experiment. The ages of the participants varied from 18 to 37, with a median age of 30. All participants were hig hly educated, having obtained at least a bachelorâ€™s degree in either the United States or Korea. Most (28 out of 32) were attending or had graduated from the University of Florida or Santa Fe Community College at the time of the study. All participants r eported having no difficulties with their hearing. All participants had been staying in the United States for at least 6 months. Their length of stay ranged from 6 months to 9 years, with a median
34 stay of 20 months. All participants had at least the 6 y ears of English that are required for graduation, but the amount of additional outside English education varied widely, from no outside education to 5 years, with a median of 8 months. Participants were recruited through my contacts in Gainesvilleâ€™s Kore an community, including acquaintances, former students, and my contacts at the Korean Baptist Church and the English Language Institute. Participants included mostly graduate students at the University of Florida, but also included several spouses, former students, undergraduates at both UF and Santa Fe Community College, and students in the highest levels of the University of Floridaâ€™s English Language Institute. Because of the nature of the participants (very highly educated, living in the United States , etc.), any generalizations that are made must be made with this in mind. Procedure Informed Consent At the start of the experiment, participants read and signed the release form that is required by the University of Florida Institutional Review Board (see Appendix A). I explained each item to the participants, including the tasks that they would be asked to do, and also assigned a code number to each participant at this time. Phonemic Distinction Task The phonemic distinction task was based on the M odified Rhyme Test described in Hodgson (1980). The stimuli were single syllable words, and
35 answer choices for a single stimulus varied by either word initial sound changes or word final, but not both. Figure 8 contains examples of each type. bust just rust must lust dust bat bad back ball bar bath Figure 8 Phonemic Distinction Task Examples In the task, participants were seated in front of a laptop computer, equipped with headphones and a mouse. Then, as the task started, they listened to a word fr om the stimulus list (see Appendix B for the list of stimuli, organized by target phoneme). After the sound file was played, a list of possible answers appeared, and participants clicked on the word that they thought they heard. The first example in Figu re 8 shows the answer choices that would appear for the words rust and lust (in the experimental condition), as well as must (in the control condition). The task took approximately 10 12 minutes to for each participant to complete, including instructions and practice. There were 10 practice trials, which were not included in the analysis, and 111 experimental trials. The stimulus list was recorded by a male native speaker 1 in a quiet room using digital audio tape for the highest possible sound quality. The list was played to participants through high quality headphones in a quiet room in order to diminish any background noise and to ensure the best possible sound perception. Background Questionnaire After completing the phonemic distinction task, the participants filled out the background questionnaire. This was a paper and pencil sheet (see Appendix 1 The lis t was not recorded by me because several of the participants were familiar with my voice.
36 C), which took approximately five minutes to finish. It asked basic biographical questions, mostly relating to their educational background in English and their time spent in the United States. Lexical Decision Task Next, participants performed a lexical decision task. In this task, participants decided whether or not a letter string presented on a computer screen was a correctly spelled English wor d. First, participants saw a string (XXXXXXXXX) to direct their attention to the focal point and signal the start of a trial. This string, which also acted as the inter trial interval (thus destroying residual images in the sensory store), was displayed on the monitor for 1500 ms. Then, participants saw a stimulus and, using the mouse, clicked a button that said â€œYes, Real Wordâ€ for a yes response (i.e. the printed string is an English word), or a button that said â€œNo, Not a Real Wordâ€ for no . Both the time taken to respond (in milliseconds) and the response itself (yes or no) was recorded. Participants were given 4 seconds a select an answer. If they did not choose in this time, the program proceeded to the next trial and was counted as an error. B ecause of the limitations of the software that was used for the experiment, the response times had to be measured by clicking a mouse button. One disadvantage of this is that the response time necessarily included that time that was needed to move the mou se pointer over the desired response. To reduce the variability of this time as much as possible, participants were instructed to move the mouse pointer to the vertical line that separated the â€œyesâ€ and â€œnoâ€ response buttons between trials. This ensured that prior responses
37 would not influence the response times of the responses that followed. For example, if both number 5 and 6 required the same response, and the participant did not move the mouse pointer to the center line, the response time would be f aster than if the trials required different responses. This is because there is no movement time required in the former case, while the mouse pointer must be moved in the latter one. If participants had to move the mouse pointer the same distance in ever y case (i.e. from the center line), this movement time should have been as consistent as possible. While I could not watch every trial for compliance with this request, it was explained thoroughly before and during the practice trials, and those trials th at I did watch were not problematic in this regard. The stimuli were taken from the Lexical Decision Word List (Appendix D). The list is organized by the phonemes in words that may be confused by Korean learners of English, including /f/, /p/, /g/, /k/, /d/, /t/, /l/, and /r/. Also included is a list of control words, chosen because they do not contain the phonemes that should cause confusion in Korean learners of English. In addition, each word is listed with its psuedohomophonic and visually similar counterparts. None of the words from the phonemic distinction task was included in the lexical decision task. Each target word contains one of the phonemes of interest. The stimuli related to that word are either homophonous to the target when processed via Korean phonology (i.e. pseudohomophones) or visually similar but not homophonous to the target word. confess conpess contess Figure 9 Normal, Pseudohomophonous, and Visually Similar Variations of c onfess
38 Figure 9 is an example from the /f/ stimulu s list in Appendix D. The first entry is the target, the second is the pseudohomophone, and the third is the visually similar, non homophonous letter string. The pseudohomophones were determined as laid out in the Chapter Three. Every participant saw al l three conditions. This made it possible to use repeated measures tests. The visually similar stimuli were designed by changing the suspect grapheme to one that is visually similar, yet still pronounceable. An attempt was made to keep certain characte ristics in the visually similar condition, such as presence of ascenders (h, d, b, etc.) or descenders (g, q, p, etc.) However, this was not possible in all cases. The stimuli were randomly presented to the participants. The total time for this task was about 10 12 minutes, including practice and rest. Subjective Word Frequency Judgment Task Finally, each participant performed the subjective word frequency judgment task. The list of stimuli included in this task is identical to the normal condition wo rd list for the lexical decision task that appears in Appendix D. In this task, participants rated the targets from the lexical decision task by how often they believed they encountered the words everyday, combining all forms of encounter: listening, spea king, reading and writing. In addition, participants were able to select a â€œdonâ€™t knowâ€ answer choice, indicating unknown words. In the task, a word appeared on the computer screen and participants selected a frequency rating from 1 (very rare) to 6 (ver y common), or the â€œDonâ€™t Knowâ€ response.
