semantic ambiguity in the lexical access of verbs - Ideals - University

SEMANTIC AMBIGUITY IN THE LEXICAL ACCESS OF VERBS: HOW DATA FROM MONOLINGUALS AND BILINGUALS INFORM A GENERAL MODEL OF THE MENTAL LEXICON

BY AMY PHYLLIS SWANSON

DISSERTATION Submitted in partial fulfillment for the requirements for the degree of Doctor of Philosophy in Spanish in the Graduate College of the University of Illinois at Urbana-Champaign, 2010

Urbana, Illinois Doctoral Committee: Associate Professor Paola E. Dussias, Chair Associate Professor Susan M. Garnsey Associate Professor Anna María Escobar Associate Professor Diane Musumeci Assistant Professor Gretchen Sunderman, Florida State University

ABSTRACT

This thesis describes an extensive norming study of Spanish verbs and an online language processing study investigating whether bilingual lexical processing is nonselective (both languages are activated) when only one language is required for use. To study bilingual lexical processing, researchers have relied upon words of shared orthography and semantics between languages in order to determine how word form and meaning impact bilingual word recognition. However, because these words have been of exact form overlap through cognates (words sharing form and meaning between languages: banana in Spanish and English) and homographs (words sharing form yet differing in meaning: the English adjective red meaning net in Spanish), it has been difficult to distinguish which language(s) participants engage during processing tasks. The present research addresses this issue by investigating cognate and homographic verbs between languages. Because differences in verb morphology between Spanish and English never result in exact form overlap between languages (e.g., assist and asistir), interlingual cognate and homographic verbs between Spanish and English should ensure that participants operate in one specific language. Hence, utilizing verbs provides an original testing ground to determine if the bilingual language processor is nonselective when operating in one language and to what degree the access depends on form and meaning overlap between languages. An extensive norming study of Spanish verbs produced a reliable list of cognates and homographs with English. The online research indicated that bilingual lexical access is guided not only by form and meaning, but also by how the frequency of a word’s meanings from both languages attach to a single form. These results mirror recent discoveries in ambiguity research in monolinguals (e.g., Rodd,

ii

Gaskell and Marslen-Wilson, 2002), the implications of which suggest an overriding mechanism of language processing—not just a theory of bilingual lexical processing.

iii

ACKNOWLEDGEMENTS

I thank many kind people for their guidance to me in this journey. First and foremost, I thank my advisor who worked with me long-distance for the entire ride—Prof. Paola (Giuli) E. Dussias. Thank you for your patience, continued encouragement, generosity, hospitality, companionship in research trips and conferences, and mostly for being such a kind and decent person always. I value you as an educator and researcher and as my dear friend. I really couldn’t have done this without you! Thank you to my committee of amazing women for your patience, time, and energy in reading and guiding: Prof. Susan M. Garnsey, Prof. Anna María Escobar, Prof. Diane Musumeci, and Prof. Gretchen Sunderman. An additional thank you goes to Prof. Susan Garnsey for letting me participate in her SMG lab at the Beckman Institute. The guidance, encouragement, and training you give to your students serve as a model for all educators. I thank Prof. M. Teresa Bajo at the University of Granada, Granada, Spain, for allowing me to conduct research in her lab and appreciate her and all her students for their friendship, hospitality, and collegiality to me during my stays in Granada. I send the same appreciation to Prof. Judy Kroll at Penn State for her excellent feedback and to the Center of Language Science (and the Purple Lab) for allowing me to interact with them during visits. Crossing half the country to meet with one’s advisor and gathering research data abroad do not come cheaply. I thank the Graduate College for funding me with off-campus and on-campus dissertation grants, as well as my department, Spanish, Italian, & Portuguese (SIP), for awarding me the prestigious Darlene F. Wolf Fellowship. Additionally, I thank the Graduate College, SIP, and the University of Trento, Italy, for various conference travel grants. Finally, I thank the people closest to me in my personal life. You all know that you share in this with me—many of iv

you have told me as much, so yes, you have earned your degrees too! I have the most incredible family and I thank each one, starting of course with my parents, Emil and Joan Swanson. Of you both, I stand in awe. I thank my siblings, their spouses, and my nieces and nephews. Your love and support are unparalleled. Finally, I thank my husband, Ronald C. Sylvester, for his patience. You never gave up on me and that has made a lot of difference.

v

TABLE OF CONTENTS

CHAPTER 1: INTRODUCTION ...................................................................................................1 CHAPTER 2: LITERATURE REVIEW ......................................................................................21 CHAPTER 3: IDENTIFYING SPANISH VERBS AS COGNATES AND HOMOGRAPHS WITH ENGLISH ...........................................................................................................................53 CHAPTER 4: INVESTIGATING ONLINE INTERLINGUAL LEXICAL AND SEMANTIC ACTIVATION THROUGH SPANISH-ENGLISH COGNATE AND HOMOGRAPHIC VERBS ............................................................................................................84 CHAPTER 5: DISCUSSION AND CONCLUSION .................................................................105 REFERENCES ............................................................................................................................120 APPENDIX A: PHASE I DATA: HOMOGRAPH SELECTION. DICTIONARY DEFINITIONS, FREQUENCY. ..................................................................................................131 APPENDIX B: PHASE I DATA: COGNATE SELECTION. DICTIONARY DEFINITIONS, FREQUENCY ..................................................................................................137 APPENDIX C: PARTICIPANT CONSENT FORM IN SPANISH ..........................................142 APPENDIX D: PARTICIPANT CONSENT FORM TRANSLATED TO ENGLISH .............144 APPENDIX E: LANGUAGE BACKGROUND QUESTIONNAIRE IN SPANISH ................146 APPENDIX F: LANGUAGE BACKGROUND QUESTIONNAIRE IN ENGLISH ...............148 APPENDIX G: SYNONYM-SOLICITATION TASK FOR HOMOGRAPHS ........................151 APPENDIX H: SYNONYM-SOLICITATION TASK FOR COGNATES ...............................153 APPENDIX I: SYNONYM-CLARIFICATION TASK.............................................................155 APPENDIX J: SYNONYM-RATING TASK FOR HOMOGRAPHS, SPANISH....................158 APPENDIX K: SYNONYM-RATING TASK FOR COGNATES, SPANISH .........................164 APPENDIX L: SYNONYM-RATING TASK FOR HOMOGRAPHS, ENGLISH ..................167 APPENDIX M: SYNONYM-RATING TASK FOR COGNATES, ENGLISH ........................172 APPENDIX N: FORM-SIMILARITY RATING TASK ...........................................................185 vi

APPENDIX O: HOMOGRAPH DATA, SYNONYM-SOLICITATION TASK ......................192 APPENDIX P: COGNATE DATA, SYNONYM-SOLICITATION TASK .............................193 APPENDIX Q: HOMOGRAPH DATA, SYNONYM-CLARIFICATION TASK ...................201 APPENDIX R: COGNATE DATA, SYNONYM-CLARIFICATION TASK ..........................217 APPENDIX S: SUMMARY OF HOMOGRAPH DATA, SYNONYM-CLARIFICATION TASK ...........................................................................................................................................233 APPENDIX T: SUMMARY OF COGNATE DATA, SYNONYM-CLARIFICATION TASK ...........................................................................................................................................244 APPENDIX U: HOMOGRAPH DATA, SYNONYM-RATING TASK ...................................251 APPENDIX V: COGNATE DATA, SYNONYM-RATING TASK .........................................259 APPENDIX W: FORM SIMILARITY DATA ..........................................................................267 APPENDIX X: PICTURE BANK ..............................................................................................270 APPENDIX Y: SPANISH PROFICIENCY TEST ....................................................................271 APPENDIX Z: ENGLISH PROFICIENCY TEST ....................................................................274 APPENDIX AA: TARGET VERB LIST ...................................................................................276 APPENDIX BB: CONTROL VERB LIST ................................................................................280 APPENDIX CC: PSEUDOWORD LIST ...................................................................................281 APPENDIX DD: VERB FAMILIARITY TASK .......................................................................282

vii

CHAPTER 1: INTRODUCTION

Research on bilingual lexical processing originally focused on the question of whether bilinguals automatically (subconsciously) activate one or both languages in the mind when encountering words of either language (e.g., Chen, 1990; Chen and Leung, 1989; De Groot and Nas, 1991; Dufour and Kroll, 1995; Kroll, 1993; Kroll and Stewart, 1994; Potter, So, von Eckhart and Feldman, 1984; Weinreich, 1953—reprinted in 1963). Lexical access or lexical processing is the process of entering the mental lexicon to retrieve words and information for their use. The bilingual mental lexicon refers to the storage of all words and their lexical information. A broad view of the lexicon says that lexical representations include not only the grammatical, syntactic, semantic and pragmatic properties of words themselves, as in phonological (sound), orthographic (form-spelling), semantic (specific meaning and notions about that meaning—e.g., concreteness or abstractness), and morpho-syntactic (grammatical class—e.g., noun or verb) properties, but also information on the co-occurring items related to a word, as in directions for use of the lexical item within sentence structure and collocational properties (rules of constraint of a lexical item within phrases and context– e.g., you can launch an idea or a rocket, but you cannot launch an exam) (See recent research on the mental storage of words by Li, Shu, Liu, & Li, 2006). A selective view of bilingual lexical access posits that language input is processed only by the context-appropriate lexicon. A nonselective view holds that both language systems respond to language input in a parallel manner. In the last decade, however, empirical findings have moved researchers away from the ‘either or’ question of whether one or both languages are activated automatically, even when a

1

bilingual intends the use of only one of his or her languages. Instead, the data overwhelmingly support a generally nonselective bilingual lexical access view (Christoffels, Firk, and Schiller, 2007; Costa, Caramazza, and Sebastian-Galllés, 2000; Dijkstra and VanHeuven, 2002; Schwartz and Kroll, 2006) and an organization of the bilingual mental lexicon that connects multiple languages at both semantic and lexical levels (Kroll and Stewart, 1994; Potter, So, von Eckhart and Feldman, 1984). What remains to be revealed, however, is an understanding of the mechanisms and constraints that guide nonselective bilingual lexical activation to allow a bilingual ultimately to choose the appropriate language when retrieving words. In this introduction, a brief overview is provided to explain what it means to be bilingual, how research findings over the past three decades have shaped the models of bilingual lexical organization and access, how lexical ambiguity has been used in studying bilingualism, how the current study manipulates lexical ambiguity in innovative ways in order to understand further the mechanisms and constraints that guide nonselective bilingual lexical processing, and how monolingual and bilingual lexical processing might be considered jointly to describe a general view of language processing. THE BILINGUAL At one time, the term bilingual was perceived to mean balanced bilingual—a highly proficient, near-native or native speaker of two languages. While there are more bilingual and multilingual speakers than monolinguals in the world today, very few bilinguals of the balanced type actually exist. In order to reflect the type of bilingual that readily is found, bilingualism is viewed along a continuum. A bilingual may have achieved low or high proficiency in a second language (L2). Nearly always, a bilingual has one dominant language. Depending upon

2

individual circumstances, this dominant language does not have to be the bilingual’s first language (L1). The term bilingual, then, has a very broad connotation in the bilingualism literature, and refers to a person who actively uses two languages at some level of proficiency—a native language and a second language (L2). The term is sometimes interchangeable with "L2 learner", and levels of proficiency, age of language acquisition/learning, length of immersion time, manner of language study, amount of daily language usage, and other relevant descriptors are considered and included in the research on bilingual language processing (Kroll and Dussias, 2004). THE BILINGUAL LEXICON The original research question regarding the nature of the bilingual lexicon asked whether or not two languages are stored separately or are integrated in some way into a single-language store within bilingual memory. In 1953, Weinreich (reprinted in 1963) put forth three types of possible organizations for the bilingual lexicon: Type A: “compound”, Type B: “coordinate”, and Type C: “subordinate”. These organizational types describe two languages as stored separately within the bilingual mind (compound), or connected either by shared conceptual stores (coordinate) or via lexical links (subordinate). The conceptual level holds all real world knowledge and meanings or events to which words refer. The lexical level of the lexicon refers to word-level representations within the mind. Only aspects of word form, such as the orthography of words, are stored at this level. During L2 learning, compound bilinguals gain one lexical representation for each conceptual representation. That is, in this type of bilingual, languages are stored separately in the bilingual mind and connections between the two languages are not developed. Coordinate

3

bilinguals develop two separate lexical representations for each concept. In this case, while concepts may be shared between languages in the bilingual, word-level representations for concepts are stored separately. The subordinate view describes a process in which a second language is learned via an existing L1. A second language learner at first does not create a direct link between an L2 word and its conceptual representation, rather s/he creates a connection between the newly-learned L2 word form and the L1 word form equivalent. Potter, So, von Eckhart and Feldman (1984) carried out the first explicit test regarding the association between word organizations of two languages. Potter et al. (1984) assumed, as did Weinreich, that some type of connection between the L1 and L2 must be made during the learning of new L2 words, and put forth the Word Association Hypothesis and the Concept Mediation Hypothesis (See Figure 1.1) as possible theories that could explain how those connections are made. The Word Association Hypothesis suggests that the L1 and the L2 are connected at the lexical level and that only the L1 has access to conceptual representation. The Concept Mediation Hypothesis suggests that the L1 and L2 are connected conceptually. That is, each language has independent access to a common conceptual representation. Potter et al. (1984) also proposed a third hypothesis that mirrors the subordinate bilingualism distinction that eventually leads to coordinate bilingualism in many bilinguals, as put forth by Weinreich (1953, 1963). The Intermediate Hypothesis suggests that L2 learners first acquire lexical associations to their L1 while learning new L2 vocabulary, thus performing under the Word Association Hypothesis. Gradually, direct links are developed between the L2 words and their conceptual representations, as in the Concept Mediation Hypothesis. While the Potter et al. (1984) data supported only the Concept Mediation Hypothesis, other empirical data have found support for the Intermediate Hypothesis (e.g., Chen and Leung 1989 and Chen, 1990). 4

Figure 1.1 Concept Mediation and Word Association Models (from Potter et al., 1984).

