Computation And Language

In this paper, I give an account of some phenomena of pronominalization in Italian in terms of centering theory. After a general introduction to the Italian pronominal system, I will review centering, and then show how the original rules have to be extended or modified. Finally, I will show that centering does not account for two phenomena: first, the functional role of an utterance may override the predictions of centering; second, a null subject can be used to refer to a whole discourse segment.

Centering was formulated as a model of the relationship between attentional state, the form of referring expressions, and the coherence of an utterance within a discourse segment (Grosz, Joshi and Weinstein, 1986; Grosz, Joshi and Weinstein, 1995). In this chapter, I argue that the restriction of centering to operating within a discourse segment should be abandoned in order to integrate centering with a model of global discourse structure. The within-segment restriction causes three problems. The first problem is that centers are often continued over discourse segment boundaries with pronominal referring expressions whose form is identical to those that occur within a discourse segment. The second problem is that recent work has shown that listeners perceive segment boundaries at various levels of granularity. If centering models a universal processing phenomenon, it is implausible that each listener is using a different centering algorithm.The third issue is that even for utterances within a discourse segment, there are strong contrasts between utterances whose adjacent utterance within a segment is hierarchically recent and those whose adjacent utterance within a segment is linearly recent. This chapter argues that these problems can be eliminated by replacing Grosz and Sidner's stack model of attentional state with an alternate model, the cache model. I show how the cache model is easily integrated with the centering algorithm, and provide several types of data from naturally occurring discourses that support the proposed integrated model. Future work should provide additional support for these claims with an examination of a larger corpus of naturally occurring discourses.

In this paper we present early work on an animated talking head commentary system called {\bf Byrne}\footnote{David Byrne is the lead singer of the Talking Heads.}. The goal of this project is to develop a system which can take the output from the RoboCup soccer simulator, and generate appropriate affective speech and facial expressions, based on the character's personality, emotional state, and the state of play. Here we describe a system which takes pre-analysed simulator output as input, and which generates text marked-up for use by a speech generator and a face animation system. We make heavy use of inter-system standards, so that future versions of Byrne will be able to take advantage of advances in the technologies that it incorporates.

Recently researchers working in the LFG framework have proposed algorithms for taking advantage of the implicit context-free components of a unification grammar [Maxwell 96]. This paper clarifies the mathematical foundations of these techniques, provides a uniform framework in which they can be formally studied and eliminates the need for special purpose runtime data-structures recording ambiguity. The paper posits the identity: Ambiguous Feature Structures = Grammars, which states that (finitely) ambiguous representations are best seen as unification grammars of a certain type, here called ``interaction-free'' grammars, which generate in a backtrack-free way each of the feature structures subsumed by the ambiguous representation. This work extends a line of research [Billot and Lang 89, Lang 94] which stresses the connection between charts and grammars: a chart can be seen as a specialization of the reference grammar for a given input string. We show how this specialization grammar can be transformed into an interaction-free form which has the same practicality as a listing of the individual solutions, but is produced in less time and space.

We describe a stochastic approach to partial parsing, i.e., the recognition of syntactic structures of limited depth. The technique utilises Markov Models, but goes beyond usual bracketing approaches, since it is capable of recognising not only the boundaries, but also the internal structure and syntactic category of simple as well as complex NP's, PP's, AP's and adverbials. We compare tagging accuracy for different applications and encoding schemes.

Increasingly, inheritance hierarchies are being used to reduce redundancy in natural language processing lexicons. Systems that utilize inheritance hierarchies need to be able to insert words under the optimal set of classes in these hierarchies. In this paper, we formalize this problem for feature-based default inheritance hierarchies. Since the problem turns out to be NP-complete, we present an approximation algorithm for it. We show that this algorithm is efficient and that it performs well with respect to a number of standard problems for default inheritance. A prototype implementation has been tested on lexical hierarchies and it has produced encouraging results. The work presented here is also relevant to other types of default hierarchies.

This paper proposes an analysis of classifiers into four major types: UNIT, METRIC, GROUP and SPECIES, based on properties of both Japanese and English. The analysis makes possible a uniform and straightforward treatment of noun phrases headed by classifiers in Japanese-to-English machine translation, and has been implemented in the MT system ALT-J/E. Although the analysis is based on the characteristics of, and differences between, Japanese and English, it is shown to be also applicable to the unrelated language Thai.

The maximum entropy method has recently been successfully introduced to a variety of natural language applications. In each of these applications, however, the power of the maximum entropy method is achieved at the cost of a considerable increase in computational requirements. In this paper we present a technique, closely related to the classical cluster expansion from statistical mechanics, for reducing the computational demands necessary to calculate conditional maximum entropy language models.

In this paper, a hierarchical context definition is added to an existing clustering algorithm in order to increase its robustness. The resulting algorithm, which clusters contexts and events separately, is used to experiment with different ways of defining the context a language model takes into account. The contexts range from standard bigram and trigram contexts to part of speech five-grams. Although none of the models can compete directly with a backoff trigram, they give up to 9\% improvement in perplexity when interpolated with a trigram. Moreover, the modified version of the algorithm leads to a performance increase over the original version of up to 12\%.

We address the problem of automatically constructing a thesaurus (hierarchically clustering words) based on corpus data. We view the problem of clustering words as that of estimating a joint distribution over the Cartesian product of a partition of a set of nouns and a partition of a set of verbs, and propose an estimation algorithm using simulated annealing with an energy function based on the Minimum Description Length (MDL) Principle. We empirically compared the performance of our method based on the MDL Principle against that of one based on the Maximum Likelihood Estimator, and found that the former outperforms the latter. We also evaluated the method by conducting pp-attachment disambiguation experiments using an automatically constructed thesaurus. Our experimental results indicate that we can improve accuracy in disambiguation by using such a thesaurus.