How learners remember words in their second language: The impact of individual differences in perception, cognitive abilities, and vocabulary size
Previous research on the lexical encoding of L2 contrasts has shown that perception is likely necessary but not sufficient for accurate lexical representations. Even when learners are able to perceive a contrast, they often struggle with lexical encoding, and studies have shown that perception ability explains only some of the variability in lexical encoding accuracy. My dissertation examines what determines lexical encoding accuracy by examining not only the relationship between the perception of different L2 contrasts and their accuracy in phonolexical representations, but also the relationship between lexical encoding and a range of individual differences, specifically, inhibitory control, attention control, phonological short-term memory, and L2 vocabulary size. Findings indicate that the factors that affect lexical representations depend on which sounds are being encoded. When representations contain sounds that are differentiated along a dimension not used in the native language (i.e., /tap-trill/), learners with higher phonological short-term memory have an advantage, likely because they are better able to hold the relevant phonetic details in memory long enough to be transferred to long-term representations. Differences in perception ability matter most for sounds that are perceptually difficult to distinguish (i.e., /tap-d/). Finally, second language vocabulary size is the strongest factor in predicting lexical encoding across almost all contrasts, such that a larger vocabulary predicts greater accuracy. This is presumably because knowing more words entails the presence of more phonological neighbors, which puts pressure on learners’ phonological system to differentiate these minimally different words with more detailed representations. In addition, a larger vocabulary is indicative of more experience with the language, and exemplars for words that are based on more input are likely better defined. You can download a copy of my dissertation here and the poster from LabPhon17 based on this study here.
Difficulty lexically encoding Spanish rhotics: L1 phonological grammar or variability in the input?
In a project related to my dissertation, I examine whether L2 learners' difficulty lexically encoding the Spanish /tap-trill/ contrast stems from 1) the lack of a rhotic contrast in the L1 or 2) trill variation in the L2. My results so far suggest that the majority of intermediate learners have not accurately encoded a difference between the tap and trill due to their L1 phonological grammar. In contrast, some advanced learners have a clear asymmetry in that they accept tap for trill but not trill for tap, some accept either rhotic but can identify the canonical pronunciation, whereas others are accurate across all tasks. This suggests that many advanced learners have correctly encoded the rhotics in lexical representations, and the acceptance of tap for trill by some members of this group is likely due to trill variability in the input, while others have not encoded a difference. You can view my slides from my presentation at CASPSLAP 2020 here.
Evaluating non-native perception tasks
My Indiana University colleagues Ryan Lidster and Franzi Kruger are conducting a number of studies with me in which we examine the relationship between various non-native perception tasks: free classification, perceptual assimilation, identification, AXB, and oddity (and likely similarity judgment in the future). We've tested the perception of Finnish vowels and Finnish length by Japanese and American English listeners, German vowels by American English listeners, and Arabic consonants by American English listeners. Here are some highlights from our findings so far:
L2 lexical representations: Alternate or fuzzy?
In previous research (Daidone & Darcy, 2014; Darcy, Daidone, & Kojima, 2013) we've found that learners have difficulty rejecting incorrect pronunciations of words with new phonemes, regardless of whether they can accurately perceive these new sounds. For example, English-speaking learners of German often will accept a non-word with a back vowel as a real word when the real word actually contains a front-rounded vowel, e.g. they will accept [konɪç] as a word when the word is actually könig [kønɪç] 'king.' However, from this research it was unclear if learners' lexical representations were fuzzy or alternate. By fuzzy, we assume that a difficult novel phoneme is underspecified such that the learner would accept many different possible realizations of a word, e.g. könig as [kønɪç], [konɪç], [kunɪç], or even [kenɪç]? If a learner has alternate lexical representations, then we hypothesize that they would only have a limited number of possible realizations that they will accept, e.g. könig only as [kønɪç] or [konɪç]. The current project with Isabelle Darcy, Franzi Kruger, and Ryan Lidster tries to determine which of these two possibilities is true for learners' lexical representations of words containing front-rounded vowels in German.
Previous research on the lexical encoding of L2 contrasts has shown that perception is likely necessary but not sufficient for accurate lexical representations. Even when learners are able to perceive a contrast, they often struggle with lexical encoding, and studies have shown that perception ability explains only some of the variability in lexical encoding accuracy. My dissertation examines what determines lexical encoding accuracy by examining not only the relationship between the perception of different L2 contrasts and their accuracy in phonolexical representations, but also the relationship between lexical encoding and a range of individual differences, specifically, inhibitory control, attention control, phonological short-term memory, and L2 vocabulary size. Findings indicate that the factors that affect lexical representations depend on which sounds are being encoded. When representations contain sounds that are differentiated along a dimension not used in the native language (i.e., /tap-trill/), learners with higher phonological short-term memory have an advantage, likely because they are better able to hold the relevant phonetic details in memory long enough to be transferred to long-term representations. Differences in perception ability matter most for sounds that are perceptually difficult to distinguish (i.e., /tap-d/). Finally, second language vocabulary size is the strongest factor in predicting lexical encoding across almost all contrasts, such that a larger vocabulary predicts greater accuracy. This is presumably because knowing more words entails the presence of more phonological neighbors, which puts pressure on learners’ phonological system to differentiate these minimally different words with more detailed representations. In addition, a larger vocabulary is indicative of more experience with the language, and exemplars for words that are based on more input are likely better defined. You can download a copy of my dissertation here and the poster from LabPhon17 based on this study here.