Images

Images

L2

L2 L1

L1

Concepts

Concepts

Concept Mediation Model

Word Association Model

The fact that L2 vocabulary acquisition sometimes shifts from its reliance upon L1 connections for meaning to direct conceptual connections suggested to researchers that an asymmetry may exist in the strength of lexical-conceptual connections between languages in the bilingual (Kroll and De Groot, 1997). To account for this possible asymmetry—a proficiency based asymmetry, Kroll (1993) and Kroll and Stewart (1994) proposed the Revised Hierarchical Model for the bilingual lexicon (See Figure 1.2). This model suggests that when a bilingual learns an L2, lexical connections are formed between the two languages. As a bilingual becomes more fluent in the L2, stronger direct links are established between the L2 and the conceptual store, although these links will not cause the already existing strong lexical links from the L1 to the L2 to disappear. The lexical store for the L2 is purposefully smaller than the lexical store for the L1 because even for highly proficient bilinguals, it is assumed that more words are known in the L1 than in the L2. While data exist in support of the Revised Hierarchical Model (Dufour and Kroll, 1995; Keatley, Spinks, and De Gelder, 1994; Kroll, 1993; Kroll and Stewart, 1994; Sholl, Sankaranarayanan, and Kroll, 1995), there are data against the hypothesis (e.g., Altarriba and Mathis, 1997; De Groot and Nas, 1991). 5

Figure 1..2 Hierarchical Bilinguaal Model (from Kroll & Stewart, 19994).

RS THAT SHAPE S A MODEL M OF F BILINGU UAL ORGAN NIZATION N AND ACC CESS FACTOR Such conflictting researchh data indicatte that the orrganization of o the bilinguual lexicon is i much moore than just a question of o whether or not two lannguages in thhe bilingual mind are “separatee” or “integrated”. Manyy lexicon-exxternal variabbles, such ass the mannerr in which bilingualls learn an L2 2 (in a classrroom or in an a immersionn setting), thhe levels of proficiency p o of bilingualls, age of acq quisition, andd the types of o experimenntal tasks em mployed in reesearch (e.g.., lexical deecision or word w naming— —defined att the end of this t chapter in i the sectionn on terms) all play a role in determining whether two languuages are connnected in thhe bilingual mind (Greenn, 1993). While W a modeel on bilinguual lexical orrganization and a access must m accountt for lexiconexternal notions n such h as whether a bilingual is balanced or o dominantt in one languuage, a moddel must first describe ho ow orthograpphy, phonoloogy, semanttics, and otheer lexical prooperties interact between languages. Inndeed, much h attention inn research haas been givenn to words thhat are orthoographically similar between lang guages (e.g., De Groot annd Nas, 19911; De Groot,, Delmar, and Lupker, 20000; Dijkstra, Van Jaarsveeld, and Brinnke, 1998; Scarborough,, Gerard, andd Cortese, 19984; Gerard and 6

Scarborough, 1989). Cognates are words that share the same or similar lexical form and meaning but usually differing phonological properties between languages (e.g., hotel in Spanish and English). Because of the interlingual orthographic form overlap of cognates, they are utilized in research and are compared to noncognates in order to determine if words are connected between languages via orthographic form. Noncognates are direct semantic translations between two languages that do not share lexical orthographic form (e.g., perro in Spanish means dog in English). In research, noncognates are activated quickly when their translations are presented together, thus suggesting that their concepts are connected in some way (e.g., De Groot and Nas, 1991). Cognates are shown to be recognized quickly as compared to noncognates, thus suggesting interlingual connections at both lexical and conceptual levels (e.g., Gerard and Scarborough, 1989; De Groot and Nas, 1991). In addition, De Groot and Keijzer (2000) carried out research involving the learning of L2 concrete words, abstract words, cognates, and noncognates, and found that cognates and concrete words were remembered more successfully than noncognates and abstract words from the beginning of foreign language learning. These types of empirical results have led recent researchers away from the either-or question of a separate or shared bilingual lexicon, and toward investigating the varying circumstances under which two languages work together in the bilingual mind at lexical (word) and/or conceptual (semantic) levels within the lexicon. De Groot (1992, 1993) proposes that the bilingual lexicon is organized by distributed features, an idea fashioned after distributed models of speech production (e.g., Dell and O’Seaghdha, 1992). By focusing on the aspects of words that appear to be associated with lexical or conceptual processing, De Groot (1992, 1993; Kroll and De Groot, 1997) put forth a theory of the bilingual lexicon that relies on three levels of representation: a lexical level that includes only aspects of word form, a conceptual level that 7

includes real world knowledge and the meanings of the objects and events to which words refer, and a lemma-level that is sensitive to syntactic constructions and mappings between lexical and conceptual features. This model, referred to as the Distributed Lexical/Conceptual Feature Model (See Figure 1.3), permits shared aspects of word form (lexical features) and meaning (conceptual features) between two languages that are interconnected via language-specific lemmas, thus allowing two languages in bilingual memory to function either autonomously or in an integrated fashion. Figure 1.3 Distributed Lexical/Conceptual Feature Model. Adapted from Kroll and De Groot, 1997.

Lexical Features

L1

Lemmas

L2

Conceptual Features

Dong, Gui and MacWhinney (2005) combined notions from all the aforementioned models to create The shared (distributed) asymmetrical model (See Figure 1.4). This model and similar versions (e.g., the Unified Model, MacWhinney, 2005 and 2007) are extensions of MacWhinney’s monolingual language processing model, the Competition Model (Bates and MacWhinney, 1982; MacWhinney, 1987), modified to include bilingual language processing. In this model, shared conceptual features between languages are represented by a storage area of common elements, while language-specific conceptual aspects remain in smaller languagespecific storage areas. Both lexicons have access to all three conceptual areas, yet proficiency 8

asymmetry and early routing of the L2 through the L1 lexicon are accounted for in the same way as in the Revised Hierarchical Model. Dong et al. (2005) suggest that as the L2 is learned, both common and L1-specific semantic notions are linked to the L2, yet this link between the L1 and L2 weakens as the L2 learner becomes more proficient and an L2-specific conceptual store is created. As strong as the ‘L2-L2’ link might become, however, it can never be as strong as the ‘L1-L1’ link. For advanced bilinguals, the language-specific links (L1-L1 specific/common stores; L2-L2 specific/common stores) will strengthen as the cross-linguistic connections weaken. Figure 1.4 The Shared (Distributed) Asymmetrical Model. Dong, Gui, & MacWhinney, 2005.

(lexical form)

L1 elements

L2

L1

Common elements

L2 elements

(semantic elements)

While all of these models give a valuable general description of how bilingual lexical organization and access might be conceived, they do not provide a detailed account of how lexical access actually occurs in the bilingual mind. One of the only models to date to detail bilingual lexical access is the Bilingual Interactive Activation + Model (BIA+), proposed by Dijkstra and Van Heuven, 2002, as a revised version of their 1998 Bilingual Interactive 9

Activation Model (BIA). The BIA+ model is a localist computational model in the connectionist tradition. For more than 20 years, connectionist models have been generated to describe unilingual language comprehension (and production) that encompass both the developing and the adult final state language system. Bilingual lexical modeling is just beginning to emerge and only to describe lexical organization and activation for a static bilingual state. While it can account for variances in proficiency between languages (e.g., one language is more dominant than the other in a bilingual) through notions of frequency, like many localist models, it does not address how the model would change through the L2 learning and acquisition process. The BIA+ model is fashioned after the monolingual Interactive Activation (IA) model of word recognition (McClelland and Rumelhart, 1981) and consists of orthographic, phonological, and semantic representations which are connected both within and between languages. As a letter string is fed into the model, orthographic and phonological features of a word interact with and activate words of similar orthography and phonology in parallel, which in turn interact with word semantics. For example, rosa in Spanish can activate its translation rose, which in turn can activate hose or other feature-similar words. Information is fed in a bidirectional manner through the system; so as orthography > phonology > semantics activate potential word candidates in a feed forward process, semantics > phonology > orthography provide feedback activation until all lexical and semantic matches take place and the appropriate lexical candidate is selected. A final language identification node contributes to the activation process by interpreting language-specific cues and identifying the language to which a word belongs. An extra component controls how task and decision demands can affect the word identification process (See Figure 1.5).

10

Figure 1..5 The BIA+ model for bilingual b worrd recognitioon (Dijkstra & VanHeuvven, 2002).

T BIA+ mo The odel for bilinngual word recognition r i able to expplain why words is w that shaare form andd/or meaning g between lannguages are recognized quickly by the t system (ii.e., as in cognates and translattions mentioned previouusly). Yet ann interesting ambiguity phenomenon p n nd between-language in which wordds share simiilar lexical feeatures exists both within- an p , yet differ inn semantics (e.g., bug in English cann mean spy (orthograaphy and/or phonology), device orr insect). Th hese words, called c homoggraphs havee been of inteerest to reseaarchers investigaating lexical processing because b theyy allow orthoography, phoonology, andd semantics to t be manipulaated and teassed apart tow ward a betterr understandiing of how each e componnent contribuutes to the woord recognitiion process and a in particcular, how eaach interacts to resolve leexical ambigguity.

11

Within-language ambiguity: intralingual homographs Researchers from a monolingual language processing perspective have exploited the within-language phenomenon of homographs (e.g., bug meaning spy device or insect or even the verb to bother) to understand how meaning is activated during lexical parsing. At issue are the same types of questions as posed for bilinguals: are multiple word meanings (within- rather than between-language) of ambiguous intralingual homographs automatically activated and available for use in language comprehension? What are the constraints for such activation that would resolve activation in favor of only one meaning? In general, this body of research has shown that the lexical processor activates all of the meanings of an ambiguous word, but resolves the ambiguity quickly, keeping active only the relevant meaning as required by context (e.g., Simpson and Burgess, 1985; Swinney, 1979). However, additional research on monolingual lexical ambiguity has found that we must differentiate between types of ambiguity. While many researchers have found facilitation in the activation for words with multiple meanings in one language (English) (e.g., strongest effects found in the following: Azuma and Van Orden, 1997; Borowsky and Masson, 1996; Millis and Button, 1989), Rodd, Gaskell and Marslen-Wilson (2002) distinguish between lexical items that truly are ambiguous (having multiple, unrelated meanings, as in bark) and those that are polysemous (having multiple meanings of a similar sense, as in twist). By drawing this distinction, Rodd et al. were able to show that while facilitation in activation for polysemous words readily occurs, there is actually an ambiguity disadvantage for words when the meanings are clearly different. Klepousniotou (2002) reported similar findings. These results suggest that accessing homographs of multiple yet unrelated meanings creates competition in the lexical

12

processor at a semantic level, whereas processing words with moderately related meanings facilitates the activation of such words. Clearly such research findings from intralingual homographs/monolingual lexical ambiguity research cannot be ignored when considering questions of between-language ambiguity. Often, researchers in bilingual lexical access argue that a model of lexical organization and activation must accommodate speakers of multiple languages, largely arguing that being bilingual is more common than not in the world of language speakers. Likewise, researchers investigating the bilingual lexicon must utilize within-language lexical (form) and semantic relationship findings when considering between-language ambiguity resolution. Between-language ambiguity: interlingual homographs Like monolingual research, bilingual lexical processing research utilizes words that provide cross-language ambiguity in order to investigate whether multiple languages are activated automatically. Interlingual homographs are words of the same or similar orthography yet differing meanings between languages (e.g., the color adjective red means net or web in Spanish). As with cognates, because of the orthographic form overlap of interlingual homographs, researchers can manipulate the items in research to determine if lexical items are connected between languages via form overlap and if language-specific meanings automatically are activated.

