Are Figurative Interpretations of Idioms Directly Retrieved or Compositionally Built? Evidence from Eye Movement Measures of First and Second Language Reading Kyle S. Lovseth Department of Psychology McGill University, Montreal Submitted: June, 2012 A thesis submitted to McGill University in partial fulfillment of the requirements of the degree of M. A. In the Department of Psychology© Kyle S. Lovseth ii Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING TABLE OF CONTENTS ABSTRACT iv RÉSUMÉ v ACKNOWLEDGEMENTS vi CONTRIBUTIONS OF AUTHORS vii LIST OF TABLES viii LIST OF FIGURES x INTRODUCTION 1 What are Idioms? 1 Noncompositional Models of Idiom Processing 2 Compositional Models of Idiom Processing 3 Hybrid Models of Idiom Processing 4 Present Study 6 EXPERIMENT 1 8 Method 8 Participants 8 Stimuli 8 Apparatus 9 Procedure 10 Results 11 First Pass Effects on Phrase-final Nouns 13 Noun FFD 16 Noun GD 17 Disambiguating Region and Subsequent Idiom Effects 17 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING iii TABLE OF CONTENTS (CONTINUED) Disambiguating Region FPGD 20 Idiom Region TRT 21 Idiom Regressions 23 Discussion 23 EXPERIMENT 2 25 Method 27 Participants 27 Stimuli & Apparatus 27 Procedure 27 Results 28 First Pass Effects on Phrase-final Nouns 30 Noun FFD 31 Noun GD 32 Disambiguating Region and Subsequent Idiom Effects 32 Disambiguating Region FPGD 32 Idiom Region TRT 32 Idiom Regressions 34 Discussion 35 GENERAL DISCUSSION 37 REFERENCES 47 APPENDIX: Sentences used in Experiments 1 and 2 55 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING iv Abstract Idioms are part of a general class of multiword expressions that convey a figurative interpretation that is not fully determined through an on-demand compositional (syntactic and semantic) analysis of their component words (e.g., kick the bucket, save your skin). Idioms are simultaneously amenable to direct retrieval from memory, and to an on-demand compositional analysis, yet it is unclear which processes lead to figurative interpretations of idioms during comprehension. Idioms also pose challenges for second language learners, who may rely less on direct retrieval than first language users due to reduced L2 experience. In this study, first and second language users (L1 and L2, respectively) read sentences containing idioms followed by figurative- or literal-biased disambiguating regions. The results for L1 readers showed that increased familiarity but not decomposability facilitated comprehension of figurative interpretations. The results for L2 readers showed that cross-language overlap and decomposability facilitated comprehension of figurative interpretations. Thus, consistent with hybrid or multidetermined idiom processing models (Titone & Connine, 1999; Libben & Titone, 2008), first language idiom interpretations arise more from direct retrieval (indexed by familiarity) than compositional analysis. Second language idiom interpretations, however, arise from both. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING v Résumé Les expressions idiomatiques font partie d'une catégorie générale d’expressions à mots multiples qui confèrent une interprétation figurale qui n'est pas entièrement déterminée par une analyse compositionnelle sur-demande (syntaxique et sémantique) des mots qui les forment (par exemple, passer l’arme à gauche, sauver votre peau). Les expressions idiomatiques se prêtent à la fois à une récupération directe en mémoire, et à une analyse compositionnelle sur-demande. Cependant, il n'est pas clair quels processus mènent à l’interprétation figurale des idiomes lors de la compréhension. Les expressions idiomatiques posent également un défi pour les individus apprenant une langue seconde, individus qui, en raison d’une expérience L2 réduite, se basent possiblement moins sur la récupération directe que les utilisateurs de langue maternelle. Dans cette étude, des utilisateurs de langue maternelle et seconde (L1 et L2, respectivement) ont lu des phrases contenant des idiomes suivis de régions désambiguisantes biaisées figuralement ou littéralement. Pour les lecteurs L1, les résultats ont montré qu’une plus grande familiarité, mais pas une décomposabilité, facilite la compréhension des interprétations figurales. Pour les lecteurs L2, les résultats ont montré que le chevauchement translinguistique et la décomposabilité facilitent la compréhension des interprétations figurales. Ainsi, conformément aux modèles hybrides ou multidéterminés de traitement des idiomes (Titone & Connine, 1999; Libben & Titone, 2008), l’interprétation des idiomes de langue maternelle repose davantage sur la récupération directe (indexée par la familiarité) que sur l'analyse compositionnelle. Toutefois, l’interprétation des idiomes de langue seconde repose sur les deux. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING vi Acknowledgments I would like to start by thanking all of the undergraduate research assistants who helped with data collection. Japneet Kaur, Monica Bennett, Lauren de la Parra, and Lianne Morier, the time and effort each of you gave was invaluable and essential to the completion of this thesis. I would also like to thank my fellow lab members, Veronica Whitford, Irina Pivneva, Naveed Sheikh, Julie Mercier, and Maya Libben. The way we help and support each other makes our lab one of, if not THE, very best working environment in the world. It is my pleasure to work alongside each and every one of you. Next, I would like to thank all of my co-supervisors during my tenure at McGill: Caroline Palmer, Michael Wagner, Fred Genesee, and Jelena Ristic. They were always available when I had questions, and pushed me to do my best work. I owe all of my success to my supervisor, Debra Titone. Her enthusiasm for what she does is infectious and it has a hand in every facet of my research. Thank you for taking a chance on me. Finally, I would like to thank my mother, brother, grandmother, and grandfather to whom this thesis is dedicated. I hope I have made each of you proud. This research was supported by funding from the NSERC Create: Training in Auditory Cognitive Neuroscience Fellowship. In addition, thank you for the support from the Centre for Research on Language, Mind, and Brain for helping to peak my interest and guide me in exploring new statistical methods. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING vii Contributions of Authors This thesis is based on a manuscript submitted to the Journal of Memory and Language. The first author, Kyle Lovseth, was responsible for experimental design, data collection, statistical analyses, data interpretation, and writing all sections of the manuscript. The co-author, Debra Titone, was responsible for providing input on the experimental design and the data interpretation. In addition, the co –author provide valuable input and revisions to the introduction and discussion of the manuscript. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING viii List of Tables Table 1 Example Sentences From Id-Id, Id-Lit, and Lit-Lit Conditions 9 Table 2 Minimum Values, Maximum Values, Means, and Standard Deviations 10 for Measurements (Familiarity, L1/L2 Overlap, Decomposability, Literal Plausibility) of Idiom Normative Values, and Reading Proficiency and Character Length for Noun and Idiom Phrase in Each Condition (Idiom, Literal) and Disambiguating Region in Each Condition (Id-Id, Id-Lit, Lit-Lit) Table 3 Definitions for Eye Tracking Measures: First Fixation Duration, First 12 Pass Reading Gaze Duration, Total Reading Time, and Idiom Regressions Table 4 Total Number of Observations, Means, and Standard Deviations Across 14 All Participants for All Eye Movement Measurements (Noun FFD, Noun GD, Disambiguating FPGD, Idiom TRT, Idiom Regressions) of Native Language Readers In Each Condition (Idiom, Literal, Id-Id, Id-Lit, Lit-Lit) for High and Low Levels of Familiarity and Decomposability for Native English Speakers in Experiment 1 Table 5 Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Noun Region First Fixation Duration and Noun Region Gaze Duration for Native English Speakers in Experiment 1 16 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING ix List of Tables (Continued) Table 6 Effect Size, Standard Error, and t-Values for Fixed Effects and Variances 19 for Random Effects Derived From Optimal LMER Models for the Dependent Variables Disambiguating Region First Pass Gaze Duration, Idiom Region Total Reading Time, and Idiom Region Regressions for Native English Speakers in Experiment 1 Table 7 Total Number of Observations, Means, and Standard Deviations Across 29 All Participants for All Eye Movement Measurements (Noun FFD, Noun GD, Disambiguating FPGD, Idiom TRT, Idiom Regressions) of Non-native Language Readers In Each Condition (Idiom, Literal, Id-Id, Id-Lit, Lit-Lit) for High and Low Levels of L1/L2 Overlap and Decomposability for Non-Native English Speakers in Experiment 2 Table 8 Effect Size, Standard Error, and t-Values for Fixed Effects and Variances 30 for Random Effects Derived From Optimal LMER Models for the Dependent Variables Noun Region First Fixation Duration and Noun Region Gaze Duration for Non-Native English Speakers in Experiment 2 Table 9 Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Disambiguating Region First Pass Gaze Duration, Idiom Region Total Reading Time, and Idiom Region Regressions for Non-Native English Speakers in Experiment 2 33 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING x List of Figures Figure 1 Noun gaze duration as a function of decomposability for the idiom and 18 literal conditions in native language readers Figure 2 Disambiguating region first pass gaze duration as a function of familiarity 21 and decomposability for the Id-Id, Id-Lit, and Lit-Lit conditions in native language readers Figure 3 Idiom region total reading time as a function of familiarity for the Id-Id, 22 Id-Lit, and Lit-Lit conditions in native language readers Figure 4 Proportion of regressions to idiom region as a function of familiarity for 23 the Id-Id, Id-Lit, and Lit-Lit conditions in native language readers Figure 5 Noun first fixation duration as a function of L1/L2 overlap for the idiom 31 and literal conditions in non-native language readers Figure 6 Idiom region total reading time as a function of decomposability for the 34 Id-Id, Id-Lit, and Lit-Lit conditions in non-native language readers Figure 7 Proportion of regressions to idiom region as a function of decomposability for the Id-Id, Id-Lit, and Lit-Lit conditions in non-native language reader 35 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 1 Are Idioms Directly Retrieved or Compositionally Built? Evidence from Eye Movement Measures of First and Second Language Reading What are Idioms? Idioms are part of a general class of multiword expressions, which include other elements of non-literal language such as metaphor, sarcasm, and indirect requests. Idioms are similar to other forms of non-literal language because they convey a figurative interpretation that is not fully determined through an on-demand compositional (syntactic and semantic) analysis of their component words (Chomsky, 1980; Fraser, 1970). Thus, a standard compositional analysis of the idiom, kick the bucket, does not lead to its figurative interpretation to die, although other idioms may be relatively more “compositional” (e.g., save your skin, which means, to protect one’s self) (Nunberg, 1978; Jackendoff, 2002). Idioms differ from other classes of non-literal language, however, because their figurative meanings are conventional and frequently used within a language, and also across languages (Wray, 2002). Thus, at some level, idioms must possess unitary representations that are stored in memory, similar to non-figurative yet over-learned multiword sequences that also pervade language (e.g., conversational expressions such as How do you do?). Despite a long-standing body of research, fundamental questions about idiom processing remain unsettled. In this study, I investigate how first (L1) and second (L2) language users process idioms using eye movement measures of reading, a method that has the potential to clarify previously conflicting theoretical accounts of idiom processing. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 2 Noncompositional Models of Idiom Processing Early models of idiom processing emphasized the lexical and noncompositional nature of idioms, and thus the direct retrieval of idiomatic meaning. For example, the lexical representation hypothesis (Swinney & Cutler, 1979) posits that idioms are represented in memory as lexical entities, and that their pre-compiled figurative meanings are directly retrieved in a manner similar to that of long compound words (see also Bobrow & Bell, 1973). Support for the lexical representation hypothesis comes from the widely replicated finding that highly familiar idioms are processed more rapidly than unfamiliar non-idiomatic phrases or idioms (e.g., Cronk & Schweigert, 1992; see also Giora, 1997; Libben & Titone, 2008; Schweigert, 1986; Swinney & Cutler, 1979; Tabossi, Fanari, & Wolf, 2009). Facilitative effects of increased idiom familiarity presumably reflect direct retrieval of pre-compiled figurative meanings, which is less effortful and faster than on-demand compositional analyses. The configuration model (Cacciari & Tabossi, 1988) also posits that direct retrieval of stored idiomatic meanings leads to comprehension, although here idioms are represented as distributed word configurations that are processed non-idiomatically until enough information is encountered for the configuration to emerge (Fanari, Cacciari, & Tabossi, 2010; Tabossi, Fanari, & Wolf, 2009; Vespignani et al., 2010). Within this view, idioms are understood through direct retrieval like any other overlearned word sequence, such as songs or clichés (Tabossi, Fanari, and Wolf, 2009; but see Molinaro & Carreiras, 2010, and Vespignani et al., 2010, for electrophysiological evidence of distinct neural responses for idioms vs. non-idiomatic collocations). Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 3 Compositional Models of Idiom Processing In recent years, however, direct retrieval models have fallen out of favor because of work showing that idioms often behave like compositional word sequences (Nunberg, 1978; Nunberg, Sag, & Wasow, 1994; Gibbs, Nayak, & Cutting, 1989, Gibbs, 1992). For example, idioms can retain their figurative meanings when used creatively in syntactically or semantically modified forms (McGlone, Cacciari, & Glucksberg, 1994; Gibbs & Nayak, 1989; Gibbs, Nayak, Bolton, & Keppel, 1989; Hamblin & Gibbs, 1999). Accordingly, the figurative meaning of the idiom break the ice can be understood in a variant form, such as crack the ice. Based on such findings, Nunberg (1978) posited that idioms are more heterogeneous than previously thought, and Gibbs and colleagues demonstrated that people process distinct classes of idioms differently. For example, Gibbs and Nayak (1989, Experiment 2) found that normally decomposable idioms, whose component words overlap with a figurative meaning (pop the question), retain their figurative meaning when syntactically modified (the question was popped) more than nondecomposable idioms, whose component words are distinct from a figurative meaning (the bucket was kicked). Moreover, other studies have shown that decomposable idioms have a global processing advantage over nondecomposable idioms. For example, Gibbs, Nayak, and Cutting (1989) found that meaningfulness judgments for decomposable idioms were faster than those for nondecomposable idioms and literal control sentences. As well, nondecomposable idioms were judged to be meaningful more slowly than literal control sentences. To explain these results, Gibbs and colleagues suggested that decomposable idioms are understood directly through a compositional analysis because their idiomatic and literal interpretations are mutually reinforcing (e.g., question in the idiom pop the question refers directly to its figurative interpretation, to Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 4 propose marriage). By comparison, nondecomposable idioms are disadvantaged because their idiomatic and literal phrasal meanings are semantically distinct (e.g., kick the bucket can mean to die suddenly or to strike a bucket with one’s foot) (see also Callies & Butcher, 2007). However, much of the work showing facilitative effects of decomposability is based on tasks that arguably reflect off-line rather than on-line processing (e.g., paraphrase judgments), or studies that use small numbers of idioms that potentially vary in others ways aside from decomposability (e.g., literal plausibility, familiarity, etc.). Hybrid Models of Idiom Processing Thus, an important but unresolved question is how and when decomposability exerts an effect on comprehension. Related to this question, Titone and colleagues have also observed facilitative effects of compositionality on idiom processing but only for later integrative stages of comprehension, a finding that conflicts with predictions of strict compositional models. In an eye movement reading study, Titone and Connine (1999) found differences between decomposable and nondecomposable idioms in reading rates as a function of context. People read sentences where idioms either preceded or followed a context that supported a figurative (He tried to save his skin by getting his work done on time) or literal (He tried to save his skin by avoiding the tanning salons) meaning. Nondecomposable idioms were read more slowly when a biased (literal or figurative) context preceded the idiom compared to when it followed the idiom, suggesting that additional time was required to select between one of two active phrase-level interpretations. Decomposable idioms did not show this slowing, suggesting that semantic integration was more easily accomplished for idioms with overlapping phrase-level interpretations that had been activated. Similarly, Libben and Titone (2008) observed facilitative Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 5 effects of familiarity across a wide variety of tasks but only observed facilitative effects of decomposability for tasks that required people to judge the meaning of idioms as a whole (similar to the body of work reported by Gibbs and colleagues, but in contrast with Tabossi, Fanari, & Wolf, 2008, 2009). For example, in a task that required people to read sentences wordby-word for comprehension without any explicit judgments about the meaning of the idiom, familiarity facilitated reading times but decomposability did not. Thus, decomposability may not facilitate initial activation of an idiom's figurative meaning, but it might help in selecting between a simultaneously activated figurative and compositionally generated literal interpretation of an idiomatic string. Based on such findings, Titone and Connine (1999) posited a hybrid model of idiom comprehension that incorporates aspects of both direct retrieval and compositional processing (see also, Libben & Titone, 2008). Central to this view is that comprehenders simultaneously follow all paths to comprehension (direct retrieval and compositional analysis), although the availability of different kinds of information may follow distinct time-courses, and thus affect comprehension at different stages. Assuming that direct retrieval is a fast process that can be triggered prior to the phrase-final word, any variable that affects direct retrieval (e.g., idiomatic status, familiarity, or predictability) should exert effects at the earliest stages of comprehension, prior to the phrase-final word. In contrast, an on-demand compositional analysis is likely to be slower and more effortful than direct retrieval, and would certainly require encountering the phrase-final word to be complete. Thus, variables that affect compositional processing (e.g., degree of decomposability) only exert effects after exposure to the phrase-final word (see Prado & Ullman, 2009, for a similar logic applied to past tense morphology). Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 6 Present Study I investigated three interdependent questions about idiom processing to adjudicate between direct retrieval models, compositional models, and hybrid models of comprehension. First, are idioms directly retrieved from memory as lexical-semantic chunks, are they built compositionally, or do both operations occur simultaneously in a time-dependent fashion? Second, how do these factors affect the ability to resolve phrase-level ambiguity that arises from plausible non-idiomatic phrasal interpretations (e.g., when one literally kicks the bucket)? Finally, how do these processes differ for people with varied linguistic experiences (e.g., when native or non-native language users encounter idioms)? While prior studies have investigated similar questions, the literature lacks consensus due to heterogeneity of the idioms tested (both in terms of structural and other linguistic attributes), the measures used (e.g., lexical decision, paraphrase judgment, meaningfulness decision, which differentially tap into on-line vs. off-line processing), and whether people were required to interpret idioms figuratively or literally. Our goal is to circumvent these issues to the extent possible by studying well-controlled idioms in sentences using naturalistic eye movement measures of sentence reading (Rayner, 1998, 2009). In Experiment 1, I investigate how familiarity and decomposability modulate idiom comprehension when native English users read conventional English idioms. Idioms were selected that have a consistent grammatical structure (verb-determiner-noun) and length, and that were normed on several important linguistic dimensions in our prior work (Libben & Titone, 2008). These idioms (and control non-idiomatic literal phrases) were embedded in sentences that participants read for comprehension while their eye movements were monitored. Within these sentences, idioms were followed by disambiguating contexts that biased the figurative (Vicki knew the score well enough not to ask for an extension on the project) or literal (Vicki knew the Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 7 score because she had been following the playoffs closely) interpretation. Structuring the sentences in this way allowed us to examine reading behavior at two key comprehension points; the region of the idiomatic phrase prior to contextual information about which meaning was intended, and after a particular meaning was specified by the subsequent disambiguating context. Experiment 2 was identical to Experiment 1 except that non-native English speakers were tested to assess whether reduced participant familiarity with English idioms would increase the likelihood that a compositional analysis would lead to a figurative interpretation. I predicted that native English users would read idioms more quickly than literal phrases on the first pass due to their familiarity with these conventional idioms. I also predicted that native language users would have more difficulty reading a figuratively biased disambiguating region for low vs. high familiar idioms, relative to literal control sentences. Regarding decomposability, consistent with previous work (Titone & Connine, 1999; Libben & Titone, 2008), I predicted that increased decomposability might be important for resolving phrase-level competition between idiomatic and non-idiomatic interpretations while reading idioms on the first pass. Because nondecomposable idioms have semantically distinct phrase-level interpretations (to kick the bucket and die vs. to stub one’s toe) and decomposable idioms have semantically similar phrase-level interpretations (to save your skin and cover up a transgression vs. one’s flesh). However, based on predictions of hybrid or multidetermined models, I predicted that only their familiarity would facilitate reading for later, figuratively biased regions of sentences. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 8 Experiment 1 Method Participants Thirty-six native English users from the McGill and Montreal community participated for course credit or compensation of $10/hr. All participants had normal or corrected-to-normal vision and no self-reported history of speech or hearing disorders. Stimuli Stimuli consisted of 60 English idioms taken from Libben and Titone (2008). All idioms had a verb - determiner - noun structure (e.g., had a lark). Non-idiomatic control phrases were formed by changing the verb in each idiomatic phrase (e.g., saw a lark), thus the nouns were identical in the idiomatic and non-idiomatic conditions. Across the idiom-bearing sentences and their non-idiom controls, the verbs matched in both character length (+/-1 character) and word frequency (+/-5 counts per million, Kucera & Francis, 1967). Each idiom was embedded in one of three sentence conditions: an idiomatic expression followed by a figurative disambiguating region (Id-Id) or an idiomatic expression followed by a literal disambiguating region (Id-Lit). Non-idiomatic phrases were also presented, followed by a supporting context (Lit-Lit). Table 1 presents example sentences from each condition, and the entire set of sentences is listed in the Appendix. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 9 Table 1. Example Sentences From Id-Id, Id-Lit, and Lit-Lit Conditions Condition Id-Id Id-Lit Lit-Lit Sentence Ruby had a lark when she switched her family’s sugar to salt as a joke. Ruby had a lark when she was a child but now wanted a parrot instead. Ruby saw a lark at the pet store and thought that it was very beautiful. Independent ratings of familiarity, decomposability, and literal plausibility for each idiom were obtained from Libben and Titone (2008). Idioms used had a mean familiarity rating of 3.40 on a five-point Likert scale (1 = Low familiarity; 5 = High familiarity), a mean decomposability proportion of 0.43 (0 = Not decomposable; 1 = Fully decomposable), and a mean literal plausibility rating of 3.87 on a five-point Likert scale (1 = Low literal plausibility; 5 = High literal plausibility). Average total reading times (log transformed) were also calculated for all filler sentences for each subject as a measure of overall reading speed or ability. Finally, total character lengths were obtained for the noun, idiom, and disambiguating regions. For both the idiom and disambiguating regions, there were no significant differences in either character length between any of the specified conditions. Nouns were identical across all conditions. Measures of familiarity, decomposability, literal plausibility, reading proficiency, and character length are shown in Table 2. Apparatus An Eye-Link 1000 tower mounted system (SR-Research, Ontario, Canada) was used to record eye movements (1 kHz sampling rate). Viewing was binocular but eye movements were recorded from the right eye only. Stimuli were presented on a 21” ViewSonic CRT monitor with a screen resolution of 1024 x 768 pixels, using EyeTrack 7.10 software developed at UMass Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 10 Amherst (http://www.psych.umass.edu/eyelab/software). Text was presented on a single line in yellow 10-point Monaco font on a black background. Participants’ eyes were positioned 71 cm from the monitor, thus 1 degree of visual angle comprised of approximately 3 characters of text. Table 2. Minimum Values, Maximum Values, Means, and Standard Deviations for Measurements (Familiarity, L1/L2 Overlap, Decomposability, Literal Plausibility) of Idiom Normative Values, and Reading Proficiency and Character Length for Noun and Idiom Phrase in Each Condition (Idiom, Literal) and Disambiguating Region in Each Condition (Id-Id, Id-Lit, Lit-Lit) Condition Idiom Normative Values Reading Proficiency Noun Idiom Phrase Disambiguating Region Idiom Literal Id-Id Id-Lit Lit-Lit Measurement Familiarity (Exp.1) L1/L2 Overlap (Exp.2) Decomposability Literal Plausibility log-Total Reading Time (Exp.1) log-Total Reading Time (Exp.2) Length (characters) Length (characters) Length (characters) Min 1.60 0.00 0.00 1.50 5.31 7.47 3 11 11 42 43 43 Max 4.80 0.92 0.97 5.00 6.08 8.49 8 20 20 63 62 61 Mean 3.40 0.28 0.43 3.87 5.71 8.07 4.67 15.78 15.68 54.25 53.80 53.42 Procedure Each participant completed a consent form and online language background questionnaire. Participants were instructed to read sentences naturally while their eye movements were recorded. Before each trial, participants were instructed to direct their gaze towards a small fixation circle, the calibration point, appearing in the center of the monitor. Once the experimenter noted that the participant’s gaze was fixed upon the calibration point, s/he initiated the trial. Trials consisted of the appearance of a fixation square in a location equivalent to that of SD 0.85 0.24 0.24 0.77 0.20 0.25 1.24 2.16 2.13 4.32 4.31 4.15 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 11 the first word of the sentence. Gazing at the fixation square automatically triggered the sentence for that trial. Participants were instructed to read the sentence at a normal rate for comprehension, and to press a button on a controller pad once they finished reading and understood the sentence. Participants viewed one of three sets of sentences in a fully counterbalanced and randomized order. Within each set, each idiomatic expression appeared in only one of its respective conditions. Thus, if a participant saw the idiomatic expression, broke the ice, in the IdId sentence, the next participant would see the same phrase in the Id-Lit sentence, and so on. No participant saw the same idiomatic expression more than once. In addition to the 60 sentences of primary interest, each participant read 19 “filler” sentences, 10 practice sentences, and 22 comprehension questions. This resulted in a total of 111 sentences read by each participant. The filler and practice sentences were similar to non-idiomatic literal sentences. All filler and practice sentences conveyed literal interpretations and were between 11 and 17 words in length. Comprehension questions appeared on roughly 20% of