Computation And Language

This paper presents an integrated tool for German morphology and statistical part-of-speech tagging which aims at making some well established methods widely available. The software is very user friendly, runs on any PC and can be downloaded as a complete package (including lexicon and documentation) from the World Wide Web. Compared with the performance of other tagging systems the tagger produces similar results.

Consider the problem of multinomial estimation. You are given an alphabet of k distinct symbols and are told that the i-th symbol occurred exactly n_i times in the past. On the basis of this information alone, you must now estimate the conditional probability that the next symbol will be i. In this report, we present a new solution to this fundamental problem in statistics and demonstrate that our solution outperforms standard approaches, both in theory and in practice.

This paper describes a new statistical parser which is based on probabilities of dependencies between head-words in the parse tree. Standard bigram probability estimation techniques are extended to calculate probabilities of dependencies between pairs of words. Tests using Wall Street Journal data show that the method performs at least as well as SPATTER (Magerman 95, Jelinek et al 94), which has the best published results for a statistical parser on this task. The simplicity of the approach means the model trains on 40,000 sentences in under 15 minutes. With a beam search strategy parsing speed can be improved to over 200 sentences a minute with negligible loss in accuracy.

The first step in most empirical work in multilingual NLP is to construct maps of the correspondence between texts and their translations ({\bf bitext maps}). The Smooth Injective Map Recognizer (SIMR) algorithm presented here is a generic pattern recognition algorithm that is particularly well-suited to mapping bitext correspondence. SIMR is faster and significantly more accurate than other algorithms in the literature. The algorithm is robust enough to use on noisy texts, such as those resulting from OCR input, and on translations that are not very literal. SIMR encapsulates its language-specific heuristics, so that it can be ported to any language pair with a minimal effort.

Most existing natural language interfaces to databases (NLIDBs) were designed to be used with ``snapshot'' database systems, that provide very limited facilities for manipulating time-dependent data. Consequently, most NLIDBs also provide very limited support for the notion of time. The database community is becoming increasingly interested in _temporal_ database systems. These are intended to store and manipulate in a principled manner information not only about the present, but also about the past and future. This thesis develops a principled framework for constructing English NLIDBs for _temporal_ databases (NLITDBs), drawing on research in tense and aspect theories, temporal logics, and temporal databases. I first explore temporal linguistic phenomena that are likely to appear in English questions to NLITDBs. Drawing on existing linguistic theories of time, I formulate an account for a large number of these phenomena that is simple enough to be embodied in practical NLITDBs. Exploiting ideas from temporal logics, I then define a temporal meaning representation language, TOP, and I show how the HPSG grammar theory can be modified to incorporate the tense and aspect account of this thesis, and to map a wide range of English questions involving time to appropriate TOP expressions. Finally, I present and prove the correctness of a method to translate from TOP to TSQL2, TSQL2 being a temporal extension of the SQL-92 database language. This way, I establish a sound route from English questions involving time to a general-purpose temporal database language, that can act as a principled framework for building NLITDBs. To demonstrate that this framework is workable, I employ it to develop a prototype NLITDB, implemented using ALE and Prolog.

We address the problem of structural disambiguation in syntactic parsing. In psycholinguistics, a number of principles of disambiguation have been proposed, notably the Lexical Preference Rule (LPR), the Right Association Principle (RAP), and the Attach Low and Parallel Principle (ALPP) (an extension of RAP). We argue that in order to improve disambiguation results it is necessary to implement these principles on the basis of a probabilistic methodology. We define a `three-word probability' for implementing LPR, and a `length probability' for implementing RAP and ALPP. Furthermore, we adopt the `back-off' method to combine these two types of probabilities. Our experimental results indicate our method to be effective, attaining an accuracy of 89.2%.

This paper explores commonalities and differences between \dachs, a variant of Dependency Grammar, and Lexical-Functional Grammar. \dachs\ is based on traditional linguistic insights, but on modern mathematical tools, aiming to integrate different knowledge systems (from syntax and semantics) via their coupling to an abstract syntactic primitive, the dependency relation. These knowledge systems correspond rather closely to projections in LFG. We will investigate commonalities arising from the usage of the projection approach in both theories, and point out differences due to the incompatible linguistic premises. The main difference to LFG lies in the motivation and status of the dimensions, and the information coded there. We will argue that LFG confounds different information in one projection, preventing it to achieve a good separation of alternatives and calling the motivation of the projection into question.

In this paper we present a robust parsing algorithm based on the link grammar formalism for parsing natural languages. Our algorithm is a natural extension of the original dynamic programming recognition algorithm which recursively counts the number of linkages between two words in the input sentence. The modified algorithm uses the notion of a null link in order to allow a connection between any pair of adjacent words, regardless of their dictionary definitions. The algorithm proceeds by making three dynamic programming passes. In the first pass, the input is parsed using the original algorithm which enforces the constraints on links to ensure grammaticality. In the second pass, the total cost of each substring of words is computed, where cost is determined by the number of null links necessary to parse the substring. The final pass counts the total number of parses with minimal cost. All of the original pruning techniques have natural counterparts in the robust algorithm. When used together with memoization, these techniques enable the algorithm to run efficiently with cubic worst-case complexity. We have implemented these ideas and tested them by parsing the Switchboard corpus of conversational English. This corpus is comprised of approximately three million words of text, corresponding to more than 150 hours of transcribed speech collected from telephone conversations restricted to 70 different topics. Although only a small fraction of the sentences in this corpus are "grammatical" by standard criteria, the robust link grammar parser is able to extract relevant structure for a large portion of the sentences. We present the results of our experiments using this system, including the analyses of selected and random sentences from the corpus.

This paper describes a system that leads us to believe in the feasibility of constructing natural spoken dialogue systems in task-oriented domains. It specifically addresses the issue of robust interpretation of speech in the presence of recognition errors. Robustness is achieved by a combination of statistical error post-correction, syntactically- and semantically-driven robust parsing, and extensive use of the dialogue context. We present an evaluation of the system using time-to-completion and the quality of the final solution that suggests that most native speakers of English can use the system successfully with virtually no training.

This thesis addresses automatic lexical error recovery and tokenization of corrupt text input. We propose a technique that can automatically correct misspellings, segmentation errors and real-word errors in a unified framework that uses both a model of language production and a model of the typing behavior, and which makes tokenization part of the recovery process. The typing process is modeled as a noisy channel where Hidden Markov Models are used to model the channel characteristics. Weak statistical language models are used to predict what sentences are likely to be transmitted through the channel. These components are held together in the Token Passing framework which provides the desired tight coupling between orthographic pattern matching and linguistic expectation. The system, CTR (Connected Text Recognition), has been tested on two corpora derived from two different applications, a natural language dialogue system and a transcription typing scenario. Experiments show that CTR can automatically correct a considerable portion of the errors in the test sets without introducing too much noise. The segmentation error correction rate is virtually faultless.