Unlike monolingual research findings, some research utilizing interlingual

homographs between languages found neither slowed nor facilitated activation for orthographically identical or similar words, suggesting that only the appropriate language meaning is activated (e.g., Gerard and Scarborough, 1989). Most data from interlingual homographs, however, support monolingual findings for a nonselective access view in which

13

both language meanings are automatically and simultaneously activated, as long as research tasks include both languages (e.g., Altenberg and Cairns, 1983; Beauvillain and Grainger, 1987; De Groot, Delmar, and Lupker, 2000; Dijkstra and Van Heuven, 1998; Dijkstra, Van Jaarsveld, and Ten Brinke, 1998; Grosjean, 1997, 1998, 2000). As Grosjean (1997) points out, the mere presence of multiple languages in research design can and probably does encourage the bilingual lexical parser to keep all languages activated and ready for use in language processing. Wordlevel research tasks that present two languages at a time do not really reflect the way that language is used by bilinguals. Although much evidence exists that bilinguals code-switch or change back and forth between languages under many circumstances, bilinguals also consciously make use of one language at a time. The few times that experimental items have been presented in a language-specific manner, results have been much less clear as to whether the bilingual processor automatically activates both languages (selective activation in Dijkstra, Van Jaarsveld, & Ten Brinke (1998) but nonselective results in De Groot, Delmar, & Lupker (2000). What is interesting to know, then, is if while operating in one language, bilinguals automatically activate elements of the lexicon of their other language. More research of this type is crucial toward obtaining a broader understanding of how the bilingual lexical processor works. In existing bilingual lexical research, a specific type of cognate and homograph has been exploited: one in which exact orthography is shared between languages (e.g., hotel as a SpanishEnglish cognate; red as a Spanish-English homograph). De Groot, Delmar, & Lupker (2000) suggest that when such words are presented in a language-specific study, conflicting results may be a consequence of lexical parsing strategies. That is, when participants are asked to identify words of exact orthographic overlap in a particular language (e.g., Is red a word of Spanish?), the processor simply responds the moment it finds any lexical match in the mind, regardless of 14

language ownership, telling us nothing about whether meaning ambiguity between languages has been tapped. In addition, cognates and homographs in bilingual lexical research have been limited to nouns and/or words that cross-categorize between languages (the English adjective red translates to the noun web/net in Spanish). As a consequence, bilingual lexical modeling (e.g., the BIA+ model) is based only on how very short words (the BIA+ model can only simulate the processing of 4-letter words), predominantly nouns, are processed. Van Hell (2002, see also Sunderman and Kroll, 2006) suggests that grammatical class may constrain existing models of word recognition and ultimately affect theories of word recognition both at the word level and in sentence processing. Research suggests that the processing of nouns and verbs may be very different since verbs are believed to be more abstract in meaning and to have a greater breadth of meaning than concrete nouns (Van Hell and de Groot, 1998). Bilingual lexical processing research has more or less ignored this notion by focusing on ‘all or nothing’ overlaps in all lexical aspects between languages. Yet, if one explores cognate and homograph meanings between languages, one finds a continuum of form and meaning overlap (see Van Hell and de Groot, 1998). More word-level research is needed that looks at the interaction of specific grammatical classes, such as verbs, between languages, and that exploits the notion of degrees of form and meaning overlap between languages. At the same time, research in bilingual lexical activation must consider how findings that ambiguous and polysemous words act differently in monolingual lexical processing might impact bilingual lexical access.

15

SIGNIFICANCE OF THE STUDY The present research study addresses the aforementioned gaps in the bilingual lexical processing literature in the following ways. The research design moves research from a mix of word classes in the experimental design to a single grammatical class by investigating cognate and homographic verbs between languages. Because differences in verb morphology between Spanish and English never result in exact form overlap between languages (e.g., assist and asistir), interlingual cognate and homographic verbs between Spanish and English should ensure that participants are operating in one language, thus removing potential results that could arise from strategic processing mechanisms (as suggested in De Groot et al., 2000). Because verbs are more abstract in meaning, degrees of semantic overlap between languages can more easily be teased apart and notions of meaning dominance and whether meanings are very different in nature or simply polysemous can be studied. Hence, utilizing verbs provides an original testing ground to determine if the lexical information of two languages as expressed by verbs is automatically accessed during bilingual language processing, to what degree the access depends on form and meaning overlap between languages, and whether bilingual and monolingual lexical processing mechanisms of ambiguous and polysemous words mirror each other.

16

DEFINITION OF TERMS bilingual—active speaker of two languages, regardless of level of proficiency. cognates -- words in two languages that share similar or exact orthographic features and word meaning, but different phonetic features. An example is hotel in Spanish and English (in Spanish, pronounced /otel/). early bilinguals—speakers of two languages who acquired each language simultaneously at an early age (e.g., on average age 6). grammaticality judgment task—a task in which participants decide whether a sentence is grammatically correct in a language. Often participants rate their decisions on a scale of options, from very certain to not certain of the grammaticality of the sentence. This task is usually a paper and pen task and is considered an "off-line" task, meaning that participants can have as long as needed consider the grammaticality of the sentences. homographic or false cognate verbs—words between languages that share similar lexical form, but different meanings. An example is asistir in Spanish, which could (mistakenly) appear to mean assist in English, but instead means to attend. interlingual homographs—words between languages that share the same lexical form, but different meanings. An example is red in English, which means net in Spanish. intralingual homographs—ambiguous words within one language that have multiple meanings. An example is bug, which could refer to a listening device for spying or an insect. lexical decision task—an experimental task in which a letter string appears on a computer screen and participants indicate whether the letter string is a legal word in a particular language. 17

The letter strings that form non-words are created by changing only one letter of a legitimate word, while ensuring that the new letter combinations are legal to whatever language is being tested (e.g., computer might be changed by one letter to camputer, with the cam combination being one that is acceptable in English (campaign, camping). Response times (RTs) and Error Rates (ERs) are collected and these measurements are interpreted to explain lexical connections in the mind. lexicon—a generative linguistics definition posits it as the storage in the mind of all idiosyncratic meanings of words, as well as instructions for their use within language (e.g., a noun phrase normally follow the verb to eat, and that noun phrase must be an edible object). Cognitive science views of the lexicon describe the lexicon in terms of feature/form (orthographical/phonological features) and semantic levels and seek to understand how such levels interact in word identification, and how form and semantic features are connected across lexical items. lexical access (lexical activation, word identification process)—the process of searching the lexicon to retrieve information about words. masked priming—a technique in which a target word in a task is preceded by a prime of a particular type (e.g., a semantically related word, an orthographically related word, etc.). The prime is preceded by a masking figure, often in the form of ‘#########’. The mask disappears, followed by the prime, which also disappears before being followed by a target (e.g., ##### car CAT). Participants in research are asked to respond to the primed paradigm in some way, perhaps with a lexical decision task (e.g., responding whether the target is a word of English). See also primed lexical decision task.

18

monolingual—speaker of one language without ability to use an additional language actively. neighborhood effects—a lexical item stored in memory shares orthographical (spelling) features with other lexical items. When words differ by only one letter (e.g., clue/club), they are considered to be neighbors in the lexicon. A word may have a large or small lexical neighborhood in terms of size, depending upon the number of orthographically similar neighbors a word has. For example, a word like clue has a small lexical neighborhood because only glue and club differ from it by one letter. A word with a large lexical neighborhood is sand because many words differ from it by one letter: band, send, said, sank. The effects of neighborhood size on word recognition have been tested across a variety of experimental tasks. In addition, neighborhood frequency has been tested. Neighborhood frequency refers to the presence or absence of high frequency words in a word's orthographic neighborhood. For example, neighborhood frequency measures the frequency of each of the neighbors for sand above (band, send, said, sank). Researchers have tested the effects of neighborhood frequency on word recognition using a variety of experimental tasks (e.g., naming, lexical decision) in order to determine if a word is recognized more quickly when its neighbors are of high frequency. noncognates—words between two languages that share meaning, but not lexical features. Examples are perro in Spanish meaning dog in English. primed lexical decision task—the same as the lexical decision task with an added component. Before deciding if a letter string seen on a computer screen is a legal word of a particular language, the participant first sees (or hears in a cross-modal version) an individual word, partial sentence, or sentence. The task enables the researcher to determine whether the first item seen (the prime) affects how quickly a subsequent lexical item is responded to. Inferences are made

19

according to response times and accuracy rates as to the nature of connections and activation of items in the mental lexicon. priming effect—a priming effect is said to occur if a target preceded by a prime is responded to more quickly than a comparable control. This facilitation suggests that the particular primetarget relationship being tested reflects lexical connections in the mind. pseudowords—letter strings created for the purpose of lexical decision tasks in which a legitimate word of a language (e.g., computer) is changed by one letter to create a phonologically legal but nonexistent word in the language (e.g., camputer). second language (L2) learners—people who are in the process of learning another language. Levels of proficiency vary among L2 learners, from non-proficient and in the early stages of learning to very-proficient, having reached a stable and competent level in the L2. translation recognition task—task in which participants determine whether two words presented at the same time are accurate translation equivalents between two languages.

20

CHAPTER 2: LITERATURE REVIEW

INTRODUCTION In the previous chapter, an overview of the research on bilingual lexical organization and access was given. The BIA+ model of bilingual word recognition has been constructed to explain the empirical findings that lexical (orthographic and phonological) and semantic levels interact both within- and between-languages. These findings, however, are based predominantly on the following: experiments utilizing short cognate and homographic nouns of exact orthographic overlap between languages, experiments presenting materials in both languages of the bilinguals being tested, experiments testing semantic connections through translations as opposed to through within-language synonyms, and experiments that conceive of semantic overlap between languages as ‘all or nothing’, rather than as a continuum of shared semantic overlap. The lexical decision task (unprimed or primed, masked or unmasked—defined in the previous chapter) readily is used to probe lexical and semantic relationships in the lexicon, since judgments on word legality are believed to tap both lexical and semantic levels of the lexicon. This chapter reviews the empirical data, predominantly from lexical decision tasks, that shape a current understanding of lexical and semantic connections both within- and between-languages in order to point out how the aforementioned gaps in the research on bilingual lexical organization and activation are accommodated by the current research design.

21

LEXICAL CONNECTIONS: ORTHOGRAPHY AND PHONOLOGY Monolingual, orthographic, and phonological connections Connectionist word recognition models in the literature on monolingual word processing propose that all orthographically similar lexical items are activated as potential word candidates before the lexical processor chooses the appropriate lexical item. This modeling notion comes out of empirical research that tested the effects of orthographic neighborhoods on lexical access. Lexical neighbors are defined by a metric called "N", first used by Coltheart, Davellar, Jonasson, and Besner in 1977 (reviewed in Andrews, 1997). This metric refers to the number of words that can be created from a word by changing just one letter of that word. For example, sand can be said to have many lexical neighbors: band, sang, sank, send, or said whereas a word like club only has one potential neighbor, clue. Activation of lexical neighbors is said to be modulated by a word’s frequency, so that a word that is more frequent in the corpora or more subjectively frequent to an individual receives more activation than words with low frequency. Sears, Hino, and Lupker (1995) tested the effects of a word's frequency, orthographic neighborhood size, and neighborhood frequency on English lexical decision Response Times (RTs). Neighborhood frequency refers to the presence or absence of high frequency words in a word's orthographic neighborhood. Various groups of high- and low-frequency words were tested in the lexical decision task: words with no neighbors, words with a large neighborhood, and words with a small neighborhood. Large and small neighborhood words were broken down into three other categories: words having at least one high frequency neighbor, words having many high frequency members, and words having no neighbors of high frequency. Results indicated that word recognition was facilitated for words having large neighborhoods.

22

Neighborhood frequency did not speed up RTs when high- or low-frequency words had small neighborhoods. However, low-frequency words were recognized more quickly when they had large neighborhoods containing higher frequency neighbors. This suggests that in general, for monolingual English speakers, having many orthographically similar neighbors can feed the activation of a particular word and make it easier to identify. In addition, low-frequency words that normally are slower to identify receive an activation boost from the high frequency of their neighbors. Such effects can only be found in a lexical model in which interactive connections exist among words at the orthographic level. In order to see if similar orthographic effects are found in Spanish, Carreiras, Perea, and Grainger (1997) tested the effects of orthographic neighborhood size and frequency in a Spanish monolingual lexical decision task. Words with large or small neighborhoods were tested, as were words with neighborhoods containing high or low frequency words. Like the Sears et al. data, results showed that participants responded more quickly to words with more neighbors. However, contrary to the Sears, et al. data, an inhibitory effect was found for neighborhood frequency. Words with high-frequency neighbors were responded to more slowly than those with low-frequency neighbors. Carreiras et al. suggested that the difference may be explained in part by how each language maps orthography and phonetics, with Spanish having a much more consistent mapping relationship between the two. In a language like Spanish with fairly regular spelling-to-sound correspondences, the orthographic neighbors are also phonetic neighbors (e.g., alba has as a neighbor alta and both words have the same vowel pronunciation) and this may provide more controlled phonological neighbors by which to test neighborhood frequency effects on lexical activation. In a language like English, orthography and phonetics do not necessarily correspond. A word like deaf in English has at least two neighbors that are pronounced 23

differently—dear and leaf. Carreiras et al. argue that this type of mismatch in phonology may in fact confound results in the Sears et al. data since such phonological inconsistency may affect purely orthographic neighborhood research effects. At the very least, Carreiras et al. propose that this type of phonological discrepancy must be controlled for. Additional research in English, utilizing a masked lexical decision task, has shown that a lack of consistency in orthographic mapping to identical phonology can slow the parser in low frequency word identification. One such study by Stone, Vanhoy, and Van Orden (1997) utilized an un\primed lexical decision task to investigate how monolingual English speakers responded to words containing a phonological sound group that can be spelled more than one way (e.g., /_ip/ as in heap and deep) versus words containing a sound pattern that can only be represented with one spelling in English (e.g., /_Ob/ as in probe and globe). They found that words with inconsistent sound-orthography mappings took longer to respond to than did words with just one sound-spelling correspondence in English. This can be interpreted like the Carreiras et al. (1997) data in which words that are of similar orthography and phonology (e.g., alta/alba and heap/deep force the processor to search discrete mappings more closely in order to tease out an appropriate word. Taken together, monolingual research on the effects of orthography and phonology in lexical processing make it clear that these properties interact with word frequency to affect the word identification process. It seems that the lexical processor activates potential lexical candidates based on similarity in orthography and phonology before landing on the appropriate match. Sometimes, this interaction can boost the availability of certain words, as in the case of low frequency words that have neighborhoods with high frequency words in them. However,