39 Originally, there was going to be a full fledged vocabulary test of the lexical decision stimuli, but this would have added a minimum of 40 minutes to the testing. As it is constructed, this task did not take mo re than 3 5 minutes. Hypotheses When trying to determine the role of first language phonology in reading, the phonological reading model with the verification mechanism essentially predicted two outcomes. First, pseudohomophones would have slower react ion times and/or lower accuracy scores than normal words on the lexical decision task. Second, the model predicted that phonemic distinction scores would show a positive relationship with lexical decision scores by phoneme. In other words, if words conta ining phoneme A have a high phonemic distinction accuracy score in one person, other words with phoneme A will have a high accuracy score on the lexical decision task. The section that follows shows that these hypotheses were both supported.
40 CHAPT ER 5 RESULTS AND DISCUSSION Treatment of Data Before explaining the statistical analyses, it is necessary to describe some of the data management that took place in order to obtain the most valid statistical results. In addition, each instance includes an explanation about why the step was taken. These steps included removing /g/ and /k/ from the experiment, removing the subjective word frequency judgment task, and setting up two ANOVAs for both accuracy scores and reaction time data in order to get a m ore complete picture of all of the relationships that exist between the factors. Removing /g/ and /k/ From the Experiment Although data was collected for /g/ and /k/, this data was not used in any statistical analyses. The reason for this is that there were very few trials involving /g/ and /k/ in the experimental tasks, for reasons of economy. Across the board, people who performed the initial versions of the experiment 1 responded that the tasks took too long, making participant fatigue a factor. The refore, I sacrificed many of the trials regarding /g/ and /k/ in all tasks in the final experiment. I chose to remove /g/ and /k/ trials for two reasons. First, my anecdotal data did not include much evidence of confusion between /g/ and /k/, consequentl y I felt it was not necessary to burden participants with trials that 1 The experimental tasks were evaluated by two graduate students in linguistics (one from Korea, one from the United States), who gave valuable feedback.
41 would not be very useful to my analysis. In addition, the difference between /g/ and /k/ is a voicing distinction, so /g/ and /k/ should pattern somewhat like /d/ and /t/, which were in cluded in the final experiment. A few trials of both /g/ and /k/ remained in the experiment, in order to see if they had a tendency to pattern like any of the other phonemes. However, even this casual analysis was not possible, as one (of only three) wo rds using /k/ in the lexical decision task displayed highly unusual characteristics. For the pseudohomophone , only 28 percent of respondents gave a correct â€œnoâ€ response. In the rest of the experiment, the trial with the next lowest mean accura cy score was , with 63 percent responding correctly. Since the number of data points involving /g/ and /k/ was small to begin with, removing and its related trials would have made it less likely to form any reliable judgments about /g/ a nd /k/. Removing Subjective Word Frequency Data from the Analysis I did not include an analysis of the word frequency data that was obtained in the experiment for two reasons. First, the subjective nature of the task proved to be relatively unreliable, as people cannot judge with great precision how often they encounter a word. Therefore, from the outset I knew that any analysis using the data would have to take this into account by using nonparametric tests to analyze the data. Another factor that co ntributed to the exclusion of the subjective word frequency data, and an unexpected occurrence, was the fact that at least 20 of the participants (out of 32) chose the highest rating for word
42 frequency for over half of the words, thereby relieving the task of virtually all of its predictive utility for the majority of the participants. Experimental Controls A short discussion regarding the control words that were used in the experiment is also in order. Recall from the Procedure section of the Methodolo gy chapter that controls in the lexical decision task were words that did not contain any sounds that should have caused difficulty (in terms of phonological transfer) for the participant. Since there were no ambiguous phonemes in these words, there could be no pseudohomophonous condition in the lexical decision task for these words. As this was the case, two analyses of each data set had to be conducted: one with the controls, which did not include the pseudohomophones, and another without the controls, which included the pseudohomophones. Descriptive Statistics Reaction Time Before discussing the results of each set of statistical analyses, it will be helpful to examine some of the descriptive statistics. Table 2 shows the data for reaction time from th e lexical decision task in tabular form, while Figure 10 shows the mean response time of each stimulus condition by phoneme in graphic form. Table 2 Lexical Decision Mean Reaction Time (in seconds) by Phoneme and Condition /f/ /p/ /d/ /t/ /r/ /l/ Contro l Normal 1.26 1.33 1.13 1.24 1.44 1.24 1.19 Pseudohomophone 1.39 1.45 1.28 1.40 1.42 1.38 n/a Visually Similar 1.48 1.33 1.30 1.31 1.30 1.42 1.43
