Moreton, Elliott (2007).  Learning bias as a factor in phonological
typology.  Presentation at the 26th meeting of the West Coast Conference
on Formal Linguistics, Berkeley, California, April 27-29.

This talk addresses two competing explanations for why some
phonological patterns are more common than others:

* "Analytic bias": Cognitive biases such as Universal Grammar
   make some patterns harder to learn than others.

* "Channel bias": Phonological patterns are innovated
   when subtle phonetic precursors are reinterpreted as phonological;
   the more robust the precursor, the more common the phonological
   pattern.

These explanations are hard to distinguish because they apply equally
well to much typological data.  However, evidence is presented that
each factor can shape typology in ways not explainable by the other.

(1) When other factors are controlled, phonological patterns relating
vowel height to vowel height (e.g., height harmony) are typologically
more frequent, in terms of language families, than patterns relating
height to consonant voicing.  (2) The phonetic precursor of the
height-height patterns is no larger than that of the height-voice
patterns.  (3) In an artificial-language experiment with 24 American
English speakers, a phonological height-height pattern (height
harmony) was learned better than a height-voice pattern.  This
finding, together with (2), suggests that analytic bias explains (1)
better than does precursor robustness.  (4) In another experiment, a
long-distance voicing harmony pattern -- very rare in nature -- was
also learned better than a height-voice pattern.  This suggests that
the rarity of voice-harmony patterns is due to lack of a precursor,
rather than to analytic bias.  (5) Two more experiments replicated the
height-height advantage over height-voice, and found that a
height-height pattern is also learned better than a height-backness
pattern.  This suggests that the advantageous property of the
height-height pattern is that it involves recurrence of a single
feature.

Exclusive focus on one factor or the other as the determinant of
typology is thus liable to be misleading.