24

when lexical candidates get activated that require more attention to discrete orthographyphonology mappings, the parser is slowed. Because orthography and phonology interact in monolingual lexical processing, it is logical to wonder if orthography and phonology interact between languages in the bilingual lexical activation process. If so, then a model of nonselective bilingual access would emerge at the lexical level, strengthening monolingual research findings and suggesting properties of a general lexical activation model. Between-language orthographic and phonological effects in mixed-language tasks Van Heuven, Dijkstra and Grainger (1998) set out to investigate whether interlexical orthographic connections occur by manipulating neighborhood effects on the recognition of Dutch and English words in a mixed-language lexical decision task. If words activate similar orthography across languages, then a Dutch word bons would activate not only lexical candidates from Dutch, but also lexical candidates from English, such as bond or bins. Words were tested with varying numbers of neighbors: many neighbors in both English and Dutch, many neighbors in Dutch with few neighbors in English, many neighbors in English and few in Dutch, or few neighbors in either Dutch or English. L1 Dutch-L2 English participants saw blocks of letter strings containing only English words and pseudowords or only Dutch words and pseudowords. Participants were to decide if each letter string was a word of English for the English block, and a word of Dutch for the Dutch block. Results showed that bilingual participants responded more slowly to L2 English words when they had many Dutch neighbors. The RTs were not affected, however, for L1 Dutch words with many English neighbors. This suggests that in a mixedlanguage task, the dominant L1 language (Dutch) failed to be influenced by the neighborhood

25

size of English words, whereas the less dominant L2 English was influenced by the orthography of Dutch. These results mirror those of Carreiras et al. (1997) in which words with high frequency neighbors were responded to more slowly than words with low frequency neighbors. If we consider the less dominant L2 English as being subjectively “less frequent” for participants, then the less frequent language is slowed by the more frequent Dutch neighborhood. In the VanHeuven et al. (1998) study, participants saw words from both languages presented in isolation, albeit separated by blocks, and found no effects of L2 orthography on L1 word recognition. Other research has found, however, that when words from two languages are presented in a priming situation, orthographic similarity from the L2 can affect word recognition in the L1. Bijeljac-Babic, Biardeau, and Grainger (1997, Experiment 2) tested proficient FrenchEnglish bilinguals in a masked lexical decision task to see how English L2 neighborhood primes affected the processing of L1 French targets. Prime-target pairs were of two types: a French target preceded by a French orthographic neighbor (e.g., mien – miel meaning mine - honey) compared to a control (e.g., hier - miel meaning yesterday - honey) and a French target preceded by an English orthographic neighbor (mile - miel) versus a control (meet - miel). While French monolinguals showed an inhibition effect only for French words preceded by an orthographic French neighbor, French-English bilinguals had slower RTs for both French and English orthographic neighbors. This suggests again that similar orthography is activated across languages. Just as monolingual research has shown phonological influences in lexical processing, evidence from an earlier study indicates that cross-linguistic phonological overlap affects bilingual lexical access. Nas (1983) presented Dutch-English bilinguals with an English lexical decision experiment in which nonwords were manipulated. Half were pseudo-homophones that 26

were formed by changing Dutch words so that they were orthographically similar to English words (e.g., snay) yet phonetically similar to Dutch words (e.g. snee in Dutch, which is pronounced like the English snay). That is, these words did not look like Dutch words to participants, but their pronunciation was the same as an existing Dutch word. The remaining nonwords were existing Dutch words changed by one letter that followed legal spelling-sound mapping in Dutch (prusk). These words did not sound like existing Dutch or English words, but rather looked like a possible Dutch word. The Dutch-English bilinguals rejected the crosslinguistic phonologically manipulated words (snay) more slowly than the traditional nonwords, which led to the conclusion that the phonology of the L1 is activated during L2 lexical processing. Taken together, orthographic and phonological interactions occur between words in two languages for bilinguals, just as they do between words in one language for monolinguals, suggesting general interactive activation principles at the lexical level in the word identification process. Some research suggests that the less dominant L2 is affected by the more dominant L1 but not vice-versa (e.g., neighborhood frequency in Van Heuven et al., 1998; L1 phonology on L2 phonology in Nas, 1983). Other research shows that when two languages are presented, both languages impact the activation of the other (Bijeljac-Babic et al., 1997). The question arises, then, as to whether the same principles are at work at the lexical level when bilinguals are asked to operate in a language-specific mode. Between-language orthographic and phonological effects in language-specific tasks Van Heuven, Dijkstra and Grainger (1998) conducted an additional study in which L1 Dutch-L2 English participants carried out a lexical decision task again testing Dutch-English

27

neighborhood effects, but this time researchers removed the Dutch block of words from the experiment and presented words only in one language—English. Results replicated results of the Dutch-English block experiment: between-language neighborhood size of the L1 affected RTs of L2 English targets. English words were responded to more slowly when they had many Dutch neighbors. While these data confirm that interlingual lexical properties do interact even when materials are language-specific for bilinguals, the researchers acknowledged that orthographic manipulations alone cannot explain how lexical properties interact between languages because monolingual research has shown that orthography, phonology, and word frequency all interact during word identification. In order to tease apart the roles of orthography and phonology, and at the same time consider how these features interact with semantics in bilingual lexical activation, Dijkstra, Grainger, and Van Heuven (1999) tested L1 Dutch-L2 English bilinguals with L2 English words that varied in degrees of orthographic, phonological, and semantic overlap with Dutch in a lexical decision task (Experiment 2). Items either overlapped completely in semantics, orthography, and phonology (SOP) with English (hotel), in semantics and orthography (SO: type), in semantics and phonology (SP: news—nieuws), in orthography and phonology (OP: step), in orthography (O: stage), or in phonology (P: note). Dijkstra et al. found that when English words overlapped with Dutch in SOP, SO, and O conditions, the lexical decision responses were facilitated as compared to matched English control words. Words overlapping in the P condition produced inhibition in RTs when compared to controls. Words that matched in SP and OP conditions did not vary significantly from their controls. When errors were considered, participants showed significantly fewer errors in the SOP, SO, and O conditions as compared to their controls, but OP and P conditions produced more errors than did controls. No 28

difference in errors was found for the SP condition and controls. English monolinguals were tested on the same materials in an additional experiment, but did not show any effects of interlingual conditions on lexical decisions. It appears, then, that interlingual orthographic overlap facilitates word recognition, while phonological overlap slows word recognition. Likely a null result in the SP and OP conditions are observed because the facilitatory effects of orthography (and probably semantics) cancel out the inhibitory effects of phonology. Yet other research investigating cross-language phonological activation has found that overlapping phonology between languages facilitates lexical activation. Haigh and Jared (2004, 2007) tested the activation of phonological representations for French-English bilinguals. In Experiment 1, bilinguals performed an English lexical decision task (L2 of the bilinguals) in which English-French interlingual homophones (e.g., mow in English which sounds like mot meaning word in French) were compared with English controls (mop), matched in frequency, initial letter, and English neighborhood size to the homophones. Bilinguals responded more quickly and more accurately to homophones than to controls. Monolinguals in a separate experiment showed no difference in RTs between conditions. Additionally, in Experiment 3, English-French bilinguals dominant in English performed the same task. This was the first experiment to test L2 homophone effects in L1 reading, and results showed no difference in RTs between conditions. Finally, in Experiment 6, interlingual homographs and cognates were added to the item list (e.g., coin meaning corner in French and train, a French-English cognate). This time, French phonology was activated: English-French bilinguals responded to interlingual homophones more quickly than control words. For these bilinguals, L2 phonology did not influence L1 lexical activation unless French was made more salient in the item list.

29

An additional study on interlingual phonological overlap, this time by Menenti and Indefrey (2006), is important to discuss here because it tested the effects of L2 phonology on L1 lexical activation in a unique way. In a primed lexical decision task, L1 German-L2 Dutch bilinguals, whose L2 proficiency scores placed them into Dutch native speaker range, were presented with primes in the L2 Dutch (e.g., trein) that when translated to German, (zug— pronounced tsu:k) rhymed with the L2 Dutch target (boek). Researchers wanted to know whether the L2 prime ostensibly activates its L1 translation equivalent so that L1 phonology affects L2 target lexical activation. Participants responded to both the prime and the target in separate lexical decision tasks immediately following each other (500 ms between prime lexical decision and subsequent target lexical decision). Both primes and targets were recombined to create control conditions. Results showed that for pairs in which the Dutch prime’s translation to the L1 German rhymed with the L2 target, RTs were faster than control conditions. Dutch monolingual speakers were also tested on the materials and showed no such effects. This is one of the first studies to depart from a cross-language priming task in order to test L1 effects in a purely L2 context, a crucial next step in the understanding of how languages interact in the bilingual lexical activation process. The logic that motivated the Menenti and Indefrey research question (to see if the L1 is activated ostensibly through the L2, in this case via phonology) is important to the current project because a similar motivation guides the research questions found at the end of this chapter. The fact that this research successfully showed an activation of the L1 in a purely L2 context give strength to the research design and findings of the current project. The bilingual lexical activation research conducted within the context of one language consistently shows effects of the L1 on the activation of the L2 (e.g., neighborhood effects of L1 on L2 in Van Heuven et al., 1998; phonology effects of L1 on L2 in Haigh & Jared, 2004, 2007). 30

Phonological effects of the L2 on the activation of the L1 were found in Haigh and Jared (2004, 2007) if the L1 was made salient. However, Menenti and Indefrey (2006) found effects of L2 phonology on the activation of the L1 even when the L1 had to be activated ostensibly through the L2. Research by Dijkstra et al. (1999) sought to tease apart bilingual orthographic, phonological and semantic properties on lexical activation and found that when words of two languages overlapped in orthography, facilitation in word recognition occurred. When the overlap was with phonology, inhibition occurred, but when orthography and phonology both overlapped, null results were obtained. One logical explanation is that the facilitation effects of overlapping orthography combined with the inhibition effects of overlapping phonology cancel each other out. General conclusion on lexical connections In sum, it is clear that lexical levels (orthography and phonology) interact within and across languages during lexical access. Exactly how and when this interaction occurs is not always consistent in the research findings. Monolingual research results suggest that the lexical processor activates potential lexical candidates based on similarity in orthography and phonology before landing on the appropriate word. The research by Sears, Hino, and Lupker (1995) found that the identification of low frequency words with neighborhoods containing high frequency words was facilitated by that connection to high frequency words. Research by Carreiras, Perea, and Grainger (1997), however, found just the opposite: that the identification of a low frequency word that shares a neighborhood of high frequency words is inhibited by such a connection. If both sets of results hold true, one explanation put forth by Carreiras et al. might be that when lexical candidates’ orthographic-phonological mappings require discrete identification (e.g., distinguishing between nearly identical lexical items: alba (soul) and alta (tall) or processing an 31

orthographic combination that varies in pronunciation—deaf vs. dear or leaf), additional processing time is required. Bilingual research findings also have provided mixed results as to how orthographic and phonological properties across languages affect targeted lexical activation. Van Heuven, Dijkstra and Grainger (1998) found that the identification of L2 words mirrored the results of Carreiras, et al. (1997). The recognition of L2 words (less frequent by nature of being the L2) with many L1 orthographic neighbors (higher frequency words by nature of being the L1) was inhibited by the presence of the L1 neighbors. Word recognition was not slowed by the reverse. That is, the identification of L1 words was not affected by L2 neighbors. Dijkstra, Grainger, and Van Heuven (1999) sought to tease out additional cross-language lexical properties and their influence on bilingual word recognition by breaking words into varying degrees of overlap among orthography, phonology, and semantics. They found that when orthography overlapped on its own between languages, or in combination with phonology or semantics, facilitation in identifying L2 words occurred. When phonology alone was the overlapping factor, inhibition occurred. When semantics or orthography overlapped with phonology, a null effect emerged, probably as a result of facilitative orthographic (and semantic) effects cancelling out inhibitory effects of phonology. Unique work by Menenti and Indefrey (2006), which extended the research on bilingual lexical activation to a purely within-language context, found that L1 phonology not only is activated during lexical processing of the L2, but that it occurs ostensibly. Participants were presented with primes in the L2 that if translated to the L1, would rhyme with the L2 target. Even though participants were not asked to engage in such translation or to consult the L1 at all in identifying the L2 primes and targets, the identification of those primetarget pairs that shared rhyming with the L1 was facilitated. 32

Given the body of research described in this section, within- and between-language lexical activation occurs for both orthography and phonology. The following sections consider research on semantic activation in both the monolingual and bilingual domains. SEMANTIC CONNECTIONS Monolingual ambiguity resolution: the role of meaning dominance When semantics is considered in the monolingual literature, it is researched in terms of ambiguity and how it is that the lexical system responds to cases in which words of similar or exact orthography (homographs: bug meaning insect or spy device) and phonology (homophones: to, too, two) map to different meanings. In the case of homographs, Simpson (1981) conducted word-level research to determine whether all meanings or only the most dominant meaning of an identical orthographic form are automatically activated. In a primed lexical decision task, monolingual English participants first responded whether a string homograph prime was a word of English. Immediately after, participants responded to a letter string target that reflected different meanings of the prime. The prime was one of three types: an ambiguous homograph (bank), an unrelated word (calf), or a pseudoword. The prime was followed by one of three options: the homograph's dominant meaning (money), the homograph's subordinate meaning (river), or a pseudoword. Both of the targets related in meaning to the homograph prime (bank) were unrelated in meaning to the other prime (calf). Simpson predicted that if all meanings of an ambiguous word are activated when that word is encountered, then both the dominant and subordinate meanings of the homograph should be responded to equally as fast in the second lexical decision task when preceded by the homograph. If only the dominant meaning of a homograph is activated, then the target reflecting 33

the dominant meaning of the homograph would be responded to more quickly than the target of the subordinate meaning. Results showed that participants responded significantly more quickly to both targets when targets followed the homograph as compared to the unrelated word. However, participants responded more quickly to the dominant meaning target of a homograph than to the subordinate meaning target when either target followed the homograph. Simpson summarized that an ambiguous word activates both meanings, but its dominant meaning is available more quickly than its subordinate meaning. In order to tease out the time course of dominant and subordinate meaning activation, Simpson and Burgess (1985) carried out additional word-level research on homographs. Monolingual English participants carried out a primed, lexical decision task in which the timing of the presentation of meaning-related targets was progressively delayed after the presentation of an ambiguous prime. Participants saw an ambiguous word prime on a computer screen. The prime disappeared and then was followed by a letter string for a lexical decision. Simpson and Burgess manipulated the stimulus onset asynchrony (the time from which the prime is presented until the target is presented) between the prime and the target. Results indicated that dominant word meanings for homographs were facilitated as early as 16 ms and stayed active through the longest time tested—750 ms. Subordinate meanings were facilitated only as time onset increased, from no facilitation at 16 ms to near equal to dominant facilitation by 300 ms. This activation declined, however, and by 750 ms, the subordinate meaning was no longer activated. Simpson and Burgess concluded that lexical access for ambiguous words is exhaustive—that is, within a single word context, all meanings of an ambiguous word are activated. The rate of that activation, however, is determined by meaning dominance.