43 This data suggests that there is a relationship between phonology and response speed: the normal conditio n had an average response time of about 140 ms faster than the pseudohomophonic condition across phonemes. The fact that the line representing pseudohomophones is essentially the same as the line representing the normal condition shifted up about 140 ms i s expected by the verification hypothesis, which predicted that readers would mull over homophonous graphic forms longer in order to verify their correctness. There is a glaring exception with /r/, which has a difference of only 20 ms between the two cond itions, but the rest of the data is consistent. 0.95 1.05 1.15 1.25 1.35 1.45 1.55 /f/ /p/ /d/ /t/ /r/ /l/ Control Phoneme Time (seconds) Normal Pseudohomophone Visually Similar Figure 10 Mean Reaction Times for the Lexical Decision Task by Phoneme and Stimulus Condition The visually similar condition shows no such relationship with the phoneme. The di fference in reaction time between the normal condition and the visually similar condition ranged from 0 270 ms per phoneme across tokens. This was the expected result, as readersâ€™ processing could take one of several
44 paths. In the first, they could proce ss the visually similar nonword and reject it immediately because they would quickly notice that the phonological form did not match anything in their lexicon. In the second, readers may think about the phonological form for a moment, searching for an unc ommon word, thus causing longer decision latencies. Additionally, and perhaps most importantly, the visually similar condition no longer contained the phoneme in question. Table 3 Lexical Decision and Phonemic Distinction Mean Percent Correct by Phonem e and Condition f p d t r l Control Normal 0.96 0.92 0.99 0.98 0.86 0.96 0.99 Pseudohomophone 0.93 0.91 0.92 0.93 0.92 0.93 n/a Visually Similar 0.94 0.94 0.96 0.94 0.94 0.95 0.90 Phonemic Distinction 0.93 0.89 0.99 0.99 0.91 0.82 0.98 Accuracy Scor es In Figure 12 and Table 3, the mean percentages of correct answers for the lexical decision task are shown by phoneme and by condition. The accuracy score for the phonemic distinction task by phoneme is also noted for reference. This latter task was no t included in the reaction time analyses because participants were not being timed during the completion of this task. In the lexical decision task, phonology played a role in a different way. The accuracy scores of the pseudohomophonic and visually simi lar conditions remain relatively steady across the various phonemes, but the lexical decision accuracy scores follow the phonemic distinction scores by phoneme to a large extent. Again, there is an exception (this time /l/), but there is still strong evid ence of a relationship between phonemic distinction ability and the ability to choose whether or not a printed letter string is a real word.
45 Overview of Statistical Analyses The observations regarding the relationship between phonology and lexical decis ion variables are borne out by statistical analyses. Two repeated measures ANOVAs were run for both accuracy scores and reaction times on the lexical decision data. Factors included phoneme and stimulus condition. In addition, correlations were run in o rder to compare participantsâ€™ phonemic distinction ability with their accuracy scores in the lexical decision task. 0.8 0.85 0.9 0.95 1 /f/ /p/ /d/ /t/ /r/ /l/ Control Phoneme Percent Correct Normal Pseudohomophone Visually Similar Phonemic Distinction Score Figure 11 Mean Decision Percentages by Phoneme and Stimulus Condition Lexical Decision Task Accuracy Scores F irst, to measure the effect of phoneme and stimulus condition on the percentage of correct answers on the lexical decision task, I ran two main tests: a
46 6x3 repeated measures ANOVA and a 7x2 repeated measures ANOVA. The first factor was phoneme, using /f/, /p/, /d/, /t/, /l/, and /r/ (and Control in the second ANOVA) as the different levels. The second factor was condition, using normal, pseudohomophonic, and visually similar conditions (leaving out pseudohomophones in the second analysis). 2 The statisti cs are summarized in Table 4. Tab le 4 ANOVA values for percent correct in lexical decision task by participant Analysis One (6x3) Analysis Two (7x2) F (5,27) = 3.631 F (6,26) =7.64 Phoneme p<.015 p<.001 F (2,30) =4.22 F (1,31) =1.102 Condition p<.025 p=. 302 F (10,22) =3.092 F (6,26) =4.402 Phoneme x Condition p<.015 p<.005 For the first analysis, 3 the main effects of phoneme and condition were significant ( p < .015 and <.025, respectively). In addition, their interaction was significant ( p < .015). The second analysis included the control words, and thus the pseudohomophonic condition had to be removed. In the second analysis, the main effect of phoneme was significant ( p < .001), though the main effect of condition was not ( p > .05). However, the int eraction of the two factors was significant ( p < .005). 2 I had to run the analyses separately because there could not be a pseudohomophonic condition for controls (recall that this is why they were controls), and the repeated measures ANOVA cannot be run if all levels of one factor do not contain all levels of the oth er factor. 3 For both ANOVAS, the statistical program that was used mentioned that Mauchlyâ€™s Test of Sphericity was significant, so I used the â€˜Statistics for Multivariate Testsâ€™ section listed in the results, as suggested by Brace et al. (2000). All r eported p values are based on the Pillaiâ€™s Trace statistic.
