• seminar

seminar

SEMINAR

Abstract: In this talk, I reflect on what we can learn about the processes involved in the generation of referring expressions by looking at a corpus of human-produced data. We find that the data vastly underspecifies what might be involved algorithmically, although it does rule out a number of popular algorithms for referring expression generation as candidates for models of what people do. We posit an alternative algorithmic schema which forces us to focus more clearly on the questions that need to be answered experimentally if we are to develop algorithms that emulate human behaviour on this task.

Date: 2009-06-11 10:15 - 12:00

Location:

Permalink

SEMINAR

One of the currently debated topics in Natural Language Processing is the problem of semantic role labelling. Given a sentence like "I want to reserve a flight from Geneva to Gotenborg'', how do we determine automatically that "from Geneva'' is the source of the flight and "to Gotenborg'' is the destination? The solution to this problem lies at the heart of all applications that require language understanding, such as dialogue systems, question answering, or machine translation. Recent successes of machine learning methods in statistical parsing and lexical acquisition pave the way to a learning approach for this problem too. In this talk, I will report experiments that explore learning of syntactic and semantic representations. First, we extend a state-of-the-art statistical parser to produce a richly annotated tree that identifies and labels nodes with semantic role labels as well as syntactic labels. Secondly, we explore rule-based and learning techniques to extract predicate-argument structures from this enriched output. The learning method is competitive with previous single-system proposals for semantic role labelling, yields the best reported precision, and produces a rich output. In combination with other high-recall semantic role labelling systems it yields an F-measure of 81%.

Date: 2008-05-30 15:15 - 17:00

Location: Humanisten F236

Permalink

SEMINAR

(Stanley Peters and Dag Westerståhl)

The talk develops the account of possessives begun in our book Quantifiers in Language and Logic (Oxford UP, 2006, ch. 7; familiarity with it is not presupposed). One novelty is that we consider post-nominal as well as pre-nominal possessive constructions, and that we do not focus on particular features of English. In contrast with most authors, we find the similarities between possessive constructions across languages more striking than the differences.

A main claim is that (real) possessives always involve two quantifiers: one over the 'possessors' and one over the 'possessed' objects. We show that the interplay between these two explains features of possessives that have appeared puzzling when one restricts attention (as does 95% of the literature) to very simple possessives where these quantifiers are not visible on the surface. In particular, it helps explain the role of the definite article in possessive contexts, and why it has seemed (erroneously) to almost all linguists that possessives must be definite. (Cf. the synonymous sentences "At least two teachers' pupils are bright" and "The pupils of at least two teachers are bright".)

Two further themes will be discussed (time permitting). One is the form of possessives, in view of the fact that the possessed noun may be relational or not ("Mary's sisters" vs. "Mary's cars"; this was the topic of Barbara Partee's lecture on possessives in Göteborg one year ago). The other is the ubiquity of narrowing (as observed in Barker 1995): that the first quantifier mentioned above is restricted to those individuals who 'possess' something of the relevant kind. (E.g. "Most beginning assistant professors' children were present at the opening" says nothing about the childless beginning assistant professors.) This appears to have an interesting consequence for the compositionality of possessive constructions.

 

 

Date: 2007-12-13 10:15 - 12:00

Location: F304

Permalink

SEMINAR

This paper will be presented at Bar Ilan University June 20.
http://www.cs.biu.ac.il/bisfai07/                                                                        

Multilingual generation of medical information
----------------------------------------------------------------------
Dana Dannélls, Göteborg University and
Louise Deléger, Université René Descartes, Paris

The paper addresses the problem of adapting a template-based generator from English to Swedish and French. We present the language generation system and describe the adaptation to both languages, viewing the grammatical aspects involved and explaining the modifications performed. This work illustrates how the same underlying knowledge representation can be used to generate output texts in multiple languages with only minor linguistic modifications.

Date: 2007-05-30 15:15 - 16:00

Location:

Permalink

X
Loading