34

In more recent word-level research on homographs, researchers began to think about meaning ambiguity in terms of meaning relatedness in order to see how this notion affects lexical access. Azuma and Van Orden (1997) tested word meaning relatedness for homographs in a lexical decision task. Meaning relatedness was determined by having English native speakers provide meanings for various ambiguous words, such as bank. The meaning reported most often was determined to be the dominant meaning for the ambiguous word, with all other meanings considered subordinate. Reported meanings were then given to English native speakers in pairs to rate for relatedness on a 7-point scale. Words were then divided into categories of words with many meanings (low-related or high-related) or words with few meanings (low-related or highrelated). In a lexical decision task, words with few, low-related meanings were responded to more slowly than any other word type (e.g., words with many meanings, low-or high-related, or words with few meanings, high-related). However, the authors acknowledge that the norming procedure to produce words with multiple meanings took all meanings and then set each dominant meaning against a subordinate one for a relatedness rating. It did not consider the relatedness among all meanings all together for a given word. In other words, it could be that ambiguous words had many or few truly ambiguous meanings (e.g., bank as in money or river), or simply many or few polysemous meanings—one meaning with multiple senses of that meaning (e.g., hook with all meanings reflecting the notion of two connecting items). Hence, Rodd, Gaskell and Marslen-Wilson (2002) teased out ambiguity effects in terms of true ambiguity and polysemy and found that in a lexical decision task, words that truly are ambiguous are responded to more slowly than words with many senses. These findings suggest that polysemous words may share one core meaning, thus providing no competition among meanings in terms of lexical access. Distinct meanings of ambiguous words, however, may compete for

35

activation, thus lexical access is slowed for these words. The Rodd et al. findings are new in the monolingual literature involving word ambiguity/intralingual homographs, as previous research on ambiguity consistently resulted in facilitation in the lexical access of ambiguous words , even advancing the notion of the ‘ambiguity (facilitation) effect’ (e.g., Borowsky & Masson, 1996; Hino & Lupker, 1996). Klepousniotou (2002) also teased apart ambiguity from polysemy and the results mirror those of Rodd et al. The ‘ambiguity (facilitation) effect’ of two decades of research likely seems due to the failure of researchers to tease apart words with truly ambiguous meanings versus polysemous words with one core meaning and many senses. While polysemous words do enjoy facilitated recognition in the word recognition process, the ‘ambiguity effect’ is one of inhibition, not facilitation. Following the lead from research on homographs, Berent and Van Orden (2000) suspected that the role of phonology in homophone recognition might interact with whether the dominant or subordinate meaning of a homophone is intended. In a masked prime recognition task, participants saw either a dominant meaning homophone target (board), or the subordinate meaning (bored), followed by three possible masks: a pseudo-homophone of the target that was identical in pronunciation but not in spelling to the target (BORD), a graphemic mask matched in spelling similarity to the target, but not in pronunciation (BORK), and a control mask (PRIK) sharing no letters or phonemes with the target. For each trial, participants saw the target (board or bored) presented for 28 ms, immediately followed by the pseudoword mask (BORD, BORK, or PRIK), also presented for 28 ms. A masking pattern (XXXXXX) followed each trial. Participants then wrote down the target and mask that they perceived. Results were measured in correct responses for targets (pseudowords were never reported accurately) and indicated that when the dominant meaning of the homophone (board) was followed by a phonological match 36

(BORD), there were significantly fewer reporting errors than if the subordinate meaning of the homophone (bored) was followed by the phonological match (BORD). An error meant, for example, that participants reported having observed the dominant meaning homophone form (board) when in fact the subordinate meaning (bored) had been presented. In other words, when the lexical processor encounters the orthographic form of the dominant meaning of a homophone, it benefits from the presence of its repeated phonology, probably because its spelling is more strongly linked to the phonology than is the spelling from any competitors. Conversely, for subordinate meanings, the repetition of phonological form adds more competition to an already less frequent spelling-sound match. This research strengthens the notion that dominant word meanings and their orthographic mappings are activated particularly quickly in monolingual word recognition. To summarize, for monolinguals, research on homophones and homographs suggests that dominant meanings get activated before subordinate ones. The time course for meaning activation with homographs shows that the dominant meaning is activated first (by 16 ms), followed by both meanings (equally activated at 300 ms), with the activation of the subordinate meaning fading rather quickly thereafter (by 750 ms). In single word lexical decision tasks, truly ambiguous words take longer to activate, presumably because distinct, multiple meanings are being accessed, while polysemous words are responded to more quickly in comparison, suggesting a single core representational meaning for access. Research investigating bilingual lexical access also has relied upon lexically ambiguous words across languages in the form of interlingual homographs to determine if both meanings— in this case language-specific meanings—are activated simultaneously and automatically during lexical processing. Interlingual homographs are often compared to cognates and noncognates in 37

such research in order to tease apart lexical processing effects due to multiple meaning access from any effects due to interlingual form similarity. In the next sections, the literature is reviewed for interlingual ambiguity resolution when homographs are presented in mixedlanguage and language-specific tasks. Bilingual ambiguity resolution in mixed-language tasks One of the first studies to investigate interlingual form ambiguity in bilingual lexical access was conducted by Gerard and Scarborough (1989). These researchers asked whether dual meanings of interlingual homographs are activated by investigating repetition priming effects within English and between Spanish and English for cognates (actual), noncognates (perro meaning dog), and interlingual homographs (red meaning net; fin meaning end) in a classical priming task (Experiment 2). In this task, a block of items (the prime) is mixed with pseudowords and presented for lexical decision. Many minutes later, a second block of the same items (hence repetition priming) is responded to in the same way. English monolinguals and two groups of mostly L1 Spanish-L2 English (2 participants were L1 English) bilinguals were tested. The English group saw English only items (e.g., actual-actual, dog-dog, red-red), while one bilingual group saw English primes-Spanish targets (e.g., actual-actual, dog-perro, red-red) and the other group saw Spanish primes-English targets (e.g., actual-actual, perro-dog, red-red). Interlingual homographs also were divided by language frequency, so that red in English is more frequent than fin in English. Results showed a priming effect in the monolingual and bilingual conditions for cognates and homographs, but not for noncognates. That is, identical form priming occurred, but translation priming did not occur. Additionally, monolinguals and bilinguals who saw English targets were faster to recognize high frequency words in English (red) than low frequency words in English (fin). When bilinguals saw targets in Spanish, the 38

opposite pattern emerged. Now words with high frequency in Spanish, such as fin, were responded to more quickly than homographs with less frequent Spanish readings (red). Only frequency effects of homographs in the target language slowed RTs for bilinguals. Non-target frequency of word forms did not affect RTs. Gerard and Scarborough interpreted the results in support of a selective bilingual processing view in which only one language meaning is activated. They argued that interlingual homographs should have caused interference for bilinguals in deciding their legality as words because of the disparity in meaning of these words. Researchers criticized many aspects of the Gerard and Scarborough study. First, the priming technique allows primes and targets to be viewed consciously by participants. When primes and targets are consciously available, it is unclear whether resulting effects are due to automatic processing mechanisms, or if they are due to post-lexical processing. Participants may integrate prime and target meanings together before responding to the target or may recall a prime from an episodic memory trace. The viewing of a target could reactivate the trace, thus showing an episodic memory effect, but not an effect of automatic lexical processing. Second, because the homographs were of exact orthography between languages, it would be unclear which language was being activated. What was meant to be cross-language repetition priming was indistinguishable from within-language repetition priming. De Groot and Nas (1991) tackled both problems by comparing unmasked priming and masked priming techniques on L1 Dutch-L2 English bilinguals for items of similar orthography between languages. Cognates and noncognates were manipulated with repeated, associative and unrelated priming between Dutch and English in four conditions: English-English (EE), EnglishDutch (ED), Dutch-English (DE), and Dutch-Dutch (DD) (e.g., EE repeated: ground-ground, associated: calf-cow, unrelated: bride-task; ED or DE repeated: grond-ground, associated: 39

kalf/calf-cow/koe, unrelated bruid/bride-task/taak; and DD repeated: grond-grond, associated: kalf-koe, unrelated: bruid-taak). Experiment 1 presented the prime-targets in a primed lexical decision task, but shortened the prime duration in order to try to avoid post-lexical strategies that previous research techniques may have encouraged. Participants saw a fixation mark (*) for 1000 ms, followed by the prime for 200 ms, then a blank inter-stimulus interval (ISI) for 40 ms, before the target appeared. The prime-target SOA was 240 ms, presumably too short for strategic processing to occur. Results showed that responses were significantly faster overall for targets in the L1 Dutch (DD: 515 ms, ED: 521 ms) than in the L2 English target conditions (EE: 552 ms or DE: 584 ms). Repeated targets (a 98 ms effect) were responded to more quickly than associated targets (a 55 ms effect) in comparison to unrelated targets. Associative-priming both within-and between-languages was equally significant, but repetition priming within language was stronger than between language priming. De Groot and Nas’s Experiment 2 tested the same materials in a masked prime lexical decision task. In general, the masked priming paradigm (the current technique made popular by Forster and Davis, 1984) presents a very brief prime (usually for 40 to 60 ms) surrounded by a forward mask (######) and target, both of which are presented for longer amounts of time (about 500 ms) (e.g., ###### nurse DOCTOR). Usually, the prime is presented in lower case script while the target is in uppercase so that the orthography of the prime and target do not spill into each other. Results showed that while the order of means was the same as Experiment 1, only the fastest DD condition (506 ms) was significantly different from the slowest DE condition (565 ms). Repeated targets (a 63 ms effect) again were responded to more quickly than associated targets (a 37 ms effect) in comparison to unrelated targets. Repetition priming within language was stronger than between language priming, and this time, associative priming effects 40

were larger when the target language was the L1 Dutch (DD and ED) than when it was the L2 English (EE and DE). In additional experiments, De Groot and Nas (1991) compared cognates from conditions DE and EE only with noncognates and found that overall in the unmasked experiment, responding in the EE condition was faster than the DE condition. Responding to cognates was faster than to noncognates. A repetition effect between languages was larger within than between languages, but was smaller for cognates (68 ms) than for noncognates (113 ms). The masked prime data showed that EE condition responses were faster than DE responses and responses were slightly faster for cognates than for noncognates, but not significantly so in either case. Associative priming effects were equally as large for cognates and noncognates, but repetition effects were faster now for cognates than for noncognates, with both cognate and noncognate effects reaching significance. The important finding in this research is that associative and repetition priming is found in the masked priming task for cognates and noncognates between languages, whereas they were not found for noncognates in previous research utilizing classical priming (e.g., Gerard and Scarborough, 1989). These experiments and many others (e.g., De Groot, Delmar, and Lupker, 2000; Dijkstra, De Bruijn, Schriefers and Ten Brinke, 2000; Dijkstra, Van Jaarsveld, and Ten Brinke, 1998; Gollan, Forster, and Frost, 1997) suggest that the semantics from both languages are tapped when two languages are presented in a task, as long as materials are presented in such a way as to eliminate post-lexical activation processes. The question again returns to language-specific tasks and whether semantics in both languages are activated when a task is language-specific.

41

Bilingual ambiguity resolution in language-specific tasks Dijkstra, Van Jaarsveld, and Ten Brinke (1998, Experiment 1) argued, among other things, that language intermixing in tasks may affect bilingual lexical access. Hence, Dijkstra et al. asked L1 Dutch-L2 English bilinguals to perform an English only lexical decision task on interlingual homographs (e.g., list meaning trick or guile in Dutch), cognates (e.g., hotel) and on English control words. Cognates were divided into two frequency groups: high frequency English-high frequency Dutch (HFE-HFD) and low frequency English-low frequency Dutch (LFE-LFD). Homographs were broken down into four groups: HFE-HFD, HFE-LFD, LFEHFD, LFE-LFD. Overall, the RTs for interlingual homographs relative to English control words were not statistically different, a finding reminiscent of the Gerard and Scarborough (1989) research. However, for cognates, a facilitation effect was observed when compared to English control words. Dijkstra et al. interpreted the cognate findings in support of nonselective bilingual access with the logic that because a cognate represents readings from two languages, faster responses for cognates over control items indicate that both languages contributed to the activation of the cognate, therefore facilitating its activation. The null-effect for interlingual homographs was in fact puzzling for the authors, who tried to explain the data in many ways other than evidence for language selective access for homographs. In additional language-specific research by De Groot et al. (2000, Experiment 2), highly proficient L1 Dutch-L2 English bilinguals completed a lexical decision task in which they responded to an all-English or an all-Dutch letter string list of homographs, controls, and language-appropriate legal nonwords. Homographs were divided into high frequency Dutch/low frequency English and low frequency English/high frequency Dutch. Participants were not aware that interlingual homographs were mixed into the lists. Results indicated that when Dutch 42

lists (the L1 of participants) were responded to, low-frequency homographs with a higher frequency reading in L2 English were responded to more slowly and less accurately than highfrequency homographs in L1 Dutch (low frequency homographs in L2 English). In the L2 English condition, word type and frequency were never statistically significant showing a null effect for homographic words, although low frequency English words (with high frequency Dutch reading) had a 22 ms tendency toward facilitation as compared to the control. The authors explain their results by suggesting that the bilingual processing system can either be selective or nonselective, depending on the target language of the task. De Groot et al. point out that their own data is opposite of what they would have expected. That is, an inhibitory effect was found for homographs in condition Dutch, the L1 and presumably stronger and more highly activated language of these participants. The effect was not found in the English-only condition where it would have been expected. De Groot et al. posture that it should be easier to block out the English readings of the homographs (the weaker language) in the Dutch-only condition and not as easy to block out the Dutch readings (the stronger language) during the English-only condition. De Groot et al. offer the explanation that participants perhaps treated the experiment not as a language-specific task, but as a language-neutral task. If participants performed the task in a language-neutral fashion (meaning that the processor would identify a word as belonging to any language), the availability of any meaning of the homograph (from either Dutch or English) would cause an affirmative and quicker response. De Groot et al. predict that these effects would cancel out any inhibitory effects that some homographs may have caused if participants performed the task as instructed, as a language-specific task. Their conclusion is that bilingual language processing is nonselective, but this can fail to emerge when a combination of processing modes are adopted by participants.

