Jan 24
======

Turnbull/etal:2018
------------------

I'm confused about some of the data exclusions in the study. In
category 1, I'm not sure why words with syllabic n were excluded, as
this could easily still assimilate and become syllabic m or
ŋ. Similarly, in category 3, the "'cause" tokens could also have been
used with the initial consonant coded as [k], which was also one of
the initial stops of interest.

Could the findings of the paper, which suggest a trade-off between
target and trigger word predictability in nasal assimilation, be
extended to explain related phenomena like final consonant deletion in
casual speech (e.g., "friend" pronounced as "frien")? For instance, is
a highly predictable target more likely to undergo deletion when
followed by a less predictable trigger?

Could the trade-off framework proposed by the authors be extended to
explain syntactic phenomena, such as subject omission in predictable
contexts, as observed in pro-drop languages like Spanish?

Why is there no effect of within-word predictability on partial
assimilation?  The authors mention "different mechanisms for full
assimilation versus partial assimilation," which seems like a vague
explanation; could there be other explanations for this phenomenon?

Could you explain the main idea behind linear predictive coding ?

Why are frequency and predictability estimates log-transformed?