|[an error occurred while processing this directive]||
Doctor Ellen Prince
Semantic Reference vs. Discourse Reference.
Impersonal subject pronouns in certain languages have recently received a fair amount of attention because they have a number of unusual properties. For example, Koenig 1999 notes that, unlike ordinary indefinites like the subject in 1a, French on 'one' does not introduce a discourse entity that can then be referred to anaphorically, as in 1b::
1. a. Quelqu'un-i a tué la présidente! Il-i était du Berry, paraît-il.
'Someone-i killed the (female) president! He-i was from the Berry, it seems.'
b. On-i a tué la présidente! #Il-i était du Berry, paraît-il.
'One-i killed the (female) president! #He-i was from the Berry, it seems.'
Koenig proposes to account for this difference by positing two types of indefinites, one, the well-known kind, that introduces discourse entities, and another, which he calls 'ultra-indefinites' (later 'a-definites'), that serves only to satisfy a predicate's argument positions. A serious problem with this approach is that the first sentence of 1b, just like the first sentence of 1a, entails 2; in contrast, pace Koenig, a predicate's argument positions can be satisfied, as in 3, with no such entailment:
2. A person killed the (female) president.
3. Personne n'a tué la présidente.
'No one killed the (female) president.'
I believe such approaches are misguided in assuming that a single component must simultaneously do the semantic work, e.g. accounting for the truth-conditional content of what is said, and also the pragmatic/discourse work, e.g. modeling attention in the processing of discourse, necessary for anaphora resolution. For the modeling of attention, I shall propose Centering Theory (Grosz, Joshi, and Weinstein 1995), which will be shown to account not only for the anaphora facts in 1 but also for other otherwise inexplicable properties of items like on 'one', namely that they have a defective paradigm, occurring only in the nominative and only as subjects of finite verbs, that they can in fact both introduce new entities into the discourse model and also refer to entities that are already in the discourse model, and that they cannot be focused. Moreover, such an approach makes it clear why a language might have such items and suggests in what sort of languages they are likely to occur. Data will be taken primarily from Yiddish, where me(n) 'one' appears to work exactly like French on 'one'.