43

Both the Dijkstra et al. (1998) and the De Groot et al. (2000) studies frame their results as evidence that homographic meanings for the nontarget language were not accessed when presented in a language-specific task. However, the monolingual literature shows that dominant meanings both for homographs and homophones enjoy stronger lexical connections and quicker activation than subordinate meanings, and that the activation of a homographic form with two very different meanings is inhibited due to the competition between meanings (Rodd et al., 2002; Klepousniotou, 2002). If we apply this knowledge to the De Groot et al. data and suggest that the bilingual lexical parser uses the same information in a ‘language blind’ manner, a different explanation suggests itself regarding whether both languages were activated in the De Groot et al. data. L1 Dutch-L2 English bilinguals responded in Dutch only and English only tasks. In the L1 Dutch task, low-frequency homographs (with a high frequency reading in L2 English) were responded to more slowly and less accurately than high-frequency homographs in L1 Dutch (low frequency homographs in L2 English). The strength of a high-frequency L2 English meaning attached to the interlingual homographic form makes this meaning emerge to a level that it can compete with the L1 low-frequency meaning. This causes inhibition, just as in the case of the monolingual data. Likewise, a low-frequency meaning coming from a second and often less dominant language may not be strong enough to compete with a dominant L1 Dutch meaning when Dutch is called upon for a task. In the L2 English condition, De Groot et al. found that low frequency L2 English words (with high frequency L1 Dutch readings) had a 22 ms tendency toward facilitation as compared to the control. In this case, one could speculate that the strength of the L1 meaning would make it the dominant meaning for the homograph, not only because it is of high frequency, but also because it comes from the often stronger L1. This type of ‘double dominance’ per se would 44

ensure that the dominant meaning of the homograph would be activated. This meaning dominance could boost the activation of the less dominant L2 meaning, thus creating facilitation in the activation of the homographic form, even though research shows that two different meanings create inhibited activation of a lexical form. Perhaps in this case, the frequency of a dominant form from the dominant L1 is strong enough or dominant enough that it works against any inhibition that would emerge from competing meanings. This activation could be enough to push the recognition of the homographic form toward facilitation. Of course, what is not known from these data and materials is how the homographs break down into truly ambiguous homographs or partial homographs, or how bilinguals subjectively interpret the dominance of the meanings in their own lexical system, or even how proficient and strong the L2 is in these bilinguals. Can the same logic be applied to the Dijkstra et al. (1998) data where the L2 English was presented to L1 Dutch speakers? If so, one might expect null effects to emerge in a number of the cases (i.e., LFE-LFD; HFE-HFD; HFE-LFD) due to the fact that any boost in activation from meanings attached to the dominant L1 may be cancelled out by the competition of a different meaning in the L2, the language of the task. Data for these conditions had null results: HFEHFD 548 ms/Control 554 ms; LFE-LFD 620 ms/Control 627 ms; HFE-LFD 548 ms/Control 556 ms. Where a result may be visible is in the case of LFE-HFD words, as was seen in the De Groot et al. data in the form of slight facilitation, albeit not statistically significant. In the Dijkstra et al. study, however, the LFE-HFD condition showed inhibition: LFE-HFD 609 ms/Control 558 ms. Perhaps the relative frequency and dominance between what is a low frequency English meaning and a high frequency Dutch meaning in the Dijkstra et al. study is

45

different from the De Groot et al. study such that in the former case, the Dutch L1 meaning activates only to the threshold of causing competition between L2-L1 meanings. A picture that seems to emerge when all data are considered from both monolingual and bilingual literature, is that the bilingual lexical system is not one in which items can be grouped together in order to binarily decide if a nontarget language has been activated. Most likely, as an L2 is learned, new meanings are added to existing forms, dominant and subordinate meanings are created, meanings are fully or partially shared or truly ambiguous, and the L1 varies in terms of how dominant it is over the L2. Through an inter-connected lexical system with the L1, the bilingual lexical processing system makes use of all of this information. Yet, because the interactions become more complex with the addition of an L2, the empirical support for such a system can be difficult to tease out. A very early study in which homographs were investigated may actually support this notion. Beauvillain and Grainger (1987, Experiment 2) tested how the frequency of languagespecific readings of homographs affects bilingual lexical access. Homographs were divided into high or low frequency French (HFF/LFF) readings and high or low frequency English (HFE/LFE) readings, each followed by a target to reflect the respective homograph reading (e.g., HFE four (=oven) followed by five, HFF pain (=bread) followed by beurre (butter), LFE pain followed by ache, LFF four followed by cuisine (=kitchen). One group of L1 English-L2 French bilinguals was told that the primes were in French, while another group was told that the primes were in English. An SOA presentation of 150 ms was tested. Results showed no effect for language mode (whether participants had been told to read the primes in French or English), but an effect for frequency was found. High frequency homograph prime readings produced a facilitatory response whether the target was contextually appropriate or inappropriate, whereas a 46

low frequency reading of the homograph prime did not affect lexical access in either case. This finding is very much like that found in the monolingual research where primed dominant meanings of homographs are responded to more quickly than subordinate primed meanings. Taken together, findings from the monolingual and bilingual semantic ambiguity literature might allow a slightly different interpretation of bilingual data—not in an ‘either-or’ way by asking whether the nontarget language was activated, but by utilizing monolingual research findings on meaning dominance, true ambiguity, and polysemy to show that not only are both languages activated, their lexicons are intertwined at all levels. Obviously, the strongest evidence for dual language activation comes from experiments in which both languages are presented. However, as Grosjean (1997) aptly pointed out, many methodological issues may shade experimental results. While material presentation in two languages strongly supports a nonselective view of language access for bilinguals, many researchers have had to explain null-results from language-specific materials in creative ways. More research is needed in which language-specific materials are manipulated. De Groot et al. (2000) suggested that a language neutral strategy might be responsible for mixed results in a language specific context because homographs in their study and in Dijkstra et al. (1998) were chosen with exact orthography between languages (e.g., red, fin), potentially allowing a lexical reading in whichever language first becomes available to bilinguals. It has been shown that bilingual lexical items exist on a continuum of meaning (and form) overlap—from no meaning overlap to partial meaning overlap to nearly complete semantic overlap (Van Hell and De Groot, 1998). Yet research continues to base notions of the bilingual lexicon on short, 4-letter nouns, presented binarily as ‘all or nothing’ in terms of meaning-overlap. Van Hell (2002) indicated

47

that research needs to be extended to other grammatical classes, away from 4-letter word nouns. By extending bilingual lexical research to verbs, some of these issues can be addressed. CONCLUSION, RESEARCH QUESTIONS AND PREDICTIONS Filling in the gaps: testing bilingual lexical activation with verbs This chapter reviewed the following important points relevant to the current research design. In monolingual research, lexical decision tasks have been utilized to demonstrate that multiple meanings for ambiguous words are activated. In a priming paradigm where a one word context biases meaning, this activation is exhaustive. That is, dominant meanings are activated quickly (by 16 ms), followed by the activation of multiple meanings (by 100 ms), with a return to the sole activation of a dominant meaning (by 750 ms). If words are truly ambiguous (their meanings are very different), activation is slowed due to the competition of two very different meanings mapped onto the same form. If words are polysemous so that they have one central core meaning used in a variety of senses, lexical access is speeded, presumably because multiple senses of the same core meaning boost activation of the lexical form. In the bilingual literature, when two languages are presented at once, the bilingual parser automatically activates multiple meanings of words that are ambiguous across languages, a phenomenon referred to as nonselective bilingual lexical processing. Both cross-language repetition (translation) priming and associative priming have been found when a paradigm is used that does not allow the conscious recognition of primes. Frequency between languages is believed to play a role in bilingual lexical access, as does form overlap in orthography and phonology.

48

What is not as clear from research data, however, is how the bilingual parser operates under language-specific conditions in terms of nonselective language activation. Research has shown that words overlapping in form and meaning between languages (cognates) tend to produce a facilitation effect in lexical access when participants perform lexical decision tasks in the L2, presumably because the shared features of these words in the lexicon speed up lexical activation. VanHell and Dijkstra (2002) tested the cognate effect of the L2 during processing of the L1. The study utilized Dutch-English-French trilinguals to see if cognate nouns between English and Dutch and French and Dutch were recognized more quickly by Dutch speakers conducting a lexical decision task in their native language. In response to avoiding a ‘language neutral’ processing strategy, a point raised by De Groot (2000), researchers tried to avoid utilizing nouns of exact orthography between languages. VanHell and Dijkstra found a facilitation effect for cognates as compared to noncognates of both the L2 (English) and the L3 (French) on the L1 (Dutch) of the trilinguals but only if participants were of higher proficiency in the second and third languages. However, 6 out of 20 of the Dutch-French cognates (e.g., gazon/lawn) and 3 out of the 20 Dutch-English cognates (e.g., ring) still shared exact orthography, thus not eliminating completely the potential for a language neutral processing strategy. Naming data also have shown that cognate naming in the L1 or the L2 produces a facilitative cognate effect, but the effect is much greater when participants name in the L2 (the less dominant language) (e.g., Costa, Caramazza, and Sebastian-Galles, 2000). Words that overlap in form but not meaning (interlingual homographs) have produced mixed results when framed in terms of whether the nontarget language has been activated as a whole. While some research has shown inhibition for such words when homographs are presented in an L1 context (albeit surprising to researchers), a null effect has been found when homographs are presented in

49

the L2. Researchers have tried to explain creatively why nonselectivity can be found sometimes for homographs (multiple meanings creating inhibition in lexical access), but not other times. Again, DeGroot (2000) suggests that because homographs in these studies are of exact orthography between languages, a neutral language strategy may be driving lexical decisions. That is, participants may simply see a form like red and respond that it is a word—in any lexicon available. While this may be true, an additional suggestion by this author is that results are being interpreted incorrectly. These unknowns as to how language-specific factors influence cross-language nonselectivity can be addressed if the following is considered: 1) research continues to investigate bilingual lexical activation in a language-specific context, 2) homographs are utilized that are not of exact lexical representation between languages, but that provide cues of language specificity, 3) words can be found that show a continuum of meaning overlap between languages, 4) research can be extended beyond 4-letter nouns to other grammatical classes, and 5) bilingual lexical access is investigated in terms of meaning dominance and subordinance. Because of the nature of language-specific verb infinitives and conjugations, exact orthographic homographs will not be found between languages (e.g., decidir/decide). Previous research indicates that bilingual lexical access is affected by orthographic neighborhoods, so there is justification that similar form representation between languages can be used to test bilingual lexical access. In addition, verbs can restrict and extend research to a new grammatical class. There is some indication that verbs may be processed differently than nouns. Nouns often express a concrete notion, an object that is quickly envisioned or drawn in pictures. Verbs, however, are more abstract in their meanings, often difficult to depict visually, and are likely to show a continuum of meaning overlap between languages. 50

The current research project utilizes Spanish verbs in an original study to provide clues as to how language-specificity affects cross-language activation. First, is nonselectivity replicated in the processing of interlingually ambiguous verbs between Spanish and English? If so, do monolingual research findings related to the role of meaning dominance and ambiguity help to explain the nature of cross-language activation? That is, does interlingual meaning dominance play a role in how the bilingual parser is nonselective? The project is divided into two experiment design groups. First, a series of extensive studies identifies a set of interlingual cognate, homographic, and noncognate verbs between Spanish and English. Second, these items are manipulated in an online lexical decision task to answer the following questions: 1. Is there evidence of interlingual lexical and semantic activation when L1 Spanish –L2 English bilinguals process interlingually ambiguous verbs in an L1 Spanish languagespecific task? 2. Is there a convergence of findings between L1 Spanish bilinguals and Spanish monolinguals in the processing of ambiguity? 3. Is there evidence of interlingual lexical and semantic activation when L1 English –L2 Spanish bilinguals process ambiguous verbs in an L2 Spanish language-specific task? Predictions Homographs: For homonyms to show nonselectivity, form overlap may facilitate RTs while semantic incongruency between languages should slow RTs. These two effects may cancel each other out so that no effect is found for homographs as compared to controls. For

