Schedule
Session 1 (Chair, Robert Munro)
1:15 - 1:45 Matt Adams: Predicting verb duration: isochrony in the V-bar
- Abstract ±
Recent work on predictability and frequency in lexical production provides strong support for the Probabilistic Reduction Hypothesis, which holds that lexemes are reduced in spontaneous speech when they are more predictable (Jurafsky et al. 2001; Bell et al. 1999; Jurafsky et al. 1998). In this talk I provide preliminary results from a corpus experiment that focuses on the duration of English transitive verbs. I show that frequency and local predictability of the noun following the verb have an unexpected effect on verb duration: as the noun becomes more predictable, the duration of the verb is longer. This effect is explained in terms of phonetic isochrony (Lehiste 1972), which suggests that similar units of speech have similar durations, all else being equal. The lengthening of the verb within the V-bar can be explained as a compensatory effect that results from the shorter duration of the noun. Verb duration also decreases as a function of increasing verb frequency, thereby replicating predictions about the relationship between frequency and reduction.
1:45 - 2:15 Stephanie Shih: Rhythm's role in the genitive and dative construction choice in spoken English
- Abstract ±
TBA
2:15 - 2:30 Break
Session 2 (Chair, Kyuwon Moon)
2:30 - 3:00 Sven Lauer: Free relatives with -EVER: Meaning and Use
- Abstract ±
The history of the study of the semantics of English -EVER free relatives is a history full of misunderstandings: They have been supposed to contain a modal assertion, modal presupposition or a modal conventional implicature.
This talk shows that neither is the contribution of -EVER necessarily modal, nor is this contribution asserted, presupposed or conventionally implicated.
Rather, -EVER serves to mark the fact that the referent of the free relative is not resolved with respect to its local context, which may be the domain of an outscoping quantificational operator (a modal, attitude verb, iterative aspect, adverb of quantification or individual quantifier), or the conversational `common ground'. In the latter case, the use of an -EVER free relative will trigger additional conversational implicatures which, in their content, may look very similar to the implications of -EVER free relatives under modals, but demonstrably behave differently.
The contribution of -EVER is formally modeled as a post-supposition, that is, a felicity requirement, that, unlike a presupposition, is imposed on output- rather than input-contexts.
The proposed analysis extends naturally to items in other languages, such as German IRGEND-indefinites and Spanish ALGUN, which previously have been treated very differently than -EVER free relatives, despite obvious similarities between these items.
3:00 - 3:30 Robert Munro: Linguistic features improving Nominal Semantic Role Labeling
- Abstract ±
Semantic Role Labeling for nominal predicates is more difficult than for verbal predicates, largely because syntax plays a more peripheral role. For example, in a phrase like "the investigation of the police" it is not clear whether "the police" are investigating something or being investigated. Where the syntax underspecifies, richer linguistic features can disambiguate a given role. For example, there is little ambiguity in "the investigation of the crime" as "the crime" lacks the animacy required to investigate something. This paper reports, for the first time, how features like animacy, constituent size, and interactions with sentential roles can be incorporated into a semantic role labeling system. The last of these features is the most interesting, with nominal agents much more likely to be embedded within sentential agents, and so on for other roles, demonstrating a tendency towards role harmony between sentential arguments and embedded nominal arguments. By training and testing on the NomBank corpus, it is demonstrated that these linguistic features produce a much more robust system, consistently classifying with greater accuracy than current state-of-the-art systems, especially across previously unseen predicates and arguments.
3:30 - 4:00 Tyler Schnoebelen: (Un)classifying Shabo
- Abstract ±
Shabo is one of our greatest classification puzzles. Most researchers describe it as Nilo-Saharan (Bender 1977, Bender 1983, Teferra and Unseth 1989, Teferra 1991, Fleming 1991), but each of them note how much of the core vocabulary seems to have been borrowed. In fact, this is one of the reasons Ehret (1995) takes Shabo out of Nilo-Saharan: there are just too many similarities to too many different languages, he says.
Bender (1977) was the first to attempt categorization and placed Shabo with the Surmic languages—a decision that was probably unduly influenced by the fact that most Shabo speakers also know Majang. Bender updated his position in 1983, still favoring Nilo-Saharan. By the mid-1990s, Bender was more circumspect: "We just don't know. I think it is one of several 'strange hybrids' (e.g. Kwegu, Birale, and Mao)" (p.c. with Teferra 1995).
Most of the classification work on Shabo has been done based on wordlists. I add grammatical data based on a month of living amongst the Shabo in a remote Ethiopian village. Since the Shabo traditionally live alongside the Majang and/or the Shekkacho, I include parallel data I acquired on those languages, too.
Wichmann and Holman (forthcoming) have used computational models to determine which features in the World Atlas of Languages are the most stable for predicting genetic relationships. I examine Shabo in terms of the top 40 features, comparing it with about 250 languages of the region. Features like presence of initial velar nasals, case syncretism, and word order, supplemented by lexical comparisons of the core vocabulary, show which languages Shabo has affinities with.
Beyond documenting a highly endangered language (there are perhaps 400 speakers of Shabo), this paper offers an innovative use of phylogenetic tools that biologists have developed. I demonstrate the potential power of such methods in building classifications and exploring the increasing amounts of typological data available to us.
4:00 Social!