47 These results show that, for Korean learners of English, there is a difference in how readers are able to make decisions about whether or not a printed string is a word, and this difference depends partly on whether or not the printed string â€œsounds likeâ€ its real word counterpart through L1 phonology. Additionally, some letters (or phonemes) cause more problems than others. However, exactly where these distinctions lie is difficult to see. This is because the statistical program that was used only allows post hoc tests for between subject variables, and since this was a repeated measures design, phoneme and condition were both within subject variables. Despite the limitations of the design, the s tatistical analysis shows that lexical decision accuracy scores are affected by phonology in at least three distinct ways. First, words containing certain phonemes are more difficult to process for readers whose native language phonological system does no t match that of English. Second, homophony matters; nonword letter strings that are homophonous to real words fare worse than visually similar nonword letter strings. Finally, the interaction of the two effects mattered. The effect of condition was less apparent across phonemes for the pseudohomophones and visually similar than for the normal condition. Reaction Time In addition to the data analysis of accuracy scores in the lexical decision task, an ANOVA was performed on the response time data. It duplicated the analyses of accuracy data, both in terms of format (6x3 and 7x2) and in results. Table 5 shows a summary of the statistics. The main effects of phoneme and
4 8 condition were both significant. In addition, the interaction of phoneme and condi tion was significant. The two sets of analyses clearly show that both phoneme and condition have significant effects on the ability of Korean speakers of English to choose whether or not a letter string is a real English word. These two main effects we re expected based on the predictions and hypotheses above. Table 5 ANOVA values for time in lexical decision task by participant Analysis One (6x3) Analysis Two (7x2) F (5,27) = 16.072 F (6,26) =8.303 Phoneme p<.001 p<.001 F (2,30) =10.588 F (1 ,31) =7.864 Condition p<.001 p<.01 F (10,22) =6.912 F (6,26) =15.198 Phoneme x Condition p<.001 p<.001 For varying phonemes, it was predicted that words with letters that correspond to more difficult phonemes (i.e. phonemes that are less accessible through L1 phon ology) would fare worse than words with easier phonemes. For the main effect of condition, it was predicted that pseudohomophones would cause more errors and possibly longer reaction times than the normal condition. In addition, the interaction of the t wo factors was significant. This indicates that different conditions affected words containing different phonemes differentially. This was also predicted from the model. Phonemes that were difficult for people to distinguish aurally should show a compou nded effect when combined with the pseudohomophonic condition, while easier phonemes should show a much smaller difference between the normal and pseudohomophonic conditions.
49 For the reaction time data in the lexical decision task, a Pearsonâ€™s correlatio n was run between the normal and pseudohomophone conditions by phoneme (Table 6 contains the details). It was predicted that these variables would be related because the model predicts that processing speed is related to phonology. Since both psuedohomop honic and normal trials represented the same phonological form, they should be related in some way. The results show that normal and pseudohomophonous conditions had highly correlated reactions times by phoneme. Table 6 Pearsonâ€™s Correlation values for Reaction Times Between Normal and Pseudohomophonous Conditions in the Lexical Decision Task Phoneme Correlation Coefficient P value /f/ .730 <.001 /p/ .695 <.001 /d/ .570 <.025 /t/ .789 <.001 /r/ .721 <.001 /l/ .691 <.001 Analysis By Word In additi on to the analyses conducted by subject, these effects were also investigated by treating each word individually. This was done in order to get some of the post hoc tests that were important to finding specifically where the differences were in the data. 4 4 However, by organizing the data in this way, the statistical tests lose the ability to keep within subject variance separate from between subject variance, thus losing the advantage of the gr eater power of a repeated measures design.
50 Table 7 ANOVA values for time and percent in lexical decision task by word Accuracy Time F (6,180) = 1.175 F (6,180) = 1.592 Phoneme p=.107 p=.171 F (2,180) =4.188 F (2,180) =8.047 Condition p<.02 p<.001 F (11,180) =2.235 F (11,180) = 1.625 Phoneme x Condition p<.02 p=.095 Table 7 shows that the main effect of condition was significant for both accuracy and time, while phoneme was not significant for either factor. This was not counter to my expectations, and it may be explained. There are still cer tain inter word effects that have not yet been accounted for. For example, in the lexical decision task, 97 percent of the participants answered correctly for people . However, only 87 percent of the participants correctly chose a â€œyesâ€ response for supre me . This difference could be due to word frequency effects; a possibility that I attempted to consider with the subjective word frequency task. Despite the problems that eventually led to disqualifying this test from statistical analysis, individual word s may show the effect of word frequency on reaction time and accuracy scores. For example, people and supreme did fall at opposite ends of the range in the data, with an average ranking of 1.16 for people (one of the highest ranked words) and 2.66 for sup reme (one of the lowest ranked words; again, see beginning of chapter) on a scale of 1 6, with 1 = â€œvery commonâ€, 6 = â€œvery rareâ€. The interaction of the two factors was significant, and the importance of this is the same as in the previous section.