Difficulty lexically encoding Spanish rhotics: L1 phonological grammar or variability in the input?
In a project related to my dissertation, I examine whether L2 learners' difficulty lexically encoding the Spanish /tap-trill/ contrast stems from 1) the lack of a rhotic contrast in the L1 or 2) trill variation in the L2. My results so far suggest that the majority of intermediate learners have not accurately encoded a difference between the tap and trill due to their L1 phonological grammar. In contrast, some advanced learners have a clear asymmetry in that they accept tap for trill but not trill for tap, some accept either rhotic but can identify the canonical pronunciation, whereas others are accurate across all tasks. This suggests that many advanced learners have correctly encoded the rhotics in lexical representations, and the acceptance of tap for trill by some members of this group is likely due to trill variability in the input, while others have not encoded a difference. You can view my slides from my presentation at CASPSLAP 2020 here.
Evaluating non-native perception tasks
My Indiana University colleagues Ryan Lidster and Franzi Kruger are conducting a number of studies with me in which we examine the relationship between various non-native perception tasks: free classification, perceptual assimilation, identification, AXB, and oddity (and likely similarity judgment in the future). We've tested the perception of Finnish vowels and Finnish length by Japanese and American English listeners, German vowels by American English listeners, and Arabic consonants by American English listeners. Here are some highlights from our findings so far:
- Free classification, which had previously been used to examine the perceptual similarity of dialects, languages, or non-native accents, is also useful for looking at the similarity of segments. Some nice properties of free classification are that there are no category labels or number of categories imposed by the researcher, and it's possible to analyze which dimensions (e.g. F1, roundedness) listeners are using to categorize the stimuli. Check out our ICPhS 2015 paper on this method.
- Training in linguistics (particularly phonetics/phonology) and knowledge of other languages can affect performance in a free classification task, and even more so in a perceptual assimilation task. Get our slides from New Sounds 2016 here. Email me if you'd like a copy of the manuscript we're revising.
- In a study together with Meg Cychosz at UC Berkeley, we found that grouping rates from free classification (how often stimuli were grouped together) and overlap scores from perceptual assimilation (how often stimuli were assimilated to the same set of L1 categories) were both good predictors of German vowel discrimination in oddity and AXB tasks. However, perceptual assimilation category types (e.g. "two-category," "category-goodness") were not. In addition, results from AXB and oddity tasks were highly correlated with each other (r = .974 for accuracy scores and r = .979 for d'). Get our slides from PSLLT 2017 here.
- In a study on Finnish length perception we carried out in collaboration with Lila Michaels (Indiana University alum) and Aaron Albin (Kobe University), we found that 1) overlap scores from an identification task and 2) grouping rates and MDS distances derived from a free classification task were good predictors of the discriminability of different length contrasts for both Japanese listeners, who have phonemic length in their L1, and American English listeners, who do not. See our slides from New Sounds 2019 here.
- In this series of experiments, overall we've found that free classification grouping rates (percent of the time stimuli were grouped together by participants) and overlap scores from perceptual assimilation and identification tasks (how often stimuli were assimilated to the same set of categories) are always better predictors of the accuracy order of contrasts in discrimination than perceptual assimilation category types (e.g. "two-category assimilation"). This suggests that if researchers want to predict which L2 sounds will be hard for beginners to discriminate, doing a task/analysis that examines the perceptual similarity of L2 sounds is more useful than determining how L2 sounds relate to L1 categories. Get our slides from New Sounds 2019 here.
L2 lexical representations: Alternate or fuzzy?
In previous research (Daidone & Darcy, 2014; Darcy, Daidone, & Kojima, 2013) we've found that learners have difficulty rejecting incorrect pronunciations of words with new phonemes, regardless of whether they can accurately perceive these new sounds. For example, English-speaking learners of German often will accept a non-word with a back vowel as a real word when the real word actually contains a front-rounded vowel, e.g. they will accept [konɪç] as a word when the word is actually könig [kønɪç] 'king.' However, from this research it was unclear if learners' lexical representations were fuzzy or alternate. By fuzzy, we assume that a difficult novel phoneme is underspecified such that the learner would accept many different possible realizations of a word, e.g. könig as [kønɪç], [konɪç], [kunɪç], or even [kenɪç]? If a learner has alternate lexical representations, then we hypothesize that they would only have a limited number of possible realizations that they will accept, e.g. könig only as [kønɪç] or [konɪç]. The current project with Isabelle Darcy, Franzi Kruger, and Ryan Lidster tries to determine which of these two possibilities is true for learners' lexical representations of words containing front-rounded vowels in German.