Does Infant?Directed Speech Help Phonetic Learning? A Machine Learning Investigation

A prominent hypothesis holds that by speaking to infants in infant-directed speech (IDS) as opposed adult-directed (ADS), parents help them learn phonetic categories. Specifically, two characteristics of IDS have been claimed facilitate learning: hyperarticulation, which makes the categories more separable, and variability, generalization robust. Here, we test separability robustness vowel category learning on acoustic representations uttered Japanese adults ADS, (addressed 18- 24-month olds), or read (RS). Separability is determined means a distance measure computed between five short Japanese, while assessed testing ability six different machine algorithms trained classify vowels generalize stimuli spoken novel speaker ADS. Using representations, find hyperarticulated speech, case RS, can yield better separability, increased between-speaker variability ADS yield, for some algorithms, robust However, these conclusions do not apply IDS, turned out neither separable nor compared inputs. We discuss usefulness run real data hypotheses about functional role IDS.