51 Phone mic Distinction Task In addition to these tests, several correlations were performed in an attempt to support the hypothesis that participantsâ€™ phonemic distinction abilities correlated with their lexical decision scores. Using Pearsonâ€™s correlation, I f ound that the accuracy scores on the normal condition in the lexical decision task correlated significantly (at a moderate level) with a personâ€™s overall phonemic distinction score (correlation coefficient = .394, p < .05). In other words, participants who could distinguish phonemes better aurally fared better in choosing the correct answer for normal words in the lexical decision task. It was predicted that this would also be the case with the accuracy scores in the pseudohomophonic condition, but this d id not turn out as expected (correlation coefficient = .167, p =.36). This lack of correlation will be addressed in the general discussion. General Discussion and Conclusion Discussion While not all of my hypotheses were supported to the extent predic ted, the general outcome of the experiment was very favorable to my predictions. Essentially, the theoretical analysis and the results of the experiment both support the notion that phonology affects reading. To summarize the experimental findings: accur acy rates for the pseudohomophone condition of the lexical decision task were worse than the accuracy rates for the visually similar condition; reaction times and accuracy rates varied by phoneme within the normal condition of the lexical decision task; re action time varied by phoneme
52 within the pseudohomophone condition; lexical decision accuracy of the normal condition was correlated with phonemic distinction score by phoneme; and reaction time of the pseudohomophonous and normal conditions were correlate d by phoneme in the lexical decision task. What do these results say about the process of word recognition in Korean learners of English? When the findings are combined, an interesting picture emerges. Real words are affected by different phonemes, bo th in terms of time and accuracy. The same phonemes that cause problems in listening cause problems in word recognition. Those problems appear in the form of slower decisions and higher numbers of errors for words that contain phonemes that are more diff icult to distinguish aurally. For nonwords, the situation is slightly different. Accuracy scores for pseudohomophonous nonwords are lower than those of visually similar nonwords, which are in turn lower than scores for normal words. However, the accur acy scores for nonwords do not vary noticeably by phoneme. If the scores are not different across phonemes for pseudohomophones, how can I argue that phonology affects the word recognition process? The answer lies in the verification hypothesis and in th e reaction times and decision threshold level. For normal words, readers reached their optimum efficiency levels, finding a balance between time spent on the task and accuracy levels, depending on their ability to verify the phoneme in the word. Pseudo homophones, on the other hand, reached an upper bound in terms of accuracy scores. Readers allowed their minds to process a string for up to
53 140 ms longer by phoneme (see Table 2) than usually needed under normal circumstances to make a decision, and then went with their best guess. This best guess was correct about 92 percent of the time. This follows the verification hypothesis in that it took longer to process pseudohomophonous trials than normal trials, which is not predicted by the unmodified phonol ogical model. The data available for pseudohomophones and normal words suggest that word recognition deals with difficulties in phonology by setting efficiency guidelines for making decisions. These guidelines balance the desire to make correct decisio ns with the need to make the decision quickly. Limitations Several areas of this study could be improved. The lexical decision task was probably too easy for the people who participated in it. A lot of the data was not normally distributed (hence the use of Pillaiâ€™s Trace statistic in the corrected ANOVAs) because of threshold effects. Nevertheless, the results were still highly significant. However, to obtain data that does not violate any of the assumptions of ANOVA, the lexical decision task could have been made to allow the stimuli to remain on the screen for a certain amount of time, and then mask them. In this case, participants would have had to rely on less visual data, thus forcing riskier decisions. As mentioned before, the subjective fr equency judgment task did not turn out as planned. Perhaps more explicit instructions or a scale that specified frequency exactly (once each day, once each week, etc.) would have prevented the scores from clustering around the highest frequency rating. G ood (i.e.