51

ambiguous homographs, nonselectivity may emerge as a null result due to facilitation from form versus inhibition from semantic incongruency. However, when ambiguous homographs are broken down based on whether the dominant or subordinate Spanish verb meaning overlaps with English, different results may emerge. Following the same logic that produces a cognate effect, a dominant meaning overlapped with English may produce a facilitation effect for bilinguals: the form and semantic overlap of the dominant meaning with the English meaning could boost activation of the word to the point of a facilitation effect. When the subordinate Spanish meaning overlaps with the English meaning, a null result could emerge. The dominant Spanish meaning could now compete with the boosted English/subordinate Spanish meaning to the point of competition. While this competition may not be enough to produce inhibition, it should be enough to reduce facilitation of form overlap. Cognates: Cognates are included in this research to see if a cognate facilitation effect is found in verb processing. Prior research has shown a strong cognate facilitation effect in bilingual processing of the L1 onto the L2, but a weaker or perhaps questionable cognate facilitation effect of the L2 (or L3) onto the L1 (e.g., Costa et. al., 2000; VanHell & Dijkstra, 2002). The L2 participant data are expected to show a cognate facilitation effect in the current research due to form and semantic overlap. It is less clear if the L1 participant data will show an equally strong effect, a weaker effect (as with the naming data) or perhaps even no effect (if the VanHell and Dijkstra materials jaundiced findings due to the fact that some items still were of exact orthography between languages). In order to show that materials are reliable, monolinguals should show no difference in RTs for cognates, homographs, and controls or noncognates. Based on monolingual research, if ambiguous partial homographs are of very different meanings, an inhibition effect may emerge. 52

CHAPTER 3: IDENTIFYING SPANISH VERBS AS COGNATES AND HOMOGRAPHS WITH ENGLISH

OVERVIEW OF THE NORMING STUDY The field of bilingualism takes an interest in interlingual cognates and homographs for pedagogical and research reasons. Cognates are words that overlap in orthography and meaning, with varying degrees of difference in pronunciation, between two languages (e.g., piano or train in Spanish and English). Interlingual homographs, on the other hand, are words that share form but not meaning (e.g., the word red in English means net or web in Spanish). Scholars have given interlingual homographs a variety of labels such as interlexical homographs (De Groot, Delmar, and Lupker, 2000), false cognates (Brysbaert, 1998; Gerard and Scarborough, 1989; Grainger, 1993), and false friends (Meara, 1993). Second and foreign language textbooks make early use of cognates between languages to build L2 vocabulary in a short period of time. While research shows that L2 learners acquire cognates more easily than other types of words (e.g., DeGroot and Keijzer, 2000), learners can be fooled by the disparate meanings of homographs and misguided in the pronunciation of cognates due to interference from differing L1 pronunciation (Jacobs, 2007). Psycholinguistic research in bilingualism employs cognates and interlingual homographs to ask primarily two questions: (1) Whether bilingual lexical access is nonselective—that is, whether both languages are automatically activated during lexical selection in one language alone— and (2) What factors constrain lexical activation when bilinguals intend to use only one language. The logic behind manipulating the cognate status of words is that if the language

53

processor activates both languages at once, then words sharing form and meaning (cognates— piano) should be processed more quickly than words that do not share form (noncognates—silla meaning chair), either because such words are stored together in memory or because the form/meaning overlap allows for stronger and faster lexical activation. Conversely, the meaning disparity of homographs (red meaning net) should slow the processor due to momentary confusion through activation of similar forms yet incongruent meanings. In fact, researchers overwhelmingly have found a time difference in the recognition of interlingual cognates (facilitation) and homographs (inhibition) as compared to noncognates, which has led them to conclude that the bilingual lexical processor is nonselective (e.g., Christoffels, Firk, and Schiller, 2007; Costa, Caramazza, and Sebastian-Gallés, 2000; Dijkstra and VanHeuven, 2002)—even in cases where context strongly biases a particular language (e.g., Elston-Güttler, 2000; Schwartz and Kroll, 2006; Duyck; Van Assche, and Drieghe, Hartsiuker, 2007; Van Assche, 2009; Van Assche, Duyck, Hartsuiker, and Diependaele, 2009). Additionally, cognates and homographs can be utilized in research to inform of the general mechanisms of language processing. Much psycholinguistic research conducted with monolinguals has sought to explain whether words are accessed in the mind via bottom-up (form) or top-down (meaning) processing strategies. However, the majority of the world’s speakers are bilingual (Romaine, 1995) and therefore any theory about how language works in the mind must also account for how bilinguals activate and process multiple languages. By manipulating words that overlap in form and meaning between languages, researchers are able to inform general theories on language processing. For example, many researchers have investigated how ambiguity is processed within English—that is, whether we access simultaneously all meanings of a word like bug (i.e., an insect, a listening device, to bother) or if 54

we activate only the meaning necessary for the immediate context (e.g., Azuma and Van Orden, 1997; Borowsky and Masson, 1996; Klepousniotou, 2002; Millis and Button, 1989; Rodd, Gaskell and Marslen-Wilson, 2002; Simpson and Burgess, 1985). By extending the same curiosity to how bilinguals handle words that are ambiguous between languages (homographs), researchers can focus on broader language processing questions: not whether monolinguals activate multiple meanings of ambiguous words like bug or whether bilinguals activate the meanings in both languages of a homograph such as red but rather, how the language processor in general handles ambiguity. Ultimately, then, there are many reasons for focusing on cognates and interlingual homographs in research and yet the existing body of research relying on such words is limited to predominantly noun-noun translations between languages or mixed word-class translations, such as with the example of red (the color adjective in English vs. the noun web or net in Spanish). There is reason to believe that by expanding research to cognate and homographic verbs, researchers may be able to gain additional insight into the nuances of bilingual and general language processing. In research relying on homographs to test whether the bilingual processor is nonselective in a monolingual context (DeGroot et al., 2000; Dijkstra et al., 1998), the findings have been mixed. DeGroot (2000) suggested that because homographs in these studies were of exact orthography between languages, a neutral language strategy may have driven lexical decisions. That is, participants may simply have seen a form like red and responded that it is a word—in any lexicon available. The norming of homographic and cognate verbs builds a bank of testing items that eliminates this possible processing strategy. Due to the morphological nature of verb endings (in this case between Spanish and English), verbs will never overlap exactly in form (dividir—to divide), thus eliminating a language-neutral processing strategy due 55

to exact form overlap. In addition, verbs are more complex than the short, one-syllable nouns that have dominated language processing research in terms of word length, abstract meaning, and the role that verbs play in sentence construction. Since many of the computational models of language processing (e.g., the Bilingual Interactive Activation Model (BIA) and BIA+, Dijkstra and VanHeuven, 1998, 2002) are built to accommodate empirical findings based on short, onesyllable words, the opportunity exists to broaden such modeling to include less concise data: normed verbs that vary in length and number of syllables. In addition, verbs are often more abstract in meaning than nouns, thus providing testing items with degrees of overlap in meaning and ultimately, insight into the general nature of processing abstract meaning. Finally, through extensive norming of verb meanings, a set of items becomes available for use in future research that extends to the sentence and discourse level, relevant because of linguistic theory that suggests that verbs drive the construct of both syntax and meaning in sentences (Levin, 1993; Levin and Rappaport, 1994, 1995). How researchers derive their lists of cognates and homographs, however, has not been consistent (Grosjean 1997, 1998; Friel and Kennison, 2001). In the case of cognates, some researchers have defined them to have the same original word root (e.g., Sánchez-Casas, 1992) while others have relied upon empirical testing to determine cognate status (De Groot and Nas, 1991; Friel and Kennison, 2001; Kroll and Stewart, 1994). In their extensive work with interlingual homographs, Dijkstra and colleagues defined homographs simply as words identical in orthography but not meaning between two languages (Dijkstra, 2005). Conversely, Friel and Kennison (2001) normed German-English cognates and homographs by comparing versions of two empirical approaches used previously by researchers. One approach asked bilinguals to rate provided translations on a scale of one to seven (De Groot and Nas, 1991) while another (Kroll 56

and Stewart, 1994) relied upon native English speakers with no knowledge of Dutch or German to translate a list of Dutch words into English. Friel and Kennison found no significant differences in the two norming approaches. In the current study, in order to define verbs in Spanish that are cognates or homographs to English, an empirical approach was used. Tasks were devised first to compare dictionary definitions with actual language use. Both native speakers of English and Spanish contributed toward determining the degree to which form and meaning overlapped and diverged in Spanish verbs and which synonyms best convey those meanings. The norming resulted in three categories of verbs defined in the following ways. Cognates are words of near complete form and complete meaning overlap between Spanish and English (calmar—to calm). In homographs, form overlaps between languages, but meaning does not (e.g., estrechar looks like to stretch but means to make narrow, to tighten). Partial homographs are verbs that are ambiguous (with two distinct meanings) in Spanish, only one of which is shared with English (e.g., experimentar-experiment shares the meaning of to experiment, while Spanish has the additional meanings of to experience or to feel). The phonology differs between Spanish and English for all three verb types. SELECTION AND NORMING PROCESS OF VERBS Method The goal of the norming research was twofold: 1) to generate a list of verbs in Spanish to be classified as cognates or homographs with English and 2) to identify the verb synonyms in Spanish that most accurately convey the meaning(s) of the cognates and homographs. To this end, two characteristics of the verbs needed to be assessed: form similarity and meaning 57

similarity. The norming study was carried out in three phases, explained in detail in the Materials and Procedures section. Phase I involved the dictionary selection of potential Spanish-English verb cognates and homographs to be used in the collection of norming data of Phases II and III. In Phase II, the meanings of the target verbs were determined by functionally monolingual Spanish speakers via three tasks: a synonym-solicitation task for cognates, a synonym-solicitation task for homographs, and a synonym-clarification task. Finally, the norming data from Phase II were used to create five total tasks for Phase III. To measure the strength of synonyms in each language, two Spanish synonym-rating tasks for cognates and homographs were completed by the Spanish speakers and two English synonym-rating tasks for cognates and homographs were completed by the English speakers. To measure the degree of form overlap between Spanish and English for the target verbs, the English speakers also completed one form similarity rating task. Participants A total of 186 participants—163 functionally monolingual Spanish speakers and 19 functionally monolingual English speakers completed the eight norming tasks. The 163 Spanish participants were undergraduate psychology students from the University of Granada, Granada, Spain. Their average age was 21 years and they reported having spent 9 years on average studying English in school, with fewer than 6% of participants having spent any time abroad studying English, and only for an average of two months.1 These participants completed the five

1

Even though the UG students show some exposure to English, they were interviewed briefly in English and could not engage in conversation. This participant population readily participates as Spanish monolinguals in research carried out in the psychology department at the University of Granada. Given the nature of the Spanish educational system and the infiltration of English in world music and other cultural interests, it is impossible to find a population that has not been exposed to English. For the purposes of this study, then, the Spanish participants shall be called functionally monolingual. At best, they are Spanish monolinguals with exposure to English that does not translate into English proficiency.

58

norming tasks conducted in Spanish from Phases II and III (approximately 32 participants per task). The participants received course credit for their participation. The 19 functionally monolingual speakers of English were undergraduate students from the University of Illinois, Urbana-Champaign. They volunteered participation in the three norming tasks in English that were carried out in Phase III. The average age of this group of participants was 19 years. All English monolinguals were students in a Spanish 101 course at UIUC, a course for students who have never studied Spanish. Materials and Procedure Phase I: Selection of potential cognate and homographic verbs In order to identify cognate and homographic verbs between Spanish and English, verb lists were generated from false cognate (homograph)2 and cognate dictionaries.3 Each verb was then searched in an unabridged Spanish-English dictionary4, in monolingual synonym/antonym dictionaries for Spanish and English5, and in a monolingual English dictionary6 to gather information about the meanings of each verb and whether meanings overlapped with the translation of the target verb to an English verb similar in form. This procedure generated a list of 76 potential homographic verbs and 76 potential cognates. Homograph candidates were chosen first, followed by cognate candidates that matched homographs in frequency (based on the frequency of the infinitival form only) as found in the Alameda and Cuetos frequency

2

Hamel, B. (1998). Comprehensive Bilingual Dictionary of Spanish False Cognates. Bilingual Book Press, Los Angeles, CA. Prado, M. (1993). NTC's Dictionary of Spanish False Cognates. NTC Publishing Group, Chicago, IL. 3 Nash, R. (1997). NTC's Dictionary of Spanish Cognates. NTC Publishing Group, Chicago, IL. 4 Harper Collins Spanish Unabridged Dictionary, Sixth Edition. (2000). Harper Collins, NY, NY. 5 Diccionario Práctico Larousse Sinónimos/Antónimos. (1986). Ediciones Larousse, Marsella, México. The Double Day Roget's Thesaurus in Dictionary Form. (1987). Doubleday, NY, NY. 6 Miriam-Webster Online Dictionary. http://www.m-w.com/

59

dictionary7, which bases frequency numbers on 2,000,000 words. The lists of potential homographic and cognate target verbs, their definitions, frequencies, translations to an English verb most similar in form to Spanish, and the English verb form meanings are provided in Appendices A and B. Phase II: Solicitation and clarification of verb meaning Two synonym-solicitation and one meaning clarification tasks were created to allow for a comparison between dictionary definitions of potential target verbs obtained during materials selection and actual Spanish monolingual usage of these verbs. The focus on synonym solicitation to express meaning was desired for two reasons. First, single verb synonyms as the expression of meanings for target verbs would be the most efficient way to express meaning in an eventual off-line task to accompany the online research. Second, the establishment of strong synonyms for target verbs allows for planned future research with the target verbs. Synonym-solicitation task for homographs and cognates For the synonym-solicitation task for homographs, 30 functionally monolingual Spanish speakers were asked to generate verb synonyms in Spanish for 63 of the 76 homographic verbs identified in Phase I8. Similarly, in the synonym-solicitation task for cognates, 39 functionally monolingual Spanish speakers were asked to generate synonyms for 63 of the 76 cognate verbs identified in Phase I. Both tasks were carried out in a classroom setting during an hour-long experimental session in which participants signed a research consent

7

Alamedo, J.R. and Cuetos, F. (1995). Diccionario de frecuencias de las unidades linguísticas de castellano. Oviedo. Servicio de Publicaciones. Oviedo. 8 Norming tasks in Phases II and III do not utilize all items from Phase I. This is due in part to inadvertent error in leaving out some potential verb candidates and/or due to the fact that as each norming phase progressed, it became clear that some verbs were not viable experimental candidates.

