24.964 (MIT) Topics in Phonology / Ling 219r (Harvard) Advanced Phonology Topics in phonotactic learning
Instructors: Adam Albright and Michael Becker
Location: alternate weeks @ Harvard (Boylston 306) and MIT (32-D461)
Time: Thursdays 3-6 (see scheduling info below)
First meeting: Thurs 2/4 at Harvard
Description:
In this seminar, we will explore approaches to several related issues in the analysis and learning of phonotactic patterns. The format will be collaborative: for each issue, we will focus in detail on a small set of languages that exemplify the problem. Registered participants will make regular presentations on proposals from the literature, and we will work together to evaluate their potential applicability to the problem at hand. In each case, the goal will be to develop a possible analysis to the point of implementing a “proof of concept” model of how the pattern could be learned. The emphasis will be on both grammatical mechanisms (primarily within OT) and learning models (constraint ranking, Bayesian approaches, etc.)
Background:
We assume basic familiarity with Optimality Theory
Scheduling info:
We will meet alternate weeks at MIT and Harvard. Please note the following:
– The first meeting is 2/4 at Harvard (Boylston 306)
– We will not meet 3/18 or 3/25
– The last meeting is 5/13
Topics (subject to revision)
- Morpheme Structure Conditions
Languages frequently exhibit phonotactic restrictions that hold of roots or underived words, but are not enforced in derived forms. For example, English has no monomorphemic words ending in fricative-fricative clusters, but they may be created by affixation (myth-s, fif-th, cave-z). Such restrictions have often been taken to require a distinct level of evaluation, e.g., through conditions on lexical entries (Halle 1959), but this approach has generally been avoided in OT. We consider the arguments for various alternatives.
Representative cases:
English final clusters
Semitic OCP restrictionsSome possible tools:
Output-output faithfulness; stratal OT; domain-sensitive markedness; accidental gaps
- Exceptions to Morpheme Structure Conditions
Another challenge posed by MSCs that are not enforced in derived forms is that they are frequently tendencies rather than absolute restrictions. We review proposals for protecting individual words from conforming to broader generalizations, which include representational and grammatical mechanisms. We ask how exceptions are defined in a theory where the generalizations are gradient, or violable, or imposed on differing domains.
Representative cases:
Semitic OCP restrictions
Romance diphthongizationSome possible tools:
Constraint indexation; faithfulness to listed forms; dual route models
- Conditions on subclasses: lexical strata, parts of speech, etc.
MSCs frequently hold over a subset of the lexicon. We consider evidence that lexical strata are more than simply sets of exceptions, considering speakers’ ability to integrate potentially independent facts into broader observations. For example, in Hebrew, complex codas are only found in items with fixed stress, and in Turkish, long vowels are only found in nouns. Independent facts can also be combined gradiently: In Turkish, a long vowel reduces, but does not eliminate, the chance of finding voicing alternations in a stem-final stop.
Representative cases:
Turkish (long vowels, voicing alternations)
Japanese (syllable structure, inventory, voicing alternations)
Spanish (intervocalic voiceless stops, clusters, diphthongization)Some possible tools:
Constraint indexation; co-phonologies; hierarchical Bayesian models