54 usable) data from the subjective word frequency task would hopefully lend stronger support for the verification hypothesis, as this variance was not able to be accounted for. One solution to both of the previous problems might have been to use more difficult words in the task. The problem with this is that participants might simply not know the more difficult words. The point is not to force more errors; rather, it is to force more difficult decisions of words that are known to the participan ts.
55 APPENDIX A INFORMED CONSENT Informed Consent Protocol Title : The effects of L1 phonology on L2 written word recognition Please read this consent form carefully before you decide to participate in this study. If you have any questions, fee l free to ask the experimenter at any time. Purpose of the research study: The purpose of this study is to examine Korean peopleâ€™s reading of English. What you will be asked to do in the study: First, you will be asked to listen to a list of words in English and circle the word that you hear. Next, you will be asked to decide whether words that you see on a computer screen are English words or not. Then, you will be asked to judge how commonly you encounter some English words. Finally, you will be a sked to fill out a short questionnaire about your background. Time required: 1.5 2 hours. Risks and Benefits: We do not anticipate any risks or direct benefits for participating in this study. Compensation: None. Confidentiality: Your identity will b e kept confidential to the extent provided by law. Your information will be assigned a code number. The list connecting your name to this number will be kept in a password protected, encrypted file on a computer disk that will be kept in a locked box at my house. When the study is completed and the data has been analyzed, the list will be destroyed. Your name will not be used in any report. Voluntary participation: Your participation in this study is completely voluntary. There is no penalty for not participating. Right to withdraw from the study: You have the right to withdraw from this study at any time without consequence. Whom to contact if you have questions about the study: Investigator : Dan Hufnagle, Graduate Student, Program in Linguistic s, 4122 Turlington Hall, 392 0639 x236, email@example.com Faculty Advisor : Theresa Antes, PhD, Assistant Professor, Department of Romance Languages and Literature, 212 Dauer Hall, 392 2016 x236, firstname.lastname@example.org Whom to contact about your rights as a resear ch participant in the study: UFIRB Office Box 112250 University of Florida
56 Gainesville, FL 32611 2250 Phone: 392 0433 Agreement: I have read the procedure described above. I voluntarily agree to participate in the procedure and I have received a copy of this description. Participant: ________________________________ Date: _______________ Principal Investigator: _______________________ Date: _______________ Dan Hufnagle
57 APPENDIX B PHONEMIC DISTINCTION WORD LIST WITH POSSIBLE ANSWER CHOICES /b/ n=3 bin 1 pin 2 sin 3 tin 4 bin 5 din 6 fin bit 1 kit 2 bit 3 fit 4 sit 5 pit 6 hit bang 1 rang 2 fang 3 gang 4 bang 5 sang 6 pang /f/ n=12 feel 1 peel 2 reel 3 feel 4 heel 5 keel 6 eel fin 1 pin 2 sin 3 tin 4 win 5 din 6 fin feat 1 meat 2 feat 3 heat 4 seat 5 beat 6 peat fig 1 dig 2 wig 3 big 4 rig 5 pig 6 fig fill 1 will 2 hill 3 kill 4 pill 5 fill 6 bill fit 1 pit 2 bit 3 fit 4 sit 5 wit 6 hit fang 1 rang 2 fang 3 gang 4 bang 5 sang 6 pang fun 1 sun 2 pun 3 gun 4 fun 5 bun 6 run fail 1 hail 2 mail 3 tail 4 bail 5 fail 6 pail cuff 1 cut 2 cub 3 cuff 4 cup 5 cud 6 cuss leaf 1 leaf 2 leap 3 lead 4 leak 5 lean 6 leash puff 1 pun 2 puff 3 pup 4 puck 5 pus 6 pub /p/ n=14 pill 1 will 2 hill 3 kill 4 pill 5 fill 6 bill pit 1 pit 2 bit 3 f it 4 sit 5 wit 6 hit pun 1 sun 2 pun 3 gun 4 fun 5 bun 6 run pin 1 pin 2 sin 3 tin 4 win 5 din 6 fin pail 1 hail 2 mail 3 tail 4 bail 5 fail 6 pail pit 1 pit 2 bit 3 fit 4 sit 5 wit 6 hit pang 1 rang 2 fang 3 gag 4 bang 5 sang 6 pang
58 peel 1 peel 2 re el 3 feel 4 heel 5 keel 6 eel pin 1 pin 2 sin 3 tin 4 win 5 din 6 fin peat 1 meat 2 feat 3 heat 4 seat 5 beat 6 peat pig 1 dig 2 wig 3 big 4 rig 5 pig 6 fig leap 1 leaf 2 leap 3 lead 4 leak 5 lean 6 leash pup 1 pun 2 puff 3 pup 4 puck 5 pus 6 pub cup 1 cut 2 cub 3 cuff 4 cup 5 cud 6 cuss /d/ n=16 dip 1 sip 2 rip 3 tip 4 dip 5 hip 6 lip dill 1 dill 2 hill 3 kill 4 till 5 fill 6 bill din 1 pin 2 sin 3 tin 4 win 5 din 6 fin dot 1 hot 2 got 3 not 4 pot 5 dot 6 tot dale 1 gale 2 male 3 tale 4 bale 5 sale 6 dale dame 1 fame 2 dame 3 came 4 name 5 tame 6 game den 1 ten 2 pen 3 den 4 hen 5 then 6 men dent 1 tent 2 bent 3 went 4 dent 5 rent 6 sent kid 1 kick 2 king 3 kid 4 kit 5 kin 6 kill sad 1 sad 2 sass 3 sag 4 sack 5 sap 6 sat bad 1 bat 2 bad 3 back 4 ball 5 ban 6 bath pad 1 pass 2 pat 3 pack 4 pad 5 path 6 pan cud 1 cut 2 cub 3 cuff 4 cup 5 cud 6 cuss bud 1 bun 2 bus 3 but 4 buff 5 buck 6 bud bead 1 bean 2 beach 3 beat 4 beam 5 bead 6 beak mad 1 map 2 mat 3 math 4 man 5 mass 6 mad /t/ n =16 ten 1 ten 2 pen 3 den 4 hen 5 then 6 men tent 1 tent 2 bent 3 went 4 dent 5 rent 6 sent tip 1 sip 2 rip 3 tip 4 dip 5 hip 6 lip till 1 dill 2 hill 3 kill 4 till 5 fill 6 bill tin 1 pin 2 sin 3 tin 4 win 5 din 6 fin tot 1 tot 2 Todd 3 top 4 tall 5 tar 6 Tom tale 1 gale 2 male 3 tale 4 bale 5 sale 6 dale tame 1 fame 2 dame 3 came 4 name 5 tame 6 game but 1 bun 2 bus 3 but 4 buff 5 buck 6 bud
