<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1107">
  <Title>Stochastic phonological grammars and acceptability</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Discussion and Conclusions
</SectionTitle>
    <Paragraph position="0"> We have compared several methods of using frequency information to predict the acceptability of neologisms. Both the probability of the word and the probability of the worst part are significant correlates of acceptability. This finding is significant, because the single worst violation dominates the determination of well-formedness in almost all varsions of generative. phonology. In Chomsky and Halle (1968), morpheme structure conditions act as a filter on underlying representations. The same concept of grammatically is proposed in approaches founded on Boolean logic, such as Declarative Phonology.</Paragraph>
    <Paragraph position="1"> According to Optimality Theory, &amp;quot;impossible words&amp;quot; are those in which a constraint is so strong that a null parse is prefered to a parse in which the constraint is violated. This means that impossible words are those which are egregious according to a single constraint.</Paragraph>
    <Paragraph position="2"> However, the probability of the worst part is not the best score of acceptability: the log probability of the whole word is a better measure, a result at odds with standard generative phonology and OT alike. In classical generative phonology, a UR which violates any single morpheme structure condition is ruled out absolutely. In more recent versions of generative phonology which build prosodic structure through some version of parsing or template mapping, the entire parse fails if it fails at any single point. The same idea shows up in a new guise in Optimality Theory. According to Optimality Theory, constraint violations do not interact cumulatively. A rank-ordering of constraints has the consequence that weak constraints can be violated to meet stronger ones, but there is no mechanism by which adherence to many weak constraints ameliorates the effect of a single violation of a stronger constraint. Our results indicate that these models achieve some success, but miss an important fact: the well-formedness of lexically attested parts ameliorates the unacceptability of the unattested or low-frequency parts. When statistically valid data on acceptability is gathered (as against the isolated intuitions of individual researchers/authors), it is found that deviations are partially redeemed by good parts, and that forms which are locally wellformed, in the sense that each piece is reasonably well-attested, can nonetheless be viewed as improbable overall. This finding supports the view that phonotactic constraints are probabilistic descriptions of the lexicon, and that probabilisitic generative grammars are a more psychologically realistic model of phonological competence than standard generative phonology and Optimality Theory.</Paragraph>
  </Section>
class="xml-element"></Paper>