trials. Participants responded to the question by pressing an appropriately labeled controller pad. Overall accuracy on the comprehension questions was 93%, indicating that participants were attentive during the experiment. Results Linear mixed effects (LME) models were used to analyze all eye movement data. LME has been used in other reading studies (e.g., Kliegl, Risse, & Laubrock, 2007), and has advantages in statistical power over separate analyses of variance (ANOVA) across subjects and items (Baayen, Davidson, Bates, 2008; Quene & van den Bergh, 2008). I used the lme4 package Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 12 (Bates & Sarkar, 2006) implemented for use with the R Project for Statistical Computing (R Development Core Team, 2006; Baayen, 2008; Baayen, Davidson, Bates, 2008). Our dependent variables (Table 3) included noun first fixation duration (Noun FFD) and noun gaze duration (Noun GD) as measures of idiom comprehension prior to encountering the disambiguating region. Comprehension measures for the disambiguating region included first pass gaze duration of the disambiguating region (Disambiguating FPGD), total reading time for the entire idiomatic phrase (Idiom TRT), and proportion of trials in which the idiom was regressed into from the disambiguating region (Idiom Regressions). First fixation and gaze duration reflect early cognitive processes, such as lexical access; total reading time and idiom regression proportion reflect later processes, such as semantic integration (Rayner, 1998, 2009). Table 3. Definitions for Eye Tracking Measures: First Fixation Duration, First Pass Reading Gaze Duration, Total Reading Time, and Idiom Regressions Eye Tracking Measure First fixation duration First pass gaze duration / Gaze duration Total reading time Idiom regressions Definition The length of time the eyes fixate on the target word the first time they land on it The sum of all fixation durations following rightward moving saccades starting the moment the eyes land on the target region until the moment they move away The sum of all fixations and refixations on the target region during a given trial The proportion of trials in which the target region was regressed into from the disambiguating region Likelihood ratio tests (LRTs) were conducted to identify the optimal model for each dependent variable (Baayen, 2008). The LRTs evaluated goodness of fit for a given model though a series of iterative log-likelihood ratio comparisons of more complex models (i.e., Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 13 models with a greater number of fixed effects) to progressively less complex models (i.e., models with fewer fixed effects). As such, optimal models for each dependent variable varied as a function of best fit and contained only fixed effects that significantly improved the fit of the model. Markov chain Monte Carlo (MCMC) sampling tests were conducted to obtain p-values for the coefficients of all contrasts included in optimal LME models. Generalized linear mixed effects models were used for binary dependent variables (i.e., regressions to the idiom region). Before analysis, 2.2% of the total fixations were removed because of track loss (e.g., eye blinks) or because fixation durations were less than 80 ms. For each dependent variable, an upper cutoff of 4 standard deviations above the mean was established to remove extreme outliers. For Noun FFD and Noun GD, all trials in which the noun was skipped (n = 535, 24.8% of all Noun FFD and Noun GD trials), or the verb or determiner of the idiom/literal phrase was not fixated prior to fixating on the noun during the first pass (n = 104, 5% of all Noun FFD and Noun GD trials) were excluded. For Disambiguating FPGD, Idiom TRT and Idiom Regression data, trials where the entire idiomatic phrase was skipped on the first pass (n = 20, 1% of all Disambiguating FPGD, Idiom TRT and Idiom Regression trials) were excluded. Finally, because a participant could have briefly entered the idiom or disambiguating region before quickly regressing out of either region, trials from Disambiguating FPGD data if the first pass gaze duration was equal to the first fixation duration of the disambiguating region for that same trial (n = 170, 7.8% of all Disambiguating FPGD trials) were removed. Trials in which this was the case would indicate that only a single fixation occurred for an entire disambiguating region. Subject means and the number of observations for each dependent variable binned across high and low Familiarity and Decomposability in each condition are presented in Table 4. These means were included to facilitate comparison to a traditional F1 ANOVA analysis; they do not reflect the actual analytical Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 14 LME procedure, in which Familiarity and Decomposability were treated as continuous predictors. Table 4. Total Number of Observations, Means, and Standard Deviations Across All Participants for All Eye Movement Measurements (Noun FFD, Noun GD, Disambiguating FPGD, Idiom TRT, Idiom Regressions) of Native Language Readers In Each Condition (Idiom, Literal, Id-Id, Id-Lit, LitLit) for High and Low Levels of Familiarity and Decomposability for Native English Speakers in Experiment 1 Familiarity Eye-Movement Measure Noun FFD1 Condition n Idiom 1001 Literal 505 1506 All Idiom 1003 Noun GD1 Literal 502 1505 All 647 Disambiguating FPGD2 Id-Id Id-Lit 672 Lit-Lit 641 1960 All Id-Id 703 Idiom TRT1 Id-Lit 711 Lit-Lit 702 2116 All Id-Id 711 Idiom Regressions Id-Lit 715 Lit-Lit 707 2133 All 1 Values presented in milliseconds 2 Values presented in milliseconds per character Note. Standard deviations are listed in parentheses Condition Mean 227 (71) 239 (79) 231 (73) 245 (92) 261 (100) 250 (95) 35.1 (14) 36.1 (15) 34.2 (14) 35.1 (15) 737 (422) 710 (384) 741 (405) 729 (404) 30.70% 26.00% 26.70% 27.80% Low 233 (72) 245 (84) 237 (76) 255 (97) 267 (103) 259 (99) 36.0 (15) 35.7 (15) 34.1 (13) 35.3 (14) 809 (461) 761 (420) 760 (407) 777 (430) 36.80% 27.60% 25.10% 30.00% High 220 (68) 232 (73) 224 (70) 233 (86) 255 (97) 241 (90) 34.2 (14) 36.4 (15) 34.2 (16) 35.0 (15) 661 (359) 661 (340) 721 (404) 681 (369) 24.00% 24.50% 28.30% 25.60% Decomposability Low 232 (75) 239 (77) 235 (76) 258 (103) 263 (99) 260 (102) 35.4 (15) 37.3 (15) 34.4 (14) 35.7 (15) 729 (394) 731 (407) 744 (380) 735 (393) 31.40% 25.90% 26.50% 27.90% High 221 (66) 238 (81) 227 (71) 232 (77) 259 (102) 241 (87) 34.8 (14) 34.8 (14) 34.0 (14) 34.5 (14) 745 (447) 689 (359) 737 (430) 723 (414) 29.90% 26.20% 26.90% 27.70% Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 15 First Pass Effects on Phrase-final Nouns I constructed initial LME models with Noun FFD and Noun GD as dependent variables, and subject and item as random effects. In these models, data from the Id-Id and Id-Lit conditions were combined to form a single “Idiom” condition, given that these two conditions were identical prior to the disambiguating region of the sentence. The most complex models included the interaction between the following theoretically driven predictor variables: Condition (idiom, literal), Familiarity, and Decomposability. The most complex models also included several control predictors: Trial (to account for practice or fatigue effects), Literal Plausibility (to account for phrasal ambiguity), Reading Proficiency (to account for individual differences in reading speed), and Character Length (to account for item differences in length of any target region of interest). All predictor variables were continuous except for Condition, which was categorical. All continuous predictor variables, except for Trial, were centered in an effort to reduce occurrences of co-linearity, and divided by their standard deviation prior to analysis (Baayen, 2008). The literal sentence condition was mapped to the intercept (i.e., it was the baseline). For significant effects, I report differences in means derived from Table 4 and t-values derived from Table 5. These means are included to facilitate comparison to traditional F1 analyses and do not reflect the actual LME procedure. For control predictor variables, I report only t-values for significant effects derived from the optimal LME models, also summarized in Table 5. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 16 Table 5. Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Noun Region First Fixation Duration and Noun Region Gaze Duration for Native English Speakers in Experiment 1 Noun Region First Fixation Duration Gaze Duration b SE t-value b SE t-value Fixed effects1 Primary Predictors Condition (Literal, Idiom)2 -10.23 3.82 -2.68** -15.41 4.81 -3.20** Familiarity -7.45 2.31 -3.23** -10.17 4.06 -2.51* Decomposability 5.04 5.16 0.98 Condition*Familiarity Condition*Decomposability -10.58 4.83 -2.19* Familiarity*Decomposability Condition*Familiarity*Decomposability Control Predictors Trial Literal Plausibility Reading Proficiency 9.92 3.25 3.05** 20.18 3.85 5.24*** Character Length (Intercept) 237.64 4.43 53.61*** 259.41 5.83 44.54*** Random Effects Variance Variance Subject 278.85 482.2 Item 123.47 368.54 Residual 4841.59 7628.36 * Significant at the .05 level ** Significant at the .01 level *** Significant at the .001 level 1 Fixed effects that significantly improved the model where was p<.05 during likelihood ratio tests 2 Model assumes literal condition as the baseline across conditions Note. All continuous fixed effects are standardized, with the exception of Trial Noun FFD. The optimal model for Noun FFD included fixed effects of Condition (Idiom vs. Literal), Familiarity, and Reading Proficiency. There was a significant effect of Condition in that nouns at the ends of idioms received shorter first fixation durations than nouns at the ends of literal phrases (-12 ms difference between Idiom and Literal; t = -2.68). There was also a significant effect of Familiarity in that nouns received shorter first fixation durations as familiarity increased (-13 ms between high and low familiarity; t = -3.23), although this did not interact with Condition. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 17 With respect to our control variables, as Reading Proficiency increased (and thus, total reading time for filler sentence decreased), nouns received shorter first fixation durations (t = 3.05). Noun GD. The optimal model for Noun GD included fixed effects of Condition (Idiom vs. Literal), Familiarity, Decomposability, Reading Proficiency, and the interaction between Condition and Decomposability. There was a significant effect of Condition in that nouns at the end of idioms received shorter gaze durations than nouns at the end of literal phrases (-16 ms difference between Idiom and Literal; t = -3.20). There was also a significant effect of Familiarity in that all nouns, irrespective of Condition, received shorter gaze durations as familiarity increased (-18 ms difference between high and low familiarity; t = -2.51). There was a significant interaction between Condition and Decomposability in that increased decomposability shortened gaze duration for the idiom condition but not for the literal condition (Idiom: -26 ms between high and low decomposability, Literal: -4 ms between high and low decomposability; t = -2.19). Figure 1 depicts this interaction. With respect to our control variables, again, nouns received shorter gaze durations as Reading Proficiency increased (t = 5.24). Disambiguating Region and Subsequent Idiom Effects I constructed LME models with Disambiguating FPGD, Idiom TRT, and Idiom Regressions as the dependent variables, and subject and item as random factors. The most complex models included the interaction between our primary predictor variables; Condition (IdId, Id-Lit, Lit-Lit), Familiarity, and Decomposability. I again included Trial, Literal Plausibility, Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 18 Figure 1. Noun Gaze Duration as a Function of Decomposability for the Idiom and Literal Conditions in Native Language Readers Reading Proficiency, and Character Length as control predictors. However, to control for character length in the Disambiguating FPGD model, I examined FPGD in milliseconds per character rather than raw values. I did this only for Disambiguating FPGD because it was the only target region in which length varied by more than three characters between conditions for a given item. For all models, the Lit-Lit condition was treated as the control condition, and thus was mapped to the intercept. For significant effects, I report differences in means derived from Table 4 and t-values derived from Table 6. For control variables, I report only t-values for significant effects derived from the optimal LME models, summarized in Table 6 Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 19 Table 6. Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Disambiguating Region First Pass Gaze Duration, Idiom Region Total Reading Time, and Idiom Region Regressions for Native English Speakers in Experiment 1 Disambiguating Region First Pass Gaze Duration b SE t-value Idiom Region Total Reading Time Regressions b SE t-value b SE z-value Fixed effects1 Primary Predictors Condition (Lit-Lit, Id-Id)2 0.76 0.64 1.20 -7.64 17.48 -0.44 Condition (Lit-Lit, Id-Lit)2 1.90 0.63 3.02** -35.94 17.43 -2.06* Familiarity 0.13 0.54 0.25 0.44 15.25 0.03 Decomposability -0.43 0.54 -0.79 Condition (Id-Id)*Familiarity -1.26 0.72 -1.74 -70.31 17.66 -3.98*** Condition (Id-Lit)*Familiarity 1.82 0.71 2.57* -40.43 17.45 -2.32* Condition (Id-Id)*Decomposability 0.29 0.71 0.41 Condition (Id-Lit)*Decomposability -1.36 0.71 -1.93*1 Familiarity*Decomposability Condition (Id-Id)*Familiarity*Decomposability Condition (Id-Lit)*Familiarity*Decomposability Control Predictors Trial -0.04 0.01 -2.44* -1.02 0.41 -2.46* Literal Plausibility -0.75 0.31 -2.44* -24.27 11.38 -2.14* Reading Proficiency 7.54 0.79 9.55*** 180.59 21.16 8.53*** Character Length 58.04 11.30 5.14*** (Intercept) 35.30 1.00 35.25*** 774.84 28.03 27.64*** Random Effects Variance Variance Subject 20.20 14384.10 Item 1.64 4540.60 Residual 129.98 106727.70 * Significant at the .05 level ** Significant at the .01 level *** Significant at the .001 level *1 p = 0.53 1 Fixed effects that significantly improved the model where was p<.05 during likelihood ratio tests 2 Model assumes literal condition as the baseline across conditions Note. All continuous fixed effects are standardized, with the exception of Trial 0.18 -0.04 0.07 -0.44 -0.22 - 0.13 0.13 0.09 0.13 0.13 - 1.45 -0.35 0.70 -3.47*** -1.72 - 0.44 0.13 3.51*** -1.16 0.15 -7.76*** Variance 0.47 0.05 n/a Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 20 Disambiguating Region FPGD. The optimal model for Disambiguating FPGD included fixed effects of Condition (Id-Id vs. Lit-Lit & Id-Lit vs. Lit-Lit), Familiarity, Decomposability, Trial, Literal Plausibility, Reading Proficiency, the interaction between Condition and Familiarity, and the interaction between Condition and Decomposability. There were no significant effects or interactions involving the Id-Id vs. Lit-Lit contrast. However, for the Id-Lit vs. Lit-Lit contrast, there was a significant effect of Condition in that first pass gaze durations for disambiguating regions in the Id-Lit condition were longer than first pass gaze durations in the Lit-Lit condition (1.9 ms/character difference between Id-Lit and Lit-Lit; t = 3.02). There was also a significant interaction between Condition and Familiarity, in that increased familiarity slowed first pass gaze durations in the Id-Lit condition compared to increases in familiarity in the Lit-Lit condition (Id-Lit: 0.7 ms/character difference between high and low familiarity, Lit-Lit: 0.1 ms/character difference between high and low familiarity; t = 2.57). Finally, there was a significant interaction between Condition and Decomposability in that increased decomposability facilitated reading for the Id-Lit condition compared to the Lit-Lit condition (IdLit: -2.5 ms/character difference between high and low decomposability, Lit-Lit: -0.4 ms/character difference between high and low decomposability; t = -1.93). All significant interactions are depicted in Figure 2. With respect to our control variables, first pass gaze durations for disambiguating regions were shorter for increases in Trial (t = -2.44), Literal Plausibility (t = -2.44), and Reading Proficiency (t = 9.55). Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 21 Figure 2. Disambiguating Region First Pass Gaze Duration as a Function of Familiarity and Decomposability for the Id-Id, Id-Lit, and Lit-Lit Conditions in Native Language Readers Idiom Region TRT. The optimal model for Idiom TRT included fixed effects of Condition, Familiarity, Trial, Literal Plausibility, Reading Proficiency, Character Length, and the interaction between Condition and Familiarity. For the Id-Lit vs. Lit-Lit contrast, there was a significant effect of Condition in that total reading times for idioms in the Id-Lit condition were shorter than total reading times for literal phrases in the Lit-Lit condition (-31 ms difference between Id-Lit and Lit-Lit; t = -2.06). For the Id-Id vs. Lit-Lit contrast, there was a significant interaction between Condition and Familiarity in that increased familiarity was more facilitative in the Id-Id condition than in the Lit-Lit condition (Id-Id: -148 ms difference between high and low familiarity, Lit-Lit: -39 ms difference between high and low familiarity; t = -3.98). For the Id-Lit vs. Lit-Lit contrast, the interaction between Condition and Familiarity was also significant in that increased familiarity was more facilitative in the Id-Lit condition than in the Lit-Lit condition (Id-Lit: -100 ms difference between high and low familiarity, Lit-Lit: -39 ms Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 22 difference between high and low familiarity; t = -2.32). Figure 3 depicts the interactions between Condition and Familiarity. With respect to our control variables, total reading times for idioms and literal phrases were shorter for increases in Trial (t = -2.46), Literal Plausibility (t = -2.14), and Reading Proficiency (t = 8.53). As well, increased Character Length yielded longer total reading times (t = 5.14). Figure 3. Idiom Region Total Reading Time as a Function of Familiarity for the Id-Id, Id-Lit, and Lit-Lit Conditions in Native Language Readers Idiom Regressions. The optimal model for Idiom Regressions included fixed effects of Condition, Familiarity, Reading Proficiency, and the interaction between Condition and Familiarity. For the Id-Id vs. Lit-Lit contrast, there was a significant interaction between Condition and Familiarity in that increases in familiarity in the Id-Id condition resulted in fewer idiom regressions, as compared to increases in familiarity in the Lit-Lit condition (Id-Id: 12.80% difference between high and low familiarity, Lit-Lit: 3.20% difference between high and Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 23 low familiarity; z = -3.47). Figure 4 depicts the interaction between condition (Id-Id vs. Lit-Lit) and familiarity. With respect to our control variables, idioms and literal phrases received fewer regressions as Reading Proficiency increased (i.e., when total reading time decreased) (z = 3.51). Figure 4. Proportion of Regressions to Idiom Region as a Function of Familiarity for the Id-Id, Id-Lit, and Lit-Lit Conditions in Native Language Readers Discussion Experiment 1 produced several key findings about idiom processing during natural reading. With respect to first pass reading (first fixation and gaze duration), participants read phrase-final nouns more quickly when embedded in conventional idiomatic phrases vs. matched literal phrases. This suggests that native language readers anticipated phrase-final words for idioms to a greater degree than they did for non-idiomatic sentences, consistent with the view that people can directly retrieve idioms prior to the phrase-final word. As well, nouns of highly decomposable idioms were read more quickly than nouns of less decomposable idioms (gaze Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 24 duration only). However, the interpretation of this result must be tempered by what readers did when they encountered later sentence regions that biased a figurative or literal interpretation of the idiom. To the extent that increased decomposability led to the on-demand computation of a figurative meaning, it should also be associated with faster processing of figurative-biased disambiguating regions, and perhaps even interference for literal-biased disambiguating regions. To the extent that it did not lead to a figurative interpretation but rather allowed readers to more easily resolve any phrasal ambiguity occurring at the end of the phrase, one would not expect a facilitative effect for figurative-biased disambiguating regions. As we will see, the data favor the second option. Consider first, the data for sentences where idioms were followed by idiom-biased contexts (Id-Id) vs. non-idiomatic literal sentences (Lit-Lit). With respect to reading rates of the disambiguating region, Id-Id sentences were comparable to Lit-Lit sentences, suggesting that participants read a figurative-biased context after an idiom no differently than what would have occurred for a non-idiomatic literal sentence. Increased familiarity also yielded significantly fewer regressions to the idiom region, and significantly shorter total reading times of the idiom region for Id-Id sentences compared to Lit-Lit sentences. In contrast with familiarity, decomposability failed to modulate any reading measure for Id-Id sentences relative to Lit-Lit sentences. Consider next, the data for sentences where idioms were followed by literal-biased contexts (Id-Lit) vs. non-idiomatic literal sentences. Reading rates for the disambiguating regions were significantly shorter for sentences containing idioms than for non-idiomatic literal sentences, however, idiom facilitation was significantly greater for high vs. low decomposable idioms. Thus, increased decomposability led to an enhanced (rather than impaired) ability to Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 25 read a literal, non-figurative, context. Increased familiarity was also associated with faster idiom total reading time. Taken together, the first pass effects on the noun combined with the later effects starting at the disambiguating region suggest that people directly retrieve idiomatic configurations during on-line comprehension, and that direct retrieval leads to comprehension of the figurative meaning. In contrast, increased decomposability only helped participants to interpret idioms literally, for example, in the case of the Id-Lit sentences (see also Titone & Connine, 1999). Thus, the results of Experiment 1 are most consistent with idiom processing models that posit a time-dependent role for both direct retrieval and compositional processes (see also Titone & Connine, 1999; Libben & Titone, 2008). Experiment 2 Experiment 2 extended the results of Experiment 1 by investigating how non-native language users, specifically French-English bilinguals for whom English is a second language, read the identical materials under the same conditions as Experiment 1. Second language research demonstrates that non-native language users, especially those who acquired their L2 as an adult, experience great difficulty comprehending and producing idioms in a native-like manner (Conklin & Schmitt, 2008; Nekrasova, 2009; Tabossi, Wolf, & Koterle, 2009; Weinart, 1995; Wood, 2006). For example, Siyanova-Chanturia, Conklin, Schmitt (2011) had native and non-native English users read paragraphs containing idiomatic expressions or literal phrases while their eye movements were recorded. They found that participants read L1 idioms significantly faster than literal phrases, while there was no difference between idioms and literal Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 26 phrases for L2 reading. However, Siyanova-Chanturia et al. did not explicitly examine the effects of linguistic dimensions on comprehension. Of relevance here, L2 idiom processing has been show to be particularly affected by linguistic dimensions such as compositionality and cross-language overlap (Abel, 2003; Irujo, 1993; Laufer, 2000; Steinel, Hulstijn, & Steinel, 2007). Abel showed that non-native English users were more likely to rate English idioms as semantically decomposable compared to our prior study of native English users (Titone & Connine, 1994a; see also, Bortfield, 2003). Presumably, L2 users rely more heavily on compositional processes to interpret idioms figuratively because they have less overall familiarity with L2 idioms, which precludes direct retrieval from memory. Interestingly, however, direct retrieval during L2 idiom processing may still occur to the extent that an L2 idiom has an L1 counterpart. For example, Irujo (1993) showed that cross-language similarity (the degree to which an idiom existed in multiple languages) was a better predictor of whether Spanish idioms would be translated into English than phrasal frequency or decomposability (see also Charteris-Black, 2002; Laufer, 2000). Thus, Experiment 2 addresses two unresolved questions about idiom processing. First, do idioms still exhibit the same processing advantage over literal phrases for L2 readers who are globally less familiar with English idioms? Second, and most importantly, do compositional processes play a more prominent role during idiom comprehension for L2 users, for whom familiarity with English idioms is lower and direct retrieval is less likely? Finally, to what extent does the cross-language similarity of idioms modulate comprehension for L2 users? Regarding the first question, I predicted that idiom-final nouns would lose their processing advantage over nouns embedded in non-idiomatic sentences for L2 users, and may actually incur a processing cost. Regarding the second question, I predicted that increased Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 27 decomposability would exert a significant role in appreciating an idiom's figurative meaning, in contrast with the findings of Experiment 1 with L1 users. Regarding the final question, I predicted that L2 idiom processing (for French L1 users) would be easier to the extent that English idioms overlapped with French idioms (e.g., play with fire and jouer avec le feu). Method Participants Thirty-seven non-native English users from the McGill and Montreal community, whose first language was French, participated for course credit or compensation of $10/hr. Participants completed a survey assessing both French and English language history and ability. All participants reported French as the first acquired language and their dominant language used daily. They also had normal or corrected-to-normal vision and no self-reported history of speech or hearing disorders. Stimuli & Apparatus The stimuli and apparatus were identical to those used in Experiment 1. Procedure The procedure was identical to that used in Experiment 1 except that after the reading task, participants completed a survey involving the 60 idioms used in the experiment. For each idiom, participants indicated whether they knew of an equivalent idiom in French. If an equivalent expression existed, participants wrote the equivalent French expression. From this procedure, a set of L1/L2 cross-language similarity ratings (L1/L2 Overlap) was amassed, to be Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 28 used as predictors in subsequent analyses. Again, comprehension questions occurred on 20% of trials, and accuracy across all participants was 86%, indicating that participants were attentive during the experiment. Results The analytic procedure was identical to Experiment 1, except that L1/L2 Overlap ratings were used rather than L1 Familiarity ratings as a predictor for two important reasons. First, native English familiarity ratings from Libben and Titone (2008) would be invalid indicators of idiom familiarity for non-native English users. Second, ratings of L1/L2 Overlap would allow us to test hypotheses about cross-language idiom processing effects. The L1/L2 Overlap measure is summarized in Table 2. Before analysis, 2.7% of total fixations were removed because of track loss or fixation durations less than 80 ms. Again, an upper cutoff of 4 standard deviations above the mean for each DV was established. When analyzing Noun FFD and Noun GD, all trials in which the noun was skipped (n = 528, 23.8% of all Noun FFD and Noun GD trials), or the verb or determiner of the idiom/literal phrase was not fixated prior to fixating on the noun during the first pass (n = 70, 3.0% of all Noun FFD and Noun GD trials) were excluded. When analyzing Disambiguating FPGD, Idiom TRT, and Idiom Regression data, I again excluded any trials in which the entire idiomatic phrase was skipped on the first pass (n = 12, 0.5% of all Disambiguating FPGD, Idiom TRT, and Idiom Regression trials). Finally, I again removed trials from Disambiguating FPGD data if the first pass gaze duration was equal to the first fixation duration of the disambiguating region for that same trial (n = 166, 7.5% of all Disambiguating FPGD trials). Similar to Experiment 1, subject means and the number of observations for each dependent variable binned Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 29 across high and low L1/L2 Overlap and Decomposability in each condition are presented in Table 7. These means were included to facilitate comparison to traditional F1 analyses; they do not reflect the actual LME procedure, in which L1/L2 Overlap and Decomposability were treated as continuous predictors. Table 7. Total Number of Observations, Means, and Standard Deviations Across All Participants for All Eye Movement Measurements (Noun FFD, Noun GD, Disambiguating FPGD, Idiom TRT, Idiom Regressions) of Non-native Language Readers In Each Condition (Idiom, Literal, Id-Id, Id-Lit, Lit-Lit) for High and Low Levels of L1/L2 Overlap and Decomposability for Non-Native English Speakers in Experiment 2 Eye-Movement Measure 1 Condition n Idiom 1097 Literal 516 1613 All Idiom 1093 Noun GD1 Literal 517 1610 All 681 Disambiguating FPGD2 Id-Id Id-Lit 673 Lit-Lit 681 2035 All Id-Id 730 Idiom TRT1 Id-Lit 729 Lit-Lit 732 2191 All Id-Id 735 Idiom Regressions Id-Lit 735 Lit-Lit 735 2205 All 1 Values presented in milliseconds 2 Values presented in milliseconds per character Note. Standard deviations are listed in parentheses Noun FFD Condition Mean 249 (85) 239 (79) 246 (84) 271 (107) 266 (108) 270 (107) 41.2 (15) 41.5 (16) 41.4 (15) 41.4 (15) 874 (460) 856 (447) 825 (411) 852 (440) 40.00% 36.90% 37.30% 38.00% L1/L2 Overlap Decomposability Low High Low High 252 (88) 230 (70) 245 (83) 274 (111) 264 (111) 271 (111) 40.4 (15) 42.6 (16) 41.1 (15) 41.4 (15) 937 (488) 887 (463) 846 (425) 890 (460) 45.10% 38.30% 40.40% 41.30% 246 (83) 246 (85) 246 (84) 268 (103) 268 (107) 268 (104) 41.9 (16) 40.6 (15) 41.5 (14) 41.4 (15) 819 (426) 829 (430) 807 (398) 818 (418) 35.50% 35.60% 34.50% 35.20% 248 (87) 239 (79) 245 (84) 275 (114) 272 (112) 274 (113) 41.0 (16) 42.5 (17) 42.3 (15) 42.0 (16) 930 (481) 884 (468) 821 (392) 878 (451) 45.00% 40.10% 37.50% 40.90% 250 (84) 238 (79) 246 (83) 267 (100) 261 (104) 265 (101) 41.5 (15) 40.5 (14) 40.4 (14) 40.8 (15) 818 (430) 828 (422) 830 (429) 825 (427) 35.10% 33.50% 37.00% 35.20% Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 30 First Pass Effects on Phrase-final Nouns I constructed initial LME models identical to those used in Experiment 1, with the exception that L1/L2 Overlap replaced Familiarity as a primary predictor. For significant effects, I report differences in means derived from Table 7 and t-values derived from Table 8. For control predictor variables, I report only t-values for significant effects derived from optimal LME models, summarized in Table 8. Table 8. Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Noun Region First Fixation Duration and Noun Region Gaze Duration for Non-Native English Speakers in Experiment 2 Noun Region First Fixation Duration Gaze Duration b SE t-value b SE t-value Fixed effects1 Primary Variables Condition (Literal, Idiom)2 9.42 4.09 2.30* L1/L2 Overlap 5.23 3.6 1.45 Decomposability Condition*L1/L2 Overlap -10.19 4.12 -2.47* Condition*Decomposability L1/L2 Overlap*Decomposability Condition*L1/L2 Overlap*Decomposability Control Variables Trial Literal Plausibility -6.05 2.21 -2.74** -11.65 3.55 -3.28** Reading Proficiency 16.39 5.18 3.17** 26.58 5.57 4.77*** Character Length 10.69 3.57 2.99** (Intercept) 239.51 5.94 40.30*** 268.95 6.08 44.21*** Random Effects Variance Variance Subject 839.56 913.28 Item 70.21 374.27 Residual 5791.20 9468.48 * Significant at the .05 level ** Significant at the .01 level *** Significant at the .001 level 1 Fixed effects that significantly improved the model where was p<.05 during likelihood ratio tests 2 Model assumes literal condition as the baseline across conditions Note. All continuous fixed effects are standardized, with the exception of Trial Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 31 Noun FFD. The optimal model for Noun FFD included Condition (Idiom vs. Literal), L1/L2 Overlap, Literal Plausibility, Reading Proficiency, and the interaction between Condition and L2 Overlap. There was a significant effect of Condition in that nouns at the ends of idioms received longer first fixation durations than nouns at the ends of literal phrases (10 ms difference between Idiom and Literal; t = 2.03). There was a significant interaction between Condition and L1/L2 Overlap, in that increased L1/L2 Overlap facilitated first fixation durations in the idiom condition more than it did in the literal condition (Idiom: -6 ms between high and low L1/L2 Overlap, Literal: 16 ms between high and low L1/L2 Overlap; t = -2.47). The interaction between Condition and L1/L2 Overlap is depicted in Figure 5. With respect to our control predictors, first fixation durations were shorter as Literal Plausibility increased (t = -2.74) and as Reading Proficiency increased (t = 3.17). Figure 5. Noun First Fixation Duration as a Function of L1/L2 Overlap for the Idiom and Literal Conditions in Non-Native Language Readers Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 32 Noun GD. The optimal model for Noun GD included fixed effects of Literal Plausibility, Reading Proficiency, and Character Length. Gaze durations for nouns were shorter as Literal Plausibility (t = -3.28) and Reading Proficiency (t = 4.77) increased, and as Character Length increased (t = 2.99). Disambiguating Region and Subsequent Idiom Effects I again constructed LME models identical to those used in Experiment 1 except that Familiarity was replaced with L1/L2 Overlap. Again, for significant effects, I report differences in means derived from Table 7 and t-values derived from Table 9. For control predictor variables, I report only t-values for significant effects, summarized in Table 9. Disambiguating Region FPGD. The optimal model for Disambiguating FPGD included only the fixed effect of Reading Proficiency. Disambiguating regions received shorter first pass gaze durations as reading proficiency increased (t = 3.97). Idiom Region TRT. The optimal model for Idiom TRT included fixed effects of Condition, Decomposability, Trial, Literal Plausibility, Reading Proficiency, Character Length, and the interaction between Condition and Decomposability. The Id-Lit vs. Lit-Lit contrast was not significant alone or in any interaction. However, for the Id-Id vs. Lit-Lit contrast, total reading times for idioms in the Id-Id condition were longer than those for literal phrases in the Lit-Lit condition (49 ms difference; t = 2.49). There was also a significant interaction between Condition and Decomposability in that increased decomposability was more facilitative in the IdId condition than in the Lit-Lit condition (Id-Id: -112 ms difference between high and low Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 33 Table 9. Effect Size, Standard Error, and t-Values for Fixed Effects and Variances for Random Effects Derived From Optimal LMER Models for the Dependent Variables Disambiguating Region First Pass Gaze Duration, Idiom Region Total Reading Time, and Idiom Region Regressions for Non-Native English Speakers in Experiment 2 Disambiguating Region First Pass Gaze Duration b SE t-value Idiom Region Total Reading Time b SE t-value b Fixed effects1 Primary Variables Condition (Lit-Lit Id-Id)2 47.07 2 Condition (Lit-Lit Id-Lit) 30.87 L1/L2 Overlap Decomposability 6.14 Condition (Id-Id)*L1/L2 Overlap Condition (Id-Lit)*L1/L2 Overlap Condition (Id-Id)*Decomposability -66.19 Condition (Id-Lit)*Decomposability -26.09 L1/L2 Overlap*Decomposability Condition (Id-Id)*L1/L2 Overlap*Decomposability Condition (Id-Lit)*L1/L2 Overlap*Decomposability Control Variables Trial -1.88 Literal Plausibility -32.14 Reading Proficiency 8.34 0.87 9.56*** 208.21 Character Length 69.23 (Intercept) 41.39 0.91 45.69*** 884.17 Random Effects Variance Subject 25.81 Item 3.39 Residual 134.06 * Significant at the .05 level ** Significant at the .01 level *** Significant at the .001 level 1 Fixed effects that significantly improved the model where was p<.05 during likelihood ratio tests 2 Model assumes literal condition as the baseline across conditions Note. All continuous fixed effects are standardized with the exception of Trial 18.88 18.89 15.97 18.91 18.88 - 2.49* 1.63 0.38 -3.50*** -1.38 - 0.45 -4.19*** 11.82 -2.72** 17.45 11.93*** 11.66 5.94*** 26.34 33.57*** Variance 9136.20 4659.50 130143.70 0.14 -0.02 -0.15 0.01 -0.33 -0.07 - Regressions SE z-value 0.12 0.17 0.06 0.09 0.12 0.12 - 1.17 -0.15 -2.32* 0.09 -2.82** -0.63 - 0.56 0.14 3.97*** -0.65 0.16 -4.15*** Variance 0.62 0.05 n/a Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 34 decomposability, Lit-Lit: 9 ms difference between high and low decomposability; t = -3.50). The interaction between Condition (Id-Id vs. Lit-Lit) and Decomposability is depicted in Figure 6. With respect to our control variables, total reading times for idioms and literal phrases were shorter as Trial (t = -4.19), Literal Plausibility (t = -2.72), and Reading Proficiency (t = 11.93) increased, and as Character Length decreased (t = 5.94). Figure 6. Idiom Region Total Reading Time as a Function of Decomposability for the Id-Id, IdLit, and Lit-Lit Conditions in Non-Native Language Readers Idiom Regressions. The optimal model for Idiom Regressions included fixed effects of Condition, L1/L2 Overlap, Decomposability, Reading Proficiency, and the interaction between Condition and Decomposability. There was a significant effect of L1/L2 Overlap in that idioms and literal phrases were regressed into less often as L1/L2 Overlap increased (-6.10% difference between high and low L1/L2 Overlap; z = -2.32). For the Id-Id vs. Lit-Lit contrast, there was a significant interaction between Condition and Decomposability in that increases in decomposability in the Id-Id condition resulted in fewer idiom regressions than that found for the Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 35 Lit-Lit condition (Id-Id: -9.90% difference between high and low familiarity, Lit-Lit: 0.05% difference between high and low decomposability; z = -2.82). The Condition and Decomposability interaction is depicted in Figure 7. With respect to our control predictors, as Reading Proficiency increased, idioms and literal phrases received fewer regressions (z = 3.97). Figure 7. Idiom Region Total Reading Time as a Function of Decomposability for the Id-Id, IdLit, and Lit-Lit Conditions in Non-Native Language Readers Discussion Experiment 2 produced several key findings regarding L2 idiom comprehension. First, in contrast with native English users (Experiment 1), non-native English users read idiom-final nouns more slowly than nouns occurring at the ends of literal phrases. This suggests that L2 readers did not directly retrieve the phrase-final words of English idioms, but rather they experienced comprehension difficulty when idiom-final nouns were encountered. However, the costs associated with idiom-final nouns were significantly modulated by whether the idiom in Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 36 question also existed in French, their L1. In particular, phrase-final nouns for uniquely English idioms were read more slowly than phrase-final nouns of literal sentences, whereas phrase-final nouns for English idioms that also existed in French were read more quickly than phrase-final nouns of literal sentences. Thus, similar to prior work for idioms (e.g., Irujo, 1993), and also for single words (Duyck, van Assche, Drieghe, Hartsuiker, 2007; Libben & Titone, 2009; Van Assche et al., 2011), idioms that have cross-language similarity are easier for readers to comprehend in their L2. A second important finding was that L2 users had difficulty processing figurative biased contexts when they encountered the disambiguating region. That is, L2 users spent more total time reading idiomatic expressions than literal phrases when the idioms were followed by figurative contexts. However, difficulty processing figurative-biased contexts was mediated by the decomposability of the idioms. As idioms increased in decomposability, L2 users made fewer regressions and spent less total time reading idioms after reading figurative-biased contexts. Thus, unlike Experiment 1, where increased decomposability facilitated comprehension of a literal interpretation of idioms, increased decomposability in Experiment 2 facilitated comprehension of a figurative interpretation for L2 readers. Presumably this is because nonnative language users are less familiar with idioms in their L2, and are less likely to directly retrieve them from memory (Abel, 2003; Irujo, 1993; Laufer, 2000; Tabossi, Wolf, & Koterle, 2009). Consequently, they must rely on compositional analyses to lead them towards an understanding of the idioms’ figurative meaning. Taken together, the results of Experiment 2 build upon those of Experiment 1, and provide support for models of idiom comprehension that incorporate both direct retrieval and compositional processes in the L1, as well as the L2. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 37 General Discussion I examined the degree to which L1 and L2 readers directly retrieved or compositionally constructed the figurative meanings of idioms during natural reading. In two experiments, native English (Experiment 1) and non-native English (Experiment 2) participants read sentences containing idioms or matched literal phrases, followed by subsequent contexts that specifically biased their figurative or literal meanings. In both experiments, several eye movement measures were analyzed that reflected both early and late stages of comprehension. In Experiment 1, L1 readers processed idiom-final nouns more easily than literal phrasefinal nouns on the first pass. This early comprehension advantage for idioms over literal phrases is similar to that reported in previous studies (e.g., Cronk & Schweigert, 1992; Gibbs, 1980, McGlone, Cacciari, & Glucksberg, 1994; Swinney & Cutler, 1979). While increased familiarity significantly facilitated reading times for nouns in the L1, the amount of facilitation did not differ between idioms and their literal counterparts as a function of familiarity. In contrast, increased decomposability had a facilitative effect on early reading times for idiom-final nouns as compared to literal phrase-final nouns. This result could have arisen for at least two different reasons. One possibility is that increased decomposability facilitates a compositional analysis leading to a figurative interpretation an idiomatic string, in agreement with the work of Gibbs and colleagues (e.g., Gibbs, Nayak, & Bolton, & Keppel, 1989). A second possibility is that increased decomposability leads to reduced phrasal or lexical ambiguity for idioms (see Titone & Connine, 1999; Colombo 1998). In this view, nondecomposable idioms would be analogous to homographs that possess semantically distinct meanings, whereas decomposable idioms would be analogous to polysemous words that possess semantically overlapping meanings. Thus, like Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 38 what has been found for homonyms and polysemous words, nondecomposable idioms may elicit processing difficulty because people must select between two distinct meanings, whereas decomposable idioms do not elicit processing difficulty because the directly retrieved figurative meaning and literal interpretation of the phrase are semantically compatible (see Frazier & Rayner, 1990; Klepousniotou, Titone, & Romero, 2008, for evidence of easier comprehension for polysemous vs. homonymous words). Indeed, the pattern of data for later measures as a function of decomposability and the particular interpretation instantiated by a sentence context was most consistent with the latter option. Native English readers processed subsequent figurative-biased portions of Id-Id sentences no differently than non-idiomatic literal sentences, however, they were less likely to regress to the idiom region for Id-Id sentences to the extent that an idiom was familiar. In contrast, increased decomposability exerted no effect on L1 reading for Id-Id sentences. The increased tendency to regress to earlier portions of the sentence for low familiar idioms was also reflected in terms of reduced total reading time on the idiom region. Thus, it appears that the early facilitative effect of decomposability in the L1 was due to the figurative and literal meaning of idioms being more-or-less polysemous, or phrasally ambiguous for decomposable vs. nondecomposable idioms (e.g., Titone & Connine, 1999; Colombo, 1998). Accordingly, when L1 readers directly retrieve idiomatic meanings when first encountering the phrase-final word, if an idiomatic meaning conflicts with a normal compositional (i.e., literal) analysis of the string, comprehension will be relatively slow. In contrast, if a directly retrieved figurative meaning is semantically compatible with a normal compositional (i.e., literal) analysis of the string, comprehension will be relatively fast. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 39 The effects of decomposability and familiarity, however, were quite different for Id-Lit sentences compared to Id-Id sentences. Here, the effect of condition interacted with both familiarity and decomposability but in opposite ways. Consider first, the effects of familiarity. Increased familiarity slowed reading times for disambiguating contexts in Id-Lit sentences, consistent with the view that direct retrieval of idiomatic representations leads to a figurative interpretation. However, increased familiarity did not increase regressions to the idiom region for Id-Lit sentences. This pattern contrasts with the effect of familiarity for Id-Id sentences, where increased familiarity did not affect reading rates for the disambiguating region but did affect whether people regressed to the idiom from the disambiguating region (i.e., people regressed significantly more for low familiar idioms). Thus, when an idiom is high familiar, and its figurative meaning is relatively more dominant than its literal phrasal meaning, readers directly retrieve its figurative meaning such that a subsequent context that is literally biased is harder but not impossible to read (i.e., it slows reading but does not cause a regressive eye movement). This pattern of data suggests that the literal phrasal meaning must be available to some degree, even for highly familiar idioms (see also Peterson, Burgess, Dell, & Eberhard, 2001). In contrast, when an idiom is not familiar, and a subsequent context is figuratively biased, reading is fully disrupted in that regressive eye movements occur. Thus, the relative amount of figurative vs. literal meaning dominance may not be perfectly symmetrical for high vs. low familiar idioms. Specifically, the figurative meaning of high familiar idioms does not fully trump a literal interpretation of the phrase, whereas the literal phrase meaning of low familiar idioms does fully trump a figurative interpretation of the phrase, presumably because the reader does not have the figurative meaning represented in memory. Thus, idiom processing may have Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 40 other similarities to how lexically ambiguous words are processed with respect to the effects of meaning dominance (see Simpson, 1981; Simpson & Burgess, 1985). Consider next, the effects of decomposability on Id-Lit sentences. To the extent that decomposability leads to a figurative interpretation, one would expect that literally biased disambiguating regions of highly decomposable idioms would be more difficult to read, similar to the effects of increased familiarity. However, this was not the case, and in fact, the reverse was true. Specifically, increased decomposability facilitated reading times for disambiguating contexts in Id-Lit sentences. Thus, consistent with the results and conclusions of Titone and Connine (1999), it appears that decomposability may have more to do with the reanalysis of an idiom’s interpretation, and in particular, when a figurative meaning must be suppressed to make way for an intended literal phrasal interpretation. It would not, however, be involved in generating a figurative interpretation as there was no significant effect of decomposability on how people read figuratively biased disambiguating regions (i.e., Id-Id sentences). Although the design of Experiments 1 and 2 were not identical, non-native English L2 readers in Experiment 2 yielded a different overall pattern of effects than native English L1 readers for both early and late measures when reading English idioms. First, regarding early reading measures, L2 readers demonstrated overall difficulty comprehending idioms vs. literal sentences, which most likely arose from a lack of familiarity with figurative meanings of English, L2 idioms (but see Siyanova-Chanturia, Conklin, & Schmitt, 2011). Interestingly, however, L2 idiom processing was less difficult when L2 idioms possessed L1 translation equivalents (i.e., the English idiom he broke the ice has the French equivalent il a brisé la glace, whereas the English idiom she had a lark has no French equivalent), consistent with other L2 learning studies (e.g., Irujo, 1993; Laufer, 2000). This effect is also intriguing when one Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 41 considers the large body of literature showing facilitation for words that overlap in form and semantics across language (i.e., cognates - words that overlap both orthographically and semantically between the L1 and L2; see Dijkstra et al., 2010; Duyck et al., 2007; Libben & Titone, 2009; Schwartz & Kroll, 2006; Titone, Libben, Mercier, Whitford, & Pivneva, in press). At the word level, cognate facilitation presumably arises from convergent lexical activation across the L1 and L2, given that bilingual language processing is overwhelmingly non-selective in nature (De Groot, 2011; Dijkstra et al., 2010). Our data suggest the same is true for idioms. However, the remarkable difference between word and “phrasal” cognates is that for phrases, component words may not be identical across languages, although their phrasal relation to each other may be highly similar. Regarding late-stage reading measures for Experiment 2, L2 readers again had difficulty computing the figurative meaning of English idioms, consistent with the pattern of early effects, as well as previous work on L2 idiom comprehension (e.g., Abel, 2003; Wray 2002). However, in contrast with what was found for L1 readers in Experiment 1, increased decomposability now facilitated reading for Id-Id sentences. Specifically, L2 readers generated fewer regressions to the idiom region when reading the disambiguating region of Id-Id sentences, to the extent that an idiom was decomposable. Similarly, L2 readers spent less total reading time on the idiom region to the extent that an idiom was decomposable. Thus, unlike L1 readers, for whom the figurative interpretations of conventional English idioms are generally represented in memory and directly retrieved during comprehension, L2 readers have no choice but to compositionally construct figurative interpretations of idioms. Thus, when an idiom is decomposable, and the products of a compositional analysis are semantically close to a figurative interpretation, comprehension is Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 42 easy. However, when an idiom is nondecomposable, and the products of a compositional analysis are semantically distant from a figurative interpretation, comprehension is difficult. Taken together, the results of Experiments 1 and 2 are key for adjudicating between different models of idiom comprehension. In contrast with the predictions of strict direct retrieval models (e.g., Swinney & Cutler, 1979; Cacciari & Tabossi, 1988), increased decomposability does indeed play a role during both L1 and L2 idiom comprehension. For L1 idiom processing, increased decomposability reduces phrasal ambiguity when an idiom is initially read, and it also helps in reanalyzing an idiom literally should a subsequent context call for that interpretation. However, in contrast with strict compositional models (Gibbs, Nayak, & Bolton, & Keppel, 1989), increased decomposability does not directly lead to a figurative interpretation of idioms during L1 idiom processing, rather figurative interpretations of conventional idioms appear instead to be directly retrieved from memory. Of note, when the same idioms are read by L2 readers, for whom English L1 idioms are less known or familiar as a group, both cross-language overlap, an index of direct retrieval for L2 readers, and increased decomposability facilitates idiom processing. Thus, the overall pattern of findings best supports a hybrid or multidetermined model of comprehension (Titone & Connine, 1999; Libben & Titone, 2008), according to which people simultaneously use all available information during comprehension (resulting from direct retrieval and compositional analysis); different kinds of information may be available at different time-courses; and different classes of information interact over time. Because all sources of information interact, if the products of direct retrieval are particularly strong or salient, they may overwhelm and functionally terminate the products of compositional processes. In contrast, if the products of direct retrieval are weak, as they might be for low familiar idioms, an ongoing Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 43 compositional analysis might instead dominate. Consistent with this view, and the data presented here, our prior work shows that non-idiomatic literal word meaning activation is reduced for highly predictable and literally implausible idioms, and that figurative meanings of highly predictable idioms are retrieved prior to encountering their phrase-final words (Titone & Connine, 1994b). Conversely, in other work, it was found that increased decomposability facilitated meaningfulness judgments for unfamiliar idioms but not highly familiar idioms (Libben & Titone, 2008). Thus, this study and our prior work contribute to the growing body of results that favor hybrid models of idiom processing such as that proposed by Titone and Connine (1999), and subsequently by others (Abel, 2003; Caillies and Butcher, 2007; Smolka, Rabanus, & Rosler, 1997; Sprenger, Levelt, & Kempen, 2006). An important corollary of the hybrid, or multidetermined approach, is that any factor that independently modulates direct retrieval or compositional processes should also modulate figurative meaning generation and integration during comprehension; these include individual differences among idioms or people. Thus, in this study I used this model to make predictions about second language learners, and many further untested predictions remain. For example, it is unclear whether the same pattern of familiarity and compositionality effects on idiom processing would hold when disambiguating contexts precede idioms during sentence comprehension. Prior contexts generally facilitate idiom processing (e.g., Fanari, Cacciari, & Tabossi, 2010; Mueller & Gibbs, 1987; Titone & Connine, 1999), however, it is still unclear whether it does so be aiding direct retrieval, compositional processing, or both. While the effects of a prior context would most likely arise from pre-activation of an idiomatic form in memory, it is possible that a prior context would make the semantic relation between an idiom’s figurative meaning and component words more salient on the first pass, thus increasing the likelihood that the figurative Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 44 interpretation is partially built compositionally as well as directly retrieved. A prior context that is figuratively biased might also cause relatively more suppression of a literal phrasal interpretation than what was found in the present study. These are important issues for further study, which are more easily characterized by a hybrid model of idiom processing compared to pure direct retrieval or compositional models. Another target for future research concerns the nature of the direct retrieval process with respect to idiom processing. One important issue here concerns what is directly retrieved during idiom processing. Direct retrieval could involve a linguistic form representation independent of a phrasal meaning, as well as the meaning-based interpretation linked in memory to a particular form. This distinction is particularly salient for idioms in that idiom-final words might be easier to process than phrase-final words for non-idiomatic phrases simply on the basis of high word co-occurrence for idioms vs. non-idioms, or differences among idioms in their structural fixedness. Regarding the latter point, classes of idioms other than verb-determiner-noun phrases, such as verb-particle constructions, are more syntactically promiscuous and may take any number of nouns as their direct objects (carry out the project, carry the project out) (e.g., Jackendoff, 2002, 2003; Konopka & Bock, 2009). Thus, direct retrieval for verb-particle idioms may also involve provisions about their structural flexibility, which interact with word cooccurrence patterns. Moreover, compositional processes may in fact play more of a role for these structurally flexible idiomatic forms. Finally, beyond the lexical form level, as seen here, idiomatic phrases have stored meaning-based interpretations that can be retrieved, and that may conflict with an ongoing literal (compositional) analysis of the string. Thus, one question is whether the semantic level of representation for idioms is in fact distinct from the lexical form level representation. Recent electrophysiological work posits different event related potentials Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 45 for idiom form vs. meaning activation (Vespignani et al., 2010; Molinaro & Carreiras, 2010), however, differences among idiomatic and non-idiomatic collocations in phrasal ambiguity in these studies preclude a straightforward interpretation of the ERP effects. In addition to questions about what is directly retrieved during idiom processing, a second issue concerns how the direct retrieval process may transpire. One option is that direct retrieval during idiom processing is a kind of template-matching process where comprehenders compare an unfolding input to fixed canonical form of a particular idiom in memory. This approach may be overly rigid, however, and unable to account for data showing that people readily comprehend and produce idioms outside of their canonical forms (e.g., Gibbs & Nayak, 1989; Gibbs & O’Brien, 1990; McGlone, Cacciari, & Glucksberg, 1994). Moreover, given the structurally flexibility of many idioms, it is unclear whether it would be possible to identify a single canonical form or template for a large proportion of idioms, and indeed, perhaps all idioms. A second option, however, is that direct retrieval is a kind of anticipatory or prediction process (e.g., Vespignani et al., 2010), where people generate lexical and semantic expectations about upcoming words as language unfolds. This view has substantial advantages over a rigid template-matching view, in that it can more easily account for the comprehension of idiom variants, and it more naturally integrates the processing of idioms with what we do normally to comprehend all forms of language. These data do not distinguish between these two options, as I only presented idioms in relatively canonical forms. However, it is possible that future studies of how the structural characteristics of idioms alter the relative balance of direct retrieval vs. compositional processing during comprehension would be better suited to this aim. To conclude, consistent with hybrid or multidetermined models of comprehension (e.g., Titone & Connine, 1999; Libben & Titone, 2008), idiom processing appears to arise from both Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 46 direct retrieval and compositional processes, which follow different time-courses for L1 and L2 readers (e.g., Libben & Titone, 2008). For L1 readers, increased familiarity leads to fast direct retrieval of the figurative meanings, although compositional processes may help for interpreting an idiom literally (Titone & Connine, 1999). For L2 readers, direct retrieval, via L1 knowledge, and compositional processes both increase the figurative interpretation of idioms. Whether such a view extends to all language contexts, to language production, or to all idiomatic forms remains open questions for future work. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 47 References Abel, B. (2003). English idioms in the first language and second language lexicon: A dual representation approach. Second Language Research, 19, 329-358. Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R. New York, NY: Cambridge University Press. Baayen, H. R., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory & Language, 59, 390-412. Bates, D., & Sarkar, D. (2006). lme4: Linear mixed-effects modeling using S4 classes R package (Version 0.9975-10) [Computer Software]. Vienna, Austria: R Foundation for Statistical Computing. Bobrow, S. A., & Bell, S. M. (1973). On catching on to idiomatic expressions. Memory & Cognition, 1, 343-346. Bortfeld, H. (2003). Comprehending idioms cross-linguistically. Experimental Psychology, 50, 217-230. Cacciari, C., & Tabossi, P. (1988). The comprehension of idioms. Journal of Memory & Language, 27, 668-683. Caillies, S., & Butcher, K. (2007). Processing of idiomatic expressions: Evidence for a new hybrid view. Metaphor & Symbol, 22, 79-108. Charteris-Black, J. (2002). Second language figurative proficiency: A comparative study of Malay and English. Applied Linguistics, 23, 104-133. Chomsky, N. (1980). Rules and representations. Behavioral and Brain Sciences, 3, 1-15. Colombo, L. (1998). Role of context in the comprehension of ambiguous Italian idioms. Syntax and Semantics, 31, 379-404. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 48 Conklin, K., & Schmit,. N. (2008). Formulaic sequences: Are they processed more quickly than nonformulaic language by native and nonnative speakers. Applied Linguistics, 29, 72-89. Cronk, B.C. & Schweigert, W.A. (1992). The comprehension of idioms: The effects of familiarity, literalness, and usage. Applied Psycholinguistics, 13, 131-146. De Groot, A. M., & Nas, G. L. (1991). Lexical representation of cognates and noncognates in compound bilinguals. Journal of Memory and Language, 30, 90-123. De Groot, A. M. B. (2011). Language and cognition in bilinguals and multilinguals: An introduction. New York-Hove: Psychology Press. Dijkstra, T., Miwa, K., Brummelhuis, B., Sappelli, M., & Baayen, H. (2010). How crosslanguage similarity and task demands affect cognate recognition. Journal of Memory and Language, 62, 284-301. Duyck, W., van Assche, E., Drieghe, D., & Hartsuiker, R. (2007). Visual word recognition by bilinguals in a sentence context: Evidence for nonselective lexical access. Journal of Experimental Psychology: Learning, Memory, & Cognition, 33, 663-679. Fanari, R., Cacciari, C., & Tabossi, P. (2010). The role of idiom length and context in spoken idiom comprehension. European Journal of Cognitive Psychology, 22, 321-334. Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 2242. Frazier, L., Rayner, K. (1990). Taking on semantic commitments: Processing multiple meanings vs. multiple senses. Journal of Memory and Language, 29, 181-200. Gibbs, R. W. (1980). Spilling the beans on understanding and memory for idioms in conversation, Memory & Cognition, 8, 149-156. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 49 Gibbs, R. W. (1992). What do idioms really mean? Journal of Memory and Language, 31, 485506. Gibbs, R. W., & Nayak, N. P. (1989). Psycholinguistic studies on the syntactic behavior of idioms, Cognitive Psychology, 21, 100-138. Gibbs, R.W., Nayak, N. P., Bolton, J. L., Keppel, M. E. (1989). Speakers’ assumptions about the lexical flexibility of idioms. Memory & Cognition, 17, 58-68. Gibbs, R.W., Nayak, N. P., & Cutting, J. C. (1989). How to kick the bucket and not decompose: Analyzability and idiom processing. Journal of Memory & Language, 28, 576-593. Gibbs, R. W., & O’Brien, J. E. (1990). Idioms and mental imagery: The metaphorical motivation for idiomatic meaning. Cognition, 36, 35-68. Giora, R. (1997). Understanding figurative and literal language: The graded salience hypothesis (Psycholinguistics). Cognitive Linguistics, 8, 183-206. Hamblin, J. L., & Gibbs, R. W. (1999). Why you can’t kick the bucket as you slowly die: Verbs in idiom comprehension. Journal of Psycholinguistic Research, 28, 25-39. Irujo, S. (1993). Steering clear: Avoidance in the production of idioms. International Review of Applied Linguistics in Language Teaching, 31, 205-219. Jackendoff, R. (2002). English particle constructions, the lexicon, and the autonomy of syntax. In N. Dehe, R. Jackendoff, A. McIntyre, & S. Urban (Eds.), Verb-particle explorations (pp. 67-94). Berlin, Germany: Mouton de Gruyter. Jackendoff, R. (2003). Foundations of language: Brain, meaning, grammar, evolution. New York, NY: Oxford University Press. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 50 Klepousniotou, E., Titone, D., & Romero, C. (2008). Making sense of word senses: The comprehension of polysemy depends on sense overlap. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34, 1534-1543. Kliegl, R., Risse, S., & Laubrock, J. (2007). Preview benefit and parafoveal-on-foveal effects on word n + 2. Journal of Experimental Psychology: Human Perception & Performance, 33, 1250-1255. Konopka, A. E., & Bock, K. (2009). Lexical or syntactic control of sentence formulation? Structural generalizations from idiom production. Cognitive Psychology, 58, 68-101. Kučera, H., & Francis, W. (1967). Computational analysis of present day American English. Providence, RI: Brown University Press. Laufer, B. (2000). Avoidance of idioms in a second language: The effect of L1-L2 degree of similarity. Studia Linguistica, 54, 186-196. Libben, M. R., & Titone, D. A. (2008). The multidetermined nature of idiom processing. Memory & Cognition, 36, 1103-1121. Libben, M. R., & Titone, D. A. (2009). Bilingual lexical access in context: evidence from eye movements during reading. Journal of Experimental Psychology: Learning, Memory, & Cognition, 35, 381-390. McGlone, M. S., Cacciari, C., & Glucksberg, S. (1994). Semantic productivity and idiom and idiom comprehension. Discourse Processes, 17, 167-190. Molinaro, N., & Carreiras, M. (2010). Electrophysiological evidence of interaction between contextual expectation and semantic integration during the processing of collocations. Biological Psychology, 83, 176-190. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 51 Mueller, R. A. G., & Gibbs, R. W. (1987). Processing idioms with multiple meanings. Journal of Psycholinguistic Research, 16, 63-81. Nekrasova, T. M. (2009). English L1 and L2 speakers' knowledge of lexical bundles. Language Learning, 59, 647-686. Nunberg, G. (1978). The Pragmatics of reference. Indiana University Linguistics Club. Nunberg, G., Sag, I.A., & Wasow, T. (1994). Idioms. Language, 70, 491-538. Peterson, R. R., Burgess, C., Dell, G. S., & Eberhard, K. A. (2001). Dissociation between syntactic and semantic processing during idiom comprehension. Journal of Experimental Psychology-Learning, Memory, and Cognition, 27, 1223-1237. Prado, E. L., & Ullman, M. T. (2009). Can imageability help us draw the line between storage and competition? Journal of Experimental Psychology- Learning, Memory, and Cognition, 35, 849-866. Quene, H., & van den Bergh, H. (2008). Examples of mixed-effects modeling with crossed random effects and with binomial data. Journal of Memory & Language, 59, 413-425. R Development Core Team. (2006). R: A language and environment for statistical computing. (Version 2.3.1) [Computer Software]. Vienna, Austria: R Foundation for Statistical Computing. Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124, 372-422. Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual search. Quarterly Journal of Experimental Psychology, 62, 1457-1506. Schwartz, A. I., & Kroll, J. F. (2006). Bilingual lexical activation in sentence context. Journal of Memory and Language, 55, 197-212. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 52 Schweigert, W. A. (1986). The comprehension of familiar and less familiar idioms. Journal of Psycholinguistic Research, 15, 33-45. Simpson, G. B. (1981). Meaning dominance and semantic context in the processing of lexical ambiguity. Journal of Verbal Learning and Verbal Behavior, 20, 120-136. Simpson, G. B., & Burgess, C. (1985). Activation and selection processes in the recognition of ambiguous words. Journal of Experimental Psychology: Human Perception and Performance, 11, 28-39. Siyanova-Chanturia, A., Conklin, K., & Schmitt, N. (2011). Adding more fuel to the fire: An eye-tracking study of idiom processing by native and non-native speakers. Second Language Reasearch, 27, 251-272. Smolka, E., Rabanus, S., & Rosler, F. (2007). Processing verbs in German idioms: Evidence against the configuration hypothesis. Metaphor and Symbol, 22, 213-231. Sprenger, S. A., Levelt, W. J. M., Kempen, G. (2006). Lexical access during the production of idiomatic phrases. Journal of Memory and Language, 54, 161-184. SR Research Ltd. (2006). Eyelink Programmer’s Guide, Version 3.0. Osgoode, Ontario, Canada: Author. Steinel, M.P., Hulstijn, J. H., & Steinel, W. (2007). Second language idiom learning in a pairedassociate paradigm: Effects of direction of learning, direction of testing, idiom imageability, and idiom transparency. Studies in Second Language Acquisition, 29, 449484. Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journal of Verbal Learning & Verbal Behavior, 18, 523-534. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 53 Tabossi, P., Fanari, R., & Wolf, K. (2008). Processing idiomatic expressions: Effects of semantic compositionality. Journal of Experimental Psychology: Learning, Memory, & Cognition, 34, 313-327. Tabossi, P., Fanari, R., & Wolf, K. (2009). Why are idioms recognized fast? Memory & Cognition, 37, 529-540. Tabossi, P., Wolf, K., & Koterle, S. (2009). Idiom syntax: Idiosyncratic or principled? Journal of Memory & Language, 61, 77-96. Titone, D.A., & Connine, C. M. (1994a). Descriptive Norms for 171 Idiomatic Expressions Familiarity, Compositionality, Predictability, and Literality. Metaphor and Symbolic Activity, 9, 247-270. Titone, D. A., & Connine, C. M. (1994b). Comprehension of idiomatic expressions: Effects of predictability and literality. Journal of Experimental Psychology: Learning, Memory, & Cognition, 20, 1126-1138. Titone, D. A., & Connine, C. M. (1999). On the compositional and noncompositional nature of idiomatic expressions. Journal of Pragmatics, 31, 1655-1674. Titone, D., Libben, M., Mercier, J., Whitford, V., & Pivneva, I. (in press). Bilingual Lexical Access during L1 Sentence Reading: The Effects of L2 Knowledge, Semantic Constraint and L1-L2 Intermixing. Journal of Experimental Psychology: Learning, Memory & Cognition. van Assche, E., Drieghe, D., Duyck, W., Welvaert, M., Hartsuiker, R. J. (2011). The influence of semantic constraints on bilingual word recognition during reading. Journal of Memory and Language, 64, 88-107. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 54 Vespignani, F., Canal, P., Molinaro, N., Fonda, S., & Cacciari, C.. (2010). Predictive mechanisms in idiom comprehension. Journal of Cognitive Neuroscience, 22, 1682-1700. Weinert, R. (1995). The role of formulaic language in 2nd-language acquisition: A review. Applied Linguistics, 16, 180-205. Wood, D. (2006). Uses and functions of formulaic sequences in second language speech: An exploration of the foundations of fluency. Canadian Modern Language Review, 63, 1333. Wray, A. (2002). Formulaic language and the lexicon. Cambridge, UK: Cambridge University Press. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 55 Appendix Sentences used in Experiments 1 and 2 Idiom lost his seat knew the score packed his bags hit the books got the message flooded the market wore the trousers smelled a rat Familiarity Rating 2.7 3.8 4.43 4.73 4.57 2.5 3 3.6 Decomposability Proportion L1/L2 Overlap 0.74 0.514 0.76 0.85 0.76 0.95 0.63 0.32 0.3 0.189 0.703 0.027 0.676 0.054 0.486 0.216 Condition Sentence Id-Id Otto lost his seat on the council after he was caught embezzling funds. Id-Lit Otto lost his seat at the football game when he got up to get some beer. Lit-Lit Otto left his seat unoccupied when he left quickly to use the bathroom. Id-Id Vicki knew the score well enough to not ask for an extension on the essay. Id-Lit Vicki knew the score because she had been following the playoffs closely. Lit-Lit Vicki told the score to the young fans who had just arrived at the game. Id-Id Bernie packed his bags when he became disillusioned with the company's ethics. Id-Lit Bernie packed his bags when he realized his plane was leaving in half an hour. Lit-Lit Bernie emptied his bags as soon as he returned home from the long vacation. Id-Id Penelope hit the books only two weeks before the dreaded final examination. Id-Lit Penelope hit the books with her hand when she sat down quickly at the desk. Lit-Lit Penelope put the books on the shelf when she returned home from the library. Id-Id Rona got the message when her friends refused to talk to her after the breakup. Id-Lit Rona got the message on the answering machine when she arrived home from school. Lit-Lit Rona put the message on the fridge so she wouldn't forget to call her mother. Id-Id Rupert flooded the market with his over-abundant soybean crop last year. Id-Lit Rupert flooded the market while fixing a leak in the store's plumbing. Lit-Lit Rupert entered the market determined to buy only the items on his list. Id-Id Cindy wore the trousers in her family in the weeks after her father died. Id-Lit Cindy wore the trousers after deciding it was too cold to wear the skirt. Lit-Lit Cindy gave the trousers to her son after she was finished hemming them. Id-Id Myrna smelled a rat when her coworker wouldn't look her in the eye yesterday. Id-Lit Myrna smelled a rat the moment she entered the cluttered and dirty pet store. Lit-Lit Myrna trapped a rat in the garage after she saw it eating the bag of birdseed. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING used his head cleared the air smacked her lips had a lark passed the buck covered her tracks scratched his head showed her cards hit a wall got the picture 4.57 3.23 2.83 1.57 2.83 4.3 3.87 2.93 3.87 4.67 0.9 0.71 0.25 0.1 0.36 0.98 0.67 0.77 0.75 0.74 0.784 0.189 0.162 0.000 0.081 0.459 0.459 0.649 0.667 0.216 56 Id-Id Ben used his head to outsmart his opponents even though they were stronger. Id-Lit Ben used his head to bump the ball straight into the opposing team's net. Lit-Lit Ben hurt his head when he sat up too quickly and struck the top bunk bed. Id-Id Royce cleared the air by completely explaining why he was late coming home. Id-Lit Royce cleared the air by opening the window and turning the exhaust fan on. Lit-Lit Royce forced the air into the flat tire before it could deflate itself again. Id-Id Alice smacked her lips at the thought of the pay raise she had been promised. Id-Lit Alice smacked her lips to feel the texture of the lipstick she had purchased. Lit-Lit Alice traced her lips in the mirror as she tried on a new shade of lipstick. Id-Id Ruby had a lark when she switched her family's sugar to salt as a joke. Id-Lit Ruby had a lark when she was a child but now wanted a parrot instead. Lit-Lit Ruby saw a lark at the pet store and thought that it was very beautiful. Id-Id Taylor passed the buck when she was accused of losing the major account. Id-Lit Taylor passed the buck when her friend didn't have enough cash for lunch. Lit-Lit Taylor saved a buck when she brought in her coupon at the grocery store. Id-Id Stacey covered her tracks so that she wouldn't be caught embezzling money. Id-Lit Stacey covered her tracks so that she couldn't be followed through the snow. Lit-Lit Stacey cleared her tracks from the garden once she was finished weeding it. Id-Id Niles scratched his head when he could not determine where his dog was hiding. Id-Lit Niles scratched his head when he returned from camp covered in mosquito bites. Lit-Lit Niles examined his head when he returned from camp covered in mosquito bites. Id-Id Norah showed her cards when she invited the handsome man to the wedding party. Id-Lit Norah showed her cards to the other poker players to show that she had bluffed. Lit-Lit Norah counted her cards to determine who had more points when the game ended. Id-Id Marge hit a wall in her renovations when she realized she was out of materials. Id-Lit Marge hit a wall while digging her garden too close to the house's foundation. Lit-Lit Marge fit a wall in between the gazebo and the path around the garden's pond. Id-Id Eric got the picture when his blind date didn't show up after 40 minutes. Id-Lit Eric got the picture when his mother finally mailed it to him overnight. Lit-Lit Eric put the picture on the mantelpiece as a reminder of his fun vacation. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING bought the farm changed her tune made a bundle lined his pockets dropped a bundle bit the bullet followed his nose trimmed his sails took the veil dropped a line 2.53 3.8 3.87 2.93 2.47 4.2 3.23 2.03 1.67 4.03 0.05 0.8 0.7 0.16 0.5 0.28 0.5 0.12 0.52 0.31 0.000 0.216 0.054 0.108 0.000 0.135 0.324 0.081 0.162 0.270 57 Id-Id Harry bought the farm unexpectedly after developing lung cancer late in life. Id-Lit Harry bought the farm from his uncle after inheriting a large sum of money. Lit-Lit Harry loved the farm and always enjoyed visiting his uncle there on holidays. Id-Id Maggie changed her tune and supported the policy when she recognized its benefits. Id-Lit Maggie changed her tune to a livelier melody when the audience was unresponsive. Lit-Lit Maggie hummed her tune happily as she returned from a successful choir practice. Id-Id Hank made a bundle over the summer working full time at the community pool. Id-Lit Hank made a bundle out of his towels before stuffing them in his suitcase. Lit-Lit Hank gave a bundle of clothes to the Salvation Army at least once a year. Id-Id Keith lined his pockets with the money he got for selling the stolen stereos. Id-Lit Keith lined his pockets with fleece after he got frostbite on his fingertips. Lit-Lit Keith sewed his pockets when he started losing change through several holes. Id-Id Jane dropped a bundle when she stayed in a five-star hotel for her vacation. Id-Lit Jane dropped a bundle when she tried to carry too much laundry up the stairs. Lit-Lit Jane carried a bundle of clothes down the street to the local Laundromat. Id-Id Larry bit the bullet and bought diamond earrings for his wife's birthday. Id-Lit Larry bit the bullet to verify the quality of the casings on his ammunition. Lit-Lit Larry hid the bullet so the police would not find the crucial evidence. Id-Id Billy followed his nose when hiring rather than checking applicant references. Id-Lit Billy followed his nose to the kitchen where his mother was baking apple pies. Lit-Lit Billy splashed his nose with cold water in an effort to relieve the swelling. Id-Id Blake trimmed his sails after he started paying for his own tuition fees. Id-Lit Blake trimmed his sails in preparation for the local yacht competition. Lit-Lit Blake raised the sails in preparation for the local yacht competition. Id-Id Rosemary took the veil because she had always wanted to devote her life to God. Id-Lit Rosemary took the veil because she didn't want to forget it before the wedding. Lit-Lit Rosemary tore the veil accidentally because the gauzy material was too fragile. Id-Id Hal dropped a line to his old roommate after remembering it was his birthday. Id-Lit Hal dropped a line to see if the fish were biting this early in the season. Lit-Lit Hal printed a line in colored ink to find out if the ink cartridge was full. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING kicked the bucket held the fort missed the mark pulled David's leg buried the hatchet faced the music hedged her bets blew a fuse lost the thread skimmed the cream 2.97 3.23 3.37 4.23 3.7 3.23 2.13 4.03 1.67 1.86 0.17 0.41 0.85 0.15 0.38 0.24 0.55 0.31 0.44 0.3 0.108 0.405 0.250 0.162 0.378 0.243 0.000 0.730 0.514 0.108 58 Id-Id Mary kicked the bucket after suddenly becoming seriously ill on the weekend. Id-Lit Mary kicked the bucket when it was blocked from her view by the chair. Lit-Lit Mary tipped the bucket to sprinkle fertilizer into her new flower garden. Id-Id Haden held the fort by taking care of his brother while their mom was at work. Id-Lit Haden held the fort high above the table while his Lego armies battled below. Lit-Lit Haden left the fort because he no longer wanted to work there in the summer. Id-Id Steve missed the mark when he claimed the Yankees won a World Series in 1960. Id-Lit Steve missed the mark when his shoulder slipped from the weight of the rifle. Lit-Lit Steve placed the mark at the place where the runners began the city marathon. Id-Id Bailey pulled David's leg because he was great fun to joke around with. Id-Lit Bailey pulled David's leg because he was stuck halfway down the slide. Lit-Lit Bailey rubbed David's leg to increase blood flow to his frozen feet. Id-Id Yvonne and Lucy buried the hatchet after both forgot why they were fighting. Id-Lit Yvonne and Lucy buried the hatchet after stealing it from the hardware store. Lit-Lit Yvonne and Lucy threw the hatchet at the approaching bear hoping to scare it. Id-Id Regan faced the music when her parents caught her returning home after curfew. Id-Lit Regan faced the music when she wanted to determine which speaker was loudest. Lit-Lit Regan hated the music and stayed out of the house when his brother practiced. Id-Id Carol hedged her bets by double majoring in computer science and accounting. Id-Lit Carol hedged her bets at the horse races by wagering on all the favorites. Lit-Lit Carol changed her bets when she learned that one of the horses hurt its leg. Id-Id Sarah blew a fuse when the material she had studied most wasn't on the test. Id-Lit Sarah blew a fuse when her bedside lamp had a severe electrical malfunction. Lit-Lit Sarah fixed a fuse that always failed when she used the plug beside the sink. Id-Id Charlie lost the thread of the story and couldn't grasp its conclusion. Id-Lit Charlie lost the thread when he dropped the sewing needle on the floor. Lit-Lit Charlie made the thread out of cotton fibers he got at the fabric store. Id-Id Heather skimmed the cream by only hiring the top graduating students for the job. Id-Lit Heather skimmed the cream by dragging a big spoon through the fresh jug of milk. Lit-Lit Heather spilled the cream when she reached across the table for the sugar bowl. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING cleared the decks broke the ice got the sack spilled the beans straddled the fence dropped a bomb spread the word turned the tables showed his teeth went to town 2.9 4.43 2.77 4.37 2.57 4.1 4.8 4.03 2.7 3.77 0.16 0.26 0.15 0.45 0.27 0.5 0.95 0.44 0.32 0.15 0.054 0.919 0.000 0.162 0.054 0.216 0.486 0.054 0.459 0.162 59 Id-Id Jeff cleared the decks by finishing his project before undertaking a new one. Id-Lit Jeff cleared the decks by sweeping away the leaves before his guests arrived. Lit-Lit Jeff cleaned the decks so they would be spotless for his party at the marina. Id-Id Bruce broke the ice by quickly introducing himself to everyone at the wedding. Id-Lit Bruce broke the ice by driving his snowmobile directly onto the thawing lake. Lit-Lit Bruce stored the ice in his cooler so he could bring it to the holiday party. Id-Id Derek got the sack after showing up late for work and having a bad attitude. Id-Lit Derek got the sack that had been hidden behind boxes in the kitchen pantry. Lit-Lit Derek put the sack in the pantry when he was finished peeling the potatoes. Id-Id Dolan spilled the beans when he mentioned the surprise party to his friend. Id-Lit Dolan spilled the beans when he tried to pour too many into the soup pot. Lit-Lit Dolan cooked the beans before he started adding vegetables to the soup pot. Id-Id Dolores straddled the fence as she couldn't decide which position to believe. Id-Lit Dolores straddled the fence as she climbed over it into her neighbor's yard. Lit-Lit Dolores destroyed the fence before building a new one with fresh cedar rails. Id-Id Winston dropped a bomb on his parents when he announced his engagement. Id-Lit Winston dropped a bomb on the enemy's base as his plane flew overhead. Lit-Lit Winston planted a bomb in the enemy's headquarters early that morning. Id-Id Bailey spread the word about the party to all her friends by sending an email. Id-Lit Bailey spread the word across the top of her poster to use up all the space. Lit-Lit Bailey feared the word that her strict father was about to say to her was no. Id-Id Jade turned the tables on his opponent by using the new tennis technique. Id-Lit Jade turned the tables upside down so that they could be easily dismantled. Lit-Lit Jade washed the tables after the customers had left the Thai restaurant. Id-Id Pete showed his teeth when he caught the local boys bullying his sister. Id-Lit Pete showed his teeth when his mother asked if he had brushed that morning. Lit-Lit Pete cleaned his teeth when his mother reminded him for the third time. Id-Id Alice went to town on the preparations for her son's elaborate wedding. Id-Lit Alice went to town to buy decorations for her son's elaborate wedding. Lit-Lit Alice came to town to visit her relatives over the Christmas holidays. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING missed the bus bore his cross lost his shirt hit the ceiling father forced her licked her wounds paved the way gave her word hit the roof bit her lip 3.97 2.5 3.47 2.97 2.77 2.86 4.2 4.6 3.97 4.03 0.47 0.5 0.56 0.18 0.7 0.45 0.85 0.91 0.16 0.74 0.459 0.162 0.270 0.216 0.514 0.162 0.216 0.703 0.162 0.541 60 Id-Id Yvette missed the bus when she didn't hand in her application on time. Id-Lit Yvette missed the bus when she forgot to set her alarm Monday morning. Lit-Lit Yvette drove the bus from the high-school to drop off the students. Id-Id Josh bore his cross the entire flight and didn't complain about the snoring man. Id-Lit Josh bore his cross down the center aisle of the church during the passion play. Lit-Lit Josh lost his cross when he dropped it in the grass on the way home from church. Id-Id Brian lost his shirt when his new hobby shop went bankrupt last winter. Id-Lit Brian lost his shirt when it blew off the clothesline during the storm. Lit-Lit Brian left his shirt at the Laundromat when he had to leave in a hurry. Id-Id Sally hit the ceiling when the course she'd prepared for was cancelled. Id-Lit Sally hit the ceiling when she tried to jump on her trampoline indoors. Lit-Lit Sally lit the ceiling on fire when the Christmas tree lights overheated. Id-Id Kaitlyn's father forced her hand by telling her where to attend high school. Id-Lit Kaitlyn's father forced her hand away from the stove before she was burned. Lit-Lit Kaitlyn's father pushed her hand away from the stove before she was burned. Id-Id Rose licked her wounds after being defeated badly in the tennis championship. Id-Lit Rose licked her wounds since there was no clean water available to wash them. Lit-Lit Rose washed her wounds with peroxide so that they would not become infected. Id-Id Jerry paved the way for his son by setting up a trust fund for the university. Id-Lit Jerry paved the way for his son so he wouldn't trip over the gravel again. Lit-Lit Jerry found the way to the top of the mountain without the help of a guide. Id-Id Vicki gave her word to Jack that she would help him paint his room the next day. Id-Lit Vicki gave her word to Jack when he started finding the scrabble game difficult. Lit-Lit Vicki felt her word was too strong for the context and regretted saying it. Id-Id Sam hit the roof when he discovered someone had taken his new golf clubs. Id-Lit Sam hit the roof to check the condition of the old and crumbling shingles. Lit-Lit Sam cut the roof into pieces with his saw after he dismantled the old barn. Id-Id Roxy bit her lip and tried to keep the plans for the surprise party a secret. Id-Lit Roxy bit her lip as she rushed through breakfast in a hurry to get to school. Lit-Lit Roxy cut her lip on a branch when she climbed too high up the cedar tree. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING got the axe pulled the plug 3.37 3.4 0.19 0.69 0.027 0.216 61 Id-Id Ramona got the axe after being repeatedly warned not to be late for work. Id-Lit Ramona got the axe after being asked to help chop kindling for the fire. Lit-Lit Ramona had the axe that she borrowed from her neighbor to chop fire wood. Id-Id Dara pulled the plug on her brother's money when she caught him gambling. Id-Lit Dara pulled the plug of the bathtub when the water began to get very cold. Lit-Lit Dara removed the plug from the bathtub when the water began to get cold. Running Head: FIRST AND SECOND LANGUAGE IDIOM PROCESSING 62
© Copyright 2026 Paperzz