59 beat 1 bean 2 beach 3 beat 4 beam 5 bead 6 beak kit 1 kick 2 king 3 kid 4 kit 5 kin 6 kill tot 1 tot 2 T odd 3 top 4 tall 5 tar 6 Tom sat 1 sad 2 sass 3 sag 4 sack 5 sap 6 sat bat 1 bat 2 bad 3 back 4 ball 5 ban 6 bath pat 1 pass 2 pat 3 pack 4 pad 5 path 6 pan cut 1 cut 2 cub 3 cuff 4 cup 5 cud 6 cuss /g/ n=3 gale 1 gale 2 male 3 tale 4 dale 5 kale 6 fail bag 1 bat 2 bad 3 back 4 ball 5 bag 6 bath pig 1 pill 2 pick 3 pip 4 pig 5 pin 6 pit /k/ n=6 cold 1 sold 2 told 3 hold 4 fold 5 gold 6 cold cot 1 hot 2 got 3 cot 4 pot 5 lot 6 tot came 1 fame 2 same 3 came 4 name 5 tame 6 game kill 1 will 2 hi ll 3 kill 4 gill 5 fill 6 bill lack 1 lass 2 lab 3 lack 4 lap 5 lag 6 lad duck 1 duck 2 dud 3 dung 4 dub 5 dug 6 done /r/ n=16 rob 1 lob 2 rob 3 bob 4 cob 5 gob 6 sob rung 1 lung 2 rung 3 sung 4 hung 5 young 6 dung rent 1 tent 2 bent 3 lent 4 dent 5 rent 6 sent rot 1 hot 2 got 3 not 4 rot 5 lot 6 tot rip 1 sip 2 rip 3 tip 4 dip 5 hip 6 lip Rick 1 kick 2 lick 3 Rick 4 pick 5 wick 6 tick red 1 led 2 shed 3 red 4 bed 5 fed 6 wed raw 1 raw 2 paw 3 law 4 jaw 5 thaw 6 saw rook 1 book 2 took 3 shook 4 cook 5 rook 6 look rust 1 bust 2 just 3 rust 4 must 5 lust 6 dust rest 1 nest 2 vest 3 lest 4 test 5 best 6 rest ray 1 way 2 may 3 ray 4 lay 5 day 6 pay
60 tier 1 teach 2 tear 3 tease 4 teal 5 team 6 teak pair 1 page 2 pane 3 pace 4 pay 5 pale 6 pare h ear 1 hear 2 health 3 heal 4 heave 5 heat 6 heap bar 1 bat 2 bad 3 back 4 ball 5 ban 6 bar /l/ n=16 lust 1 bust 2 just 3 rust 4 must 5 lust 6 dust lest 1 nest 2 vest 3 lest 4 test 5 best 6 rest lay 1 way 2 may 3 ray 4 lay 5 day 6 pay lob 1 lob 2 rob 3 bob 4 cob 5 gob 6 sob lung 1 lung 2 rung 3 sung 4 hung 5 young 6 dung lent 1 tent 2 bent 3 lent 4 dent 5 rent 6 sent lot 1 hot 2 got 3 not 4 rot 5 lot 6 tot lip 1 sip 2 rip 3 tip 4 dip 5 hip 6 lip lick 1 kick 2 lick 3 Rick 4 pick 5 wick 6 tick led 1 led 2 shed 3 red 4 bed 5 fed 6 wed law 1 raw 2 paw 3 law 4 jaw 5 thaw 6 saw look 1 book 2 took 3 shook 4 cook 5 rook 6 look heal 1 hear 2 health 3 heal 4 heave 5 heat 6 heap ball 1 bat 2 bad 3 back 4 ball 5 ban 6 bar teal 1 teach 2 tear 3 tease 4 t eal 5 team 6 teak pail 1 page 2 pane 3 pace 4 pay 5 pale 6 pare Control n=9 cuss 1 cut 2 cub 3 cuff 4 cup 5 cud 6 cuss gun 1 sun 2 pun 3 gun 4 fun 5 bun 6 run male 1 gale 2 male 3 tale 4 bale 5 sale 6 pale mark 1 hark 2 dark 3 mark 4 lark 5 park 6 b ark way 1 way 2 bay 3 ray 4 lay 5 day 6 pay hump 1 rump 2 bump 3 jump 4 lump 5 pump 6 hump meat 1 meat 2 feat 3 heat 4 seat 5 beat 6 peat tack 1 tab 2 tan 3 tam 4 tang 5 tack 6 tap cave 1 came 2 cape 3 cane 4 cake 5 cave 6 case
61 APPENDIX C B ACKGROUND QUESTIONNAIRE Background Questionnaire Participant Number: ___________ Name: ______________________ Birth year: ____________________ How long have you been in the United States? _______years ________months What is your highest education le vel from Korea ? _____________________ What is your highest education level from the US ? _____________________ When you listen to English, which sounds are the most difficult for you to hear correctly? List them here: ____________________________________ ________________ Do you speak any languages in addition to Korean and English? If you speak other languages also, list them here: ___________________________________________ Have you ever gone to an English institute in Korea ? Yes No How Long? ______ Have you ever gone to an English institute in the US ? Yes No How Long? ______ Have you ever taken an English pronunciation class? Yes No Have you ever taught Korean? Yes No Have you ever taught English? Yes No Except for a class about English, ha ve you ever taken a class that was taught in English? Yes No Are you currently a student? (UF, SFCC, ELI) Yes No Do you have normal hearing? Yes No How well can you spell in Korean Hangul ? 1. very well 2. good 3. average 4. poor 5. terrible How well do you spell i n English ? 1. very well 2. good 3. average 4. poor 5. terrible
62 When reading English, I understand: 1. almost everything. 2. most. 3. about half. 4. less than half. 5. not very much. When listening to English, I understand: 1. almost everything. 2. most. 3. about half. 4. less than half. 5. not very much.
