Poster E58, Saturday, August 18, 3:00 – 4:45 pm, Room 2000AB
Model of weighted phonological similarity for predicting bilingual lexical retrieval
Danielle Fahey1;1University of South Carolina
Introduction: Listeners decompose words into phonological representations, linking phonemic representations to concepts (Levelt et al., 1999). Bilingual speakers are thought to have one concept connected to two separate phonological representations for translational equivalents (Kroll & Stewart, 1994). To predict how quickly bilingual speakers will retrieve words in their first (L1) or second language (L2), neighborhood effects are often used to quantitatively evaluate the differences between words in the two languages of interest, comparing addition, deletions and subtractions from orthographic representations (Peeters et al., 2013). However, bilingual speakers have been shown to expand phonemes across languages, such that overlapping phonemes are perceived as a single phoneme in both languages (Flege, 1995). Because bilingual speakers assimilate phonological features across languages, studies predicting aural word retrieval should model weighted featural similarity based upon perceived featural differences. This study asked participants to rate Spanish-English cognate differences, and modeled featural perception, weighting different phonological characteristics. The resulting model can be used to predict perceived phonological similarity in studies of bilingual lexical retrieval. Methods: Thirty-five Spanish-English bilingual participants heard 150 Spanish-English cognates, and rated similarity on a 7-point Likert scale in Qualtrics. Phonology change from English to Spanish was analyzed as: stress change (StrC), number of syllables (SNC), and syllable stress location (SCL). Consonant and vowel changes were analyzed as follows, and summed to weight each type across a cognate set: consonant additions/deletions (CAD); consonant substitutions with a voicing or manner change (CSubSame); consonant substitutions with a place change (CSubDiff); consonant changes to onset (COns); consonant changes to coda (CCoda); vowel additions/deletions (VAD); vowel substitutions within similar space (VSubSame); vowel substitutions to different space (VSubDiff); changes to stressed vowels (VStress). Results: Repeated measures linear regression of mixed effects showed the following factors significantly affected similarity rating: SCL (p<0.001), CAD (p<0.001), CSubDiff (p<0.001), CCoda (p=0.001), VAD (p<0.001), VSubSame (p<0.001), VSubDiff (p<0.001), and VStress (p<0.001). All parameter weights were negative, except for the effect of VSubSame. Similarity rating = 5.4484 - .4671xSCL - .5505xCAD - .7637xCSubDiff - .1456xCCoda - .3224xVAD + .2977xVSubSame - .2430xVSubDiff - .5133xVStress Conclusions: Results demonstrate that greater phonological distance, as measured by changes to cognates’ stress placement, syllabification, consonants or vowels, causes lower ratings of cognate similarity. However, close phonetic substitutions accounted for no difference, or even perceived similarity. These results are important because they extend the results of Flege (1995) and others to categorical perception of L1 lexical items instead of L2 items only. Further, categorical perception of phonological features applies across cognates, suggesting that merely quantifying phonological differences apparent to monolinguals is not enough to provide a strong prediction of bilingual lexical retrieval. Thus, subphonemic similarity should be used when predicting cognate similarity, and the above model may be used to rate similarity across Spanish-English words.
Topic Area: Multilingualism