Abstract
Recent research has shown that seemingly identical suffixes such as word-final /s/ in English show systematic differences in their phonetic realisations. Most recently, durational differences between different types of /s/ have been found to also hold for pseudowords: the duration of /s/ is longest in non-morphemic contexts, shorter with suffixes, and shortest in clitics. At the theoretical level such systematic differences are unexpected and unaccounted for in current theories of speech production. Following a recent approach, we implemented a linear discriminative learning network trained on real word data in order to predict the duration of word-final non-morphemic and plural /s/ in pseudowords using production data by a previous production study. It is demonstrated that the duration of word-final /s/ in pseudowords can be predicted by LDL networks trained on real word data. That is, duration of word-final /s/ in pseudowords can be predicted based on their relations to the lexicon.
Highlights
Many studies on the acoustic properties of phonologically homophonous elements have shown unexpected effects of their morphological structure on their phonetic realisation
All of the following analyses make use of the following non-lexical covariates: BASEDURLOG, SPEAKINGRATE, SLIDENUMBER, and PAUSEBIN as variables concerning speech rate and continuity, preceding consonant (PREC) and following segmental type (FOLTYPE) accounting for coarticulatory effects, LIST taking into consideration potential priming effects, MONOMULTILINGUAL, GENDER, LOCATION, AGE, and SPEAKER to account for speaker-individual differences, and REAL to include potential effects of real word counterparts
In the final model including linear discriminative learning (LDL) measures as well as the AFFIX covariate as parts of the individual components resulting from the principal component analysis, and fitted according to the procedure described in Section “Model B: LDL Measures and Affix Specification,” we find main effects of the first principal component (COMPONENT1), the third principal component (COMPONENT3), DENSITY, ALC, base duration (BASEDURLOG), following pause (PAUSEBIN), following segmental type (FOLTYPE), and preceding consonant (PREC)
Summary
Linear discriminative learning as a computational model implements a discriminative view of learning. All of the following analyses make use of the following non-lexical covariates: BASEDURLOG, SPEAKINGRATE, SLIDENUMBER, and PAUSEBIN as variables concerning speech rate and continuity, PREC and FOLTYPE accounting for coarticulatory effects, LIST taking into consideration potential priming effects, MONOMULTILINGUAL, GENDER, LOCATION, AGE, and SPEAKER to account for speaker-individual differences, and REAL to include potential effects of real word counterparts. As an alternative we implement a model that uses LDL measures If these measures are predictive, they offer an explanation of the morphologically-induced phonetic differences: they emerge as a by-product of the association of form and meaning in the mental lexicon, and this association is the outcome of discriminative learning. As for random effects, random intercepts for GENDER, LOCATION, MONOMULTILINGUAL, AGE, LIST, and SPEAKER were included This full model was continuously reduced through step-wise exclusion of non-significant variables, following the aforementioned criteria. An overview of all variables used in the initial model and their distribution is given in Supplementary Table 2
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
