Novel phonotactic learning: Tracking syllable-position and co-occurrence constraints

Abstract

Are syllable-level and co-occurrence representations simultaneously available when one learns novel phonotactics? After training on word-medial consonant restrictions (e.g., word-medial onsets P/Z, codas D/F, and cross-syllable consonant clusters FP/DZ in items like baF.Pev, tiD.Zek), adults falsely recognized novel items containing restricted consonants with the same co-occurrences (e.g., FP) more often than those with different co-occurrences (e.g., FZ) when syllable-position information was kept constant (e.g., vuF.Pet vs. vuF.Zet; Experiment 1). Thus, adults tracked co-occurrence information. Additionally, even when co-occurrence information was different from training, participants more often falsely recognized novel items that contained restricted consonants in the same (e.g., onset-Z) rather than different syllable positions (e.g., coda-Z), whether the restricted consonants were in the same (word-medial, e.g., vuF.Zet vs. vuZ.Fet, Experiment 2) or different word positions (word-edge, e.g., Zut.veF vs. Fut.veZ, Experiment 3). Thus, adults also tracked syllable-level information. These findings show that adults spontaneously represent sound sequences at multiple levels.

The adaptable speaker: A theory of implicit learning in language production.

“…Experimental studies of phonotactic learning often use artificial grammar paradigms, in which language users are exposed to linguistic forms, such as single syllables, that exhibit particular constraints. Then, rule-following and rule-violating forms are presented for judgments of grammaticality (e.g., Finley & Badecker, 2009), familiarity (e.g., Bernard, 2017), or for a speeded response of some sort (e.g., Onishi et al, 2002). If the participants are infants, learning can be assessed by differences in attention to the rule-following and rule-violating forms (e.g., Chambers et al, 2003; Seidl et al, 2009).…”

Using LSTMs to Assess the Obligatoriness of Phonological Distinctive Features for Phonotactic Learning

“…Following this, our model will represent gradient constraints, and its task will be to assign gradient acceptability ratings to sequences of phonetic segments. Bernard (2017) demonstrated that humans are capable of simultaneously tracking and learning phonotactic generalizations defined at the level of word boundaries, syllable positions, and cooccurrences between adjacent phonetic segments. Our LSTM networks are capable of capturing all three types of constraints.…”

Perception of the English Phonotactics by Saudi English Majors: A Comparative Study

“…Each language has its own constraints on permissible sequences which may interact with other languages (Smolensky & Prince, 1993). Bernard (2017) says that human brain is able to track and learn phonotactics and can generalize at different levels that include word boundaries and syllable positions, stating that "phonotactic knowledge leads to enhanced speech processing enables us to use phonotactic learning and generalization as a means to better understand how humans represent speech sounds and sound sequences" (p. 138).…”