63 APPENDIX D LEXICAL DECISION TASK STIMULUS LIST BY PHONEME AND CONDITION Phoneme Normal Pseudohomophone Visually Similar /f/ confess confirm confuse joyful perfect reflect safety software transfer conpess conpirm conpuse joypul perpect replect sapety soptware transper contess conlirm contuse joydul persect reblect sasety soltware transler /p/ airplane aspect complex disprove empire expect people purple purpose supreme airflane asfect comflex disfrove emfire exfect peofle purfle purfose sufreme airklane asqect comklex diskrove emlire extect peogle purkle purgose sutreme /g/ disguise finger magnet diskuise finker maknet dispuise finper masnet /k/ conclude discard discount conglude disgard disgount conplude disnard dispount /d/ advance building index older order standard sudden under window atvance builting intex olter orter stantard sutten unter wintow abvance builping inpex olber orper stanpard suvven unser winrow /t/ center disturb doctor cender disdurb docdor cenfer disburb docfor
64 entire filter master mistake practice sentence system endire filder masder misdake pracdice sendence sysdem enmire filker masfer mispake pracmice senpence sysnem /l/ airplane building complex culture decline dislike explain problem reflect unlike airprane buirdi ng comprex curture decrine disrike exprain probrem refrect unrike airpmane buinding compfex custure decmine distike expsane probwem refnect unvike /r/ approve children corner decrease degree destroy increase nearby neutral secret applove childlen colner declease deglee destloy inclease nealby neutlal seclet appwove childwen comner decnease degmee destwoy incsease neamby neutbal secfet Control honest singing money easy someone insane manage many human measure minus nonsense season horest sinjing momey ea vy soweone intane mabage mamy hugan meature mirus nolsense seaton
65 REFERENCES Baluch, B., & Besner, D. (1991). Visual word recognition: evidence for strategic control of lexical and nonlexical routines in oral reading. Journal of Experimental Psycholog y: Learning, Memory and Cognition , 17 (4), 644 652. Bernhardt, E. B. (1991). Reading Development in a Second Language . Norwood, New Jersey: Ablex. Brace, N., Kemp, R., & Snelgar, R. (2000). SPSS for Psychologists . Mahwah, New Jersey: Lawrence Erlbaum A ssociates. Frost, R. (1998). Toward a strong phonological theory of visual word recognition: true issues and false trails. Psychological Bulletin , 123 (1), 71 99. Hodgson, W. (1980). Basic Audiological Evaluation . Baltimore: Williams & Wilkins. Larsen Freeman, D., & Long, M. H. (1991). An Introduction to Second Language Acquisition Research . New York: Longman. Lesch, M. F., & Pollatse k, A. (1998). Evidence for the use of assembled phonology in accessing the meaning of printed w ords. Journal of Experim ental Psychology: Learning, Memory and Cognition , 24 (3), 573 592. Luo, C. (1996). How is word meaning accessed in reading? Evidence from the phonologically mediated interference effect. Journal of Experimental Psychology: Learning, Memory, and Cognition , 22 (4), 883 895. Paap, K. R., & Noel, R. W. (1991). Dual route models of print to sound: still a good horse race. Psychological Research , 53 , 13 24. Peereman, R., Content, A., & Bonin, P. (1998). Is perception a two way street? The case of feedback cons istency in visual word recognition. Journal of Memory and Language , 39 , 151 174. Perfetti, C. A., & Tan, L. H. (1998). The Time Course of Graphic, Phonological, and Semantic Activation in Chinese Character Identification. Journal of Experimental Psycho logy: Learning, Memory, and Cognition , 24 (1), 101 118.
66 Smith, F. (1994). Understanding Reading . (Fifth ed.). Hillsdale, New Jersey: Lawrence Erlbaum Associates. Sohn, H. (1994). Korean . New York: Routledge. Sohn, H. (1999). The Korean Language . Ca mbridge: Cambridge University Press. Stone, G. O., & Van Orden, G. C. (1994). Building a resonance framework for word recognition using design and system principles. Journal of Experimental Psychology: Human Perception and Performance , 20 (6), 1248 1268. Stone, G. O., Vanhoy, M. D., & Van Orden, G. C. (1997). Perception is a two way street: Feedforward and feedback phonology in visual word recognition. Journal of Memory and Language , 36 , 337 369. Tan, L. H., Hoosain, R., & Siok, W. W. T. (1996). Activ ation of phonological codes before access to character meanings in written Chinese. Jour nal of Experimental Psychology: Learning, Memory, and Cognition , 22 (4), 865 882. Tan, L. H., & Perfetti, C. A. (1998). Phonological codes as early sources of constrain t in Chinese word identification: A review of current discoveries and theoretical accounts. Reading and Writing: An Interdisciplinary Journal , 10 , 165 200. Tan, L. H., & Perfetti, C. A. (1999a). Phonological and associative inhibition in the early stages of English word identification: evidence from backward masking. Journal of Experimental Psychology: Human Perception and Performance , 25 (1), 59 69. Tan, L. H., & Perfetti, C. A. (1999b). Phonological activation in visual identification of Chinese two char acter words. Journal of Experimental Psychology: Learning, Memory, and Cognition , 25 (2), 382 393. Van Orden, G. (1987). A ROWS is a ROSE: Spelling, sound, and reading. Memory and Cognition , 15 (3), 181 198. Ziegler, J. C., Benraiss, A., & Besson, M. (1999 ). From print to meaning: an electrophysiological investigation of the role of phonology in accessing word meaning. Psychophysiology , 36 , 775 785.
67 BIOGRAPHICAL SKETCH Daniel Hufnagle received his B.A. in economics and political science from the U niversity of Florida in 1998. He started in the University of Floridaâ€™s Program in Linguistics in August of 1999, and he received his M.A. and TESL certification in May of 2002. He will pursue his PhD. in linguistics at the University of Pittsburgh.