Phonological Alternation, Neighborhood Density, and Markedness in
Processing.

Abstract

Recent work suggests a pervasive link between crosslinguistically attested
sound patterns and differences in articulatory ease or accuracy of
perception (e.g. Steriade 1997). By the same token, if some kinds of
strings are more easily processed than others, we might predict this
difference to be reflected crosslinguistically in a relative
under-representation for more difficult strings. Here I report a
correlation between contextual contrast preservation and predictions on
processing efficiency and accuracy under the Neighborhood Activation Model
(NAM, Luce and Pisoni 1998 and references therein).

According to NAM, lexical entries are recognized in the context of
phonologically similar lexical entries. A stimulus input activates a
similarity set, or neighborhood, of entries in memory which all compete
for recognition. NAM predicts that entries in high-density phonological
neighborhoods will be recognized less quickly and accurately. These
predictions have been borne out by a broad series of studies of the last
decade (e.g., Goldinger et al, 1989, Cluff and Luce 1990).
 

Under variants of both one-morpheme-one-lexical entry and full-listing
morphological theories, phonological alternation is predicted to increase
the effective neighborhood density of a lexical entry. Therefore, because
the difficulty of matching an output to a lexical entry increases with
neighborhood density, the cost of alternation should also increase with
increasing neighborhood density, leading to the hypothesis that languages
will avoid alternation in high-density neighborhoods. I report here
initial tests of these predictions in the context of suffix-conditioned
stem-final alternation in Turkish and Catalan, and stress-shift-conditions
vowel alternation in Catalan.
 

My initial approach has been to use morpheme size as a proxy for the
more-difficult-to-measure neighborhood density, as the two are known to
have a strong inverse correlation, and as predicted, there is indeed a
significant tendency for small stems not to alternate, relative to larger
ones. I am now carrying out more detailed analyses to determine whether
the correlation is in fact tighter with neighborhood density as predicted
by the model. A possible alternative explanation based on morpheme-size
will be introduced, and both possible diachronic and synchronic
explanations for the pattern will be discussed.

[back]