<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1241">
  <Title>Reconciliation of Unsupervised Clustering, Segmentation and Cohesion</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3. Semantics and Synonymy
</SectionTitle>
    <Paragraph position="0"> At the semantic level, it supports our denial of the Powers 307 Unsupervised Clustering, Segmentation and Cohesion David M. W. Powers (1998) Reconciliation of Unsupervised Clustering, Segmentation and Cohesion. In D.M.W. Powers (ed.) NeMLaP3/CoNLL 98 Workshop on Paradigms and Grounding in Language Learning, ACL, pp 307-310. existence of pure synonymy. Thus words like 'too' and 'also' which are apparent synonyms have quite different syntactic constraints, while words like &amp;quot;small' and 'little' which are apparent synonyms and appear to occupy the same same syntactic role, actually have quite different connotations. Thus 'a small boy' is small for his age, whilst 'a little boy' is a young child, and 'a small little boy' combines these implications; the tendency for 'little' to prefer and be preferred when a metaphorical interpretation is appropriate is confirmed by idioms like 'a little while' and 'a little bit&amp;quot; whereas 'small' tends to have more direct connection to the underlying spatial interpretation, and when used in a metaphorical or temporal context it thus tends to reinforce the metaphor and supply additional emphasis -- contrast 'except for one little detail' and 'except for one small detail'. Of course, any examples of this sort are highly influenced by the specific language, dialect and idiolect of the speaker and may vary at each level.</Paragraph>
    <Paragraph position="1"> Many researchers have used clustering techniques to induce semantic classes (e.g. Finch, 1993), although these have tended again to be non-hierarchical except to the extent that a pairwise clustering technique induces a dendritic structure on the semantic space (although Finch did perform two levels of analysis in one experiment).</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4. Morphology and Syntax
</SectionTitle>
    <Paragraph position="0"> At the syntactic level the assumption of absence of free variation is not so controversial, and although generative grammarians have tended to treat some choices as arbitrary, e.g. the choice between active and passive, which is probably more a function of their ignoring pragmatics to focus on grammar. At the level of morphology, what may appear to be free variation synchronically usually has a diachronic explanation, and invariably involves clear complementarity in terms of the distribution of allomorph according to the syntactic role of the embedding word.</Paragraph>
    <Paragraph position="1"> The experiments of Powers (1992) demonstrated both learning of classes and hierarchical rules from the character level up to the level of simple noun phrases and simple clauses. As is the case with subjaeency, noun phrases and clauses tend to act similarly, and indeed we propose that they themselves form a complementary distribution (involving their multiple forms, including nominalized clauses and verbs: 'he wanted the girl to come', 'the girl must come', 'he decided that the girl should come', 'he decided the girl could come') and suggests a generalization of the finiteness feature of verbs should apply to both nouns and verbs and their dominated structures ('to' and 'that' axe both optional markers of the infinite form; the finite form would appear to be the default role of a verb and the unmarked form).</Paragraph>
  </Section>
class="xml-element"></Paper>