Mary Beckman pointed out that vowels [or a release] are sometimes necessary to establish the place of articulation of consonants, so phonetics and phonotactics *should* be analyzed together. Jessica points out that context also introduces variation, and that sometimes the dependency is longer distance than just the neighboring segments [please confirm! I doubt my notes here!]
Jerry Jeager (?) asks Ying and Jeff why they used a hierarchical model - she has found evidence from slips of the tongue by 18-month-olds to 5-year-olds and adults for combination of features but not hierarchical structure. Ying and Jeff answer that this is just a manner of analyzing the data, although it is possible that the hierarchy works prior to motor experience. Jessica points to a study by Jusczyk [Goodman and Baumann, 1998] where babies react to manner of articulation similarities, but not place of articulation similarities, between syllables in a list. It's possible that later one only retains the 'bottom' categories, and not the hierarchy itself.
* would the contribution of semantic cues be more helpful in the case of morphophonological alternations? Lisa D. answers that meaning helped her participants slightly, not a lot.
?? Susan - merging and resplitting <- phonotactic differences were harder
* Using TIMIT for learning features is cheating! How about the phonemic segmentation problem? Ying refers to the other part of his dissertation, where he starts from the raw acoustics.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment