Lexical Strata and Phonotactic Perplexity Minimization

We present a model of gradient phonotactics that is shown to reduce overall phoneme uncertainty in a language when the phonotactic grammar is modularized in an unsupervised fashion to create more than one sub-grammar. Our model is a recurrent neural network language model (Elman 1990), which, when applied in two separate, randomly initialized modules to a corpus of Japanese words, learns lexical subdivisions that closely correlate with two of the main lexical strata for Japanese (Yamato and Sino-Japanese) proposed by Ito and Mester (1995). We find that the gradient phonotactics learned by the model, which are based on the entire prior context of a phoneme, reveal a continuum of gradient strata membership, similar to the gradient membership proposed by Hayes (2016) for the Native vs. Latinate stratification in English.

Gradient Acceptability in Mandarin Nonword Judgment

Syllable well-formedness judgment experiments reveal that speakers exhibit gradient judgment on novel words, and the gradience has been attributed to both grammatical factors and lexical statistics (e.g., Coetzee, 2008). This study investigates gradient phonotactics stemming from the violations of four types of grammatical constraints in Mandarin Chinese: 1) principled phonotactic constraints, 2) accidental phonotactic constraints, 3) allophonic restrictions, and 4) segmental-tonal cooccurrence restrictions. A syllable well-formedness judgment experiment was conducted with native Mandarin speakers to examine how the grammatical and lexical statistics factors contribute to the variation in phonotactic acceptability judgment.

The metrical parse is guided by gradient phonotactics

Phonotactic generalisations can be computed at different levels of granularity, from a coarse-grained legal/illegal dichotomy (blick,dwick≻ *bnick, *lbick) to a fine-grained gradient of acceptability (blick≻dwick≻bnick≻lbick). This article investigates the sensitivity of the English metrical parse to the granularity of medial onset phonotactics. We present two experiments that feature pseudo-words with medial consonants and CC clusters varying in word-edge frequency and sonority (e.g.vatablick,vatadwick,vatabnick,vatalbick). The metrical parse is inferred from a hyphenation experiment and an online stress-assignment experiment. The results of both studies indicate that the parse is stochastic, and guided by relatively fine-grained phonotactic dependencies. Vocabulary simulations suggest that this level of granularity may arise because the gradient parser consistently outperforms the coarse-grained alternative across the developing lexicon.