60

form, filled out a language background questionnaire, and completed one of the norming tasks. Participants were instructed to provide up to three possible verb synonyms in infinitival form for each verb and to rate each of those synonyms as to how easy or difficult it was to think of them. Figure 3.1 provides an example of the questionnaires translated into English. All Phase II tasks are found in Appendices C-I. The participant consent form is found in Appendix C with the translation in Appendix D, the language background questionnaire in Appendix E (translation in F), the synonym-solicitation task for homographs in Appendix G, and the synonym-solicitation task for cognates in Appendix H. Figure 3.1 Example of the synonym-solicitation task, instructions and verbs translated into English. Instructions: Please indicate if you are familiar with the following verbs (yes/no). For each verb that you are familiar with, write down all the meanings that you can think of for that verb. Please write the meanings in verb infinitive form. Then, mark the degree to how difficult it was to think of each verb infinitive meaning. Ex. The verb ‘andar’ has as synonyms: to walk (the physical action), to take a walk (as in a special event), ‘to function’, as in ‘The car doesn’t andar/function’. So you choose according to the example below: caminar, pasear, funcionar. Verb

Are you familiar with this verb?

Meaning 1

Ex. andar (to walk)

yes/no

caminar

abortar

yes/no

Scale of difficulty; 4=easy, 1=difficult

Meaning 2

1234

pasear

Scale of difficulty;

Meaning 3

4=easy, 1=difficult

(to walk)

1234

(to take a walk)

1234

61

4=easy, 1=difficult funcionar

1234

(to work or function) 1234

(to abort)

Scale of difficulty;

1234

Synonym-clarification task The meanings for target verbs established by Spanish monolinguals through the synonym-solicitation task were compared to and combined with dictionary definitions to create the synonym-clarification task. The clarification task was necessary to resolve two issues. First, some discrepancy existed between the dictionary and usage definitions so it was necessary to see whether, if given a list of choices, participants would repeat the usage results of the synonym-solicitation task. Second, synonym foils were added to the norming task in order to reflect meanings that could be linked to the similar English form of the target verb. This was to ensure that Spanish monolinguals would indeed reject these meanings. For example, the verb restar looks like to rest in English, but does not mean such. A synonym candidate was added to the questionnaire for this verb: descansar (to rest). Spanish monolinguals were not expected to treat foils as synonyms to target verbs. In the synonym-clarification task, 122 experimental item candidates9 consisting of 60 of the original 76 potential homographs and 62 of the original 76 potential cognates were followed by a series of 5 possible synonyms, some of which were viable synonym candidates and some of which were not. The questionnaire was given to a new group of 40 Spanish monolinguals from the University of Granada, Granada, Spain, who completed it in a classroom setting during a 30minute experimental session. Participants signed a research consent form (Appendix C), filled out a language background questionnaire (Appendix E), and completed the norming questionnaire (Appendix I).

9

The questionnaire is numbered to 124; two numbers are missing in the numbering (51 and 79), making for 122 items.

62

A sample of the questionnaire is provided in Figure 3.2 Participants were instructed to indicate whether they were familiar with the target verbs by marking ‘yes’ or ‘no’. Participants then reviewed all possible synonyms for each verb from the set provided and circled any synonym candidates that reflected the meaning of the target verb. Finally, participants were asked to rank each synonym preference by placing a ‘1’ next to the most preferred synonym, a ‘2’ next to a second chosen, and so forth. The data were hand-coded and out of this data, a list of experimental verbs and their synonyms was established in order to create the norming questionnaires of Phase III. Figure 3.2 Example of the synonym-clarification task, instructions and verbs translated into English. Instructions: Please indicate if you are familiar with the following verbs (yes/no). For each verb that you are familiar with, choose all the related meanings. Then, number according to your preference, each one of the chosen verbs. Ex. The verb ‘andar’ has as synonyms: to walk (the physical action), to take a walk (as in a special event), ‘to function’, as in ‘The car doesn’t andar/function’. So you choose according to the example below, funcionar-3, pasear-2, caminar-1. Verb

Are you familiar with this verb?

Ex. andar (to walk)

yes/no

Admirar

yes/no

(to admire)

Meaning

Meaning

Meaning

Meaning

Meaning

funcionar3

pasear2

saltar

caminar1

dedicar

(to function)

(to take a walk)

(to jump)

(to walk)

(to dedicate)

adorar

suponer

asombrar

discutir

necesitar

(to adore)

(to suppose)

(to shock/surprise)

(to discuss)

(to need)

63

Phase III: Rating the strength of synonyms and norming form For the final series of norming questionnaires, four synonym-rating tasks were developed to yield ratings of synonym strength for cognate and homographic target items. Spanish speakers completed two Spanish synonym-rating tasks—one for cognates and one for homographs. English speakers completed the same synonym-rating tasks for cognates and homographs as Spanish speakers except that they were translated to English. The English speakers also completed a form similarity rating task indicating the degree to which the Spanish target verbs are similar in form to their English homograph counterparts. All tasks are found in Appendices J-N. Spanish synonym-rating tasks for cognates and homographs Functionally monolingual speakers of Spanish from Granada, Spain, completed both of the synonym-rating tasks in Spanish. Twenty-five participants rated 65 out of the original 76 homographs from Phase I while 29 participants rated 75 out of the original 76 cognates from Phase I.10 For both tasks, participants signed the research consent form found in Appendix C, filled out the language background questionnaire (Appendix E), and completed one of the norming tasks (Appendix J, Homographs and Appendix K, Cognates). To complete the tasks, participants were instructed to rate the degree to which each verb pair was synonymous by using a scale of 1 to 7, with 1 indicating no synonymous relationship and 7 indicating the strongest of synonymous relationships. A blank space was provided for each verb and participants were told that if a better, stronger synonym occurred to them, they were to write it in the blank and rate it. Figure 3.3 provides an example of the format for both Spanish synonym-rating tasks. 10

The syonym-rating tasks for cognates in both Spanish and English show 79 items. Four items were inadvertently listed twice on the cognate rating task: calcular, coordinar, decidir, identificar. Hence 75 items were tested.

64

Figure 3.3 Example of Spanish synonym-rating task, instructions translated into English. For each verb in bold below, indicate 1) the most common meanings of each for you and 2) to what degree each meaning is a synonym of the verb in bold, assigning a value between 1 and 7 (1 = not a synonym, 7 = a strong synonym). Note: If you find that a meaning you use is not presented, write it in the blank and include it in the values you assign. A. If for you, ‘to make notes’ is the only meaning or synonym you can choose for ‘to write’ and it is a strong synonym, then you give it a high number. If you don’t see any relationship of synonym between ‘to think’ and ‘to write’, then you assign a ‘1’ to ‘to think’. escribir (to write)

apuntar (to make notes) 1 2 3 4 5 6 7 pensar (to think) 1234567 ________ 1234567

B. If for ‘to embrace’, you use the meaning ‘to hug’, but also the meaning ‘to adopt’ as in ‘to adopt or embrace an idea’, then you write it in the blank. Then you indicate to what degree the meanings are synonyms. abrazar (to embrace)

1

2

1234567 1234567 1234567

acortar (to shorten) enlazar (to hug) adoptar (una idea) (to adopt (an idea))

VERB

MEANING

POINTS

acostar

dormir

1234567

atacar

1234567

_____________

1234567

repetir

1234567

contestar

1234567

_____________

1234567

replica

English synonym-rating tasks for cognates and homographs The English synonym-rating task for homographs and the English synonym-rating task for cognates were rated by 19 students of Spanish 101 at the University of Illinois on their first day of classes. Participants signed the research consent form (Appendix D), filled out the language background questionnaire (found in Appendix F), and completed the norming tasks 65

(Appendix L, Homographs and Appendix M, Cognates). Participants rated the degree to which each verb pair was synonymous by using a scale of 1 to 7, with 1 indicating no synonymous relationship and 7 indicating the strongest of synonymous relationships. A blank space was provided for each verb and participants were told that if a better, stronger synonym occurred to them, they were to write it in the blank and rate it. Figure 3.4 provides an example of the format for both English synonym-rating tasks Figure 3.4 Example of English version of synonym-rating task. For each verb in bold below, indicate 1) the most common meanings of each for you and 2) to what degree each meaning is a synonym of the verb in bold, assigning a value between 1 and 7 (1 = not a synonym, 7 = a strong synonym). Note: If you find that a meaning you use is not presented, write it in the blank and include it in the values you assign. Examples: A. If for you, ‘to make notes’ is the only meaning or synonym you can choose for ‘to write’ and it is a strong synonym, then you give it a high number. If you don’t see any relationship of synonym between ‘to think’ and ‘to write’, then you assign a ‘1’ to ‘to think’. to write

to make notes to think ________

1234567 1234567 1234567

B. If for ‘to embrace’, you use the meaning ‘to hug’, but also the meaning ‘to adopt’ as in ‘to adopt or embrace an idea’, then you write it in the blank. Then you indicate to what degree the meanings are synonyms. to embrace

1

2

1234567 1234567 1234567

to shorten to hug to adopt

VERB

MEANING

POINTS

to accost

to sleep

1234567

to attack

1234567

_____________

1234567

to repeat

1234567

to answer

1234567

_____________

1234567

to replicate

66

Form-similarity rating task The same 19 monolingual speakers of English who completed the synonym-similarity rating tasks also rated the degree to which target Spanish verbs are similar in form with English during the same testing session. The form-similarity rating task (Appendix N) consisted of a total of 139 verbs: 65 of the original 76 homograph candidates and 74 of the original 76 cognate candidates. For this task, participants were asked to follow the examples in the instructions of the task for rating the degree to which Spanish and English target verbs resembled each other. In the examples, if the complete English verb was found in the Spanish verb form, a ‘7’ rating was chosen. If little to no similarity was found between the verbs, a ‘1’ rating was selected. An example of the instructions and format of the form-similarity rating task can be found in Figure 3.5. Figure 3.5 Form-similarity rating task Please use the scale below to rate the degree to which you think the Spanish and English verbs look alike in each verb pair below. Keep in mind that the verbs will never look exactly alike because Spanish verbs have verb endings (-ar, -ir, -er) that are different from English. You can, however, make a form similarity judgment based on the rest of the verb. Examples for using the scale 5: You see the complete English verb form in the Spanish verb: conform

conformar

1234567

1: You see very little to no similarity between the two verbs in the pair. suspect

sospechar

1234567

VERB

VERBO

SCALE

1

accost

Acostar

1234567

2

replicate

Replicar

1234567

67

Results and discussion The data summaries and discussion are divided into two sections: Phase I/II data and Phase III data. Phase I/II data: Solicitation and clarification Dictionary and synonym-solicitation task The data from Phase II came from 69 functionally monolingual speakers of Spanish, 30 participants completing the synonym-solicitation task for homographs and 39 participants completing the synonym-solicitation task for cognates. The data were hand-coded and reviewed to find (a) the primary, secondary, and tertiary meanings for each verb, (b) the synonyms utilized by native speakers to express those meanings, and (c) the overall degree of familiarity participants had with each verb. In order to best organize the abundance and variety of synonyms provided by participants, the data were coded in the following way. The largest number of synonym tokens for a given meaning was considered to represent the strongest meaning of the verb (Meaning 1). The meaning with the next largest number of synonyms was considered to be the second meaning for the verb (Meaning 2). The third meaning of a given verb usually had a small number of tokens, thus if multiple third meanings with very few tokens were provided for a particular target verb, they were grouped together into the ‘Meaning 3’ category.11 The raw data for homographs is found in Appendix O., and for cognates, Appendix P. The data charts include the following information: Spanish target verbs; dictionary definitions of 11

In cases where multiple synonyms were given to express the same meaning of a verb, those synonyms were lumped together to form one general meaning with multiple examples for expressing that meaning. For example, the verb adorar means to adore, to worship. Participants gave that definition through a variety of synonyms: querer (to love), venerar (to worship), idolatrar (to idolize), amar (to love).

68

each Spanish verb; the number of meanings for each verb as indicated by the dictionary along with the total number of meanings provided by participants; Meanings 1, 2, and 3; the frequency with which each meaning was expressed by participants; the percentage of all responses that each meaning represents; and the best examples of synonyms to express separate meanings, along with their translations. Some errors were found on the norming tasks. The verbs estampar and consentir were included in the cognate norming task rather than the homograph norming task. This, however, was irrelevant to the data since the categories of ‘cognate’ and ‘homograph’ are arbitrary for monolingual speakers of Spanish and only are used here for the organization of potential experimental items. The verb alternar appeared on both tasks, so data were considered from the homograph task only. The verb enrollar was spelled incorrectly on the homograph task as enroller. The data still were included for this verb as participants corrected the spelling and treated the verb as enrollar in their answers. In all, the synonym-solicitation tasks produced 65 potential homographs and 60 potential cognates. Correlations were run separately on the homographs and on the cognates to compare the results of dictionary findings from Phase I with participant meanings provided in Phase II. For homographs, there was a significant correlation between the number and order of dictionary meanings and the number and order of definitions that participants provided (r=+.72, n=65, p