Datasets:

MaartenGr

/

arxiv_nlp

like 2

Utilizing the World Wide Web as an Encyclopedia: Extracting Term Descriptions from Semi-Structured Texts

In this paper, we propose a method to extract descriptions of technical terms from Web pages in order to utilize the World Wide Web as an encyclopedia. We use linguistic patterns and HTML text structures to extract text fragments containing term descriptions. We also use a language model to discard extraneous descriptions, and a clustering method to summarize resultant descriptions. We show the effectiveness of our method by way of experiments.

Computation and Language

A Novelty-based Evaluation Method for Information Retrieval

In information retrieval research, precision and recall have long been used to evaluate IR systems. However, given that a number of retrieval systems resembling one another are already available to the public, it is valuable to retrieve novel relevant documents, i.e., documents that cannot be retrieved by those existing systems. In view of this problem, we propose an evaluation method that favors systems retrieving as many novel documents as possible. We also used our method to evaluate systems that participated in the IREX workshop.

Computation and Language

Applying Machine Translation to Two-Stage Cross-Language Information Retrieval

Cross-language information retrieval (CLIR), where queries and documents are in different languages, needs a translation of queries and/or documents, so as to standardize both of them into a common representation. For this purpose, the use of machine translation is an effective approach. However, computational cost is prohibitive in translating large-scale document collections. To resolve this problem, we propose a two-stage CLIR method. First, we translate a given query into the document language, and retrieve a limited number of foreign documents. Second, we machine translate only those documents into the user language, and re-rank them based on the translation result. We also show the effectiveness of our method by way of experiments using Japanese queries and English technical documents.

Computation and Language

Tree-gram Parsing: Lexical Dependencies and Structural Relations

This paper explores the kinds of probabilistic relations that are important in syntactic disambiguation. It proposes that two widely used kinds of relations, lexical dependencies and structural relations, have complementary disambiguation capabilities. It presents a new model based on structural relations, the Tree-gram model, and reports experiments showing that structural relations should benefit from enrichment by lexical dependencies.

Computation and Language

The Use of Instrumentation in Grammar Engineering

This paper explores the usefulness of a technique from software engineering, code instrumentation, for the development of large-scale natural language grammars. Information about the usage of grammar rules in test and corpus sentences is used to improve grammar and testsuite, as well as adapting a grammar to a specific genre. Results show that less than half of a large-coverage grammar for German is actually tested by two large testsuites, and that 10--30% of testing time is redundant. This methodology applied can be seen as a re-use of grammar writing knowledge for testsuite compilation.

Computation and Language

Retrieval from Captioned Image Databases Using Natural Language Processing

It might appear that natural language processing should improve the accuracy of information retrieval systems, by making available a more detailed analysis of queries and documents. Although past results appear to show that this is not so, if the focus is shifted to short phrases rather than full documents, the situation becomes somewhat different. The ANVIL system uses a natural language technique to obtain high accuracy retrieval of images which have been annotated with a descriptive textual caption. The natural language techniques also allow additional contextual information to be derived from the relation between the query and the caption, which can help users to understand the overall collection of retrieval results. The techniques have been successfully used in a information retrieval system which forms both a testbed for research and the basis of a commercial system.

Computation and Language

Semantic interpretation of temporal information by abductive inference

Besides temporal information explicitly available in verbs and adjuncts, the temporal interpretation of a text also depends on general world knowledge and default assumptions. We will present a theory for describing the relation between, on the one hand, verbs, their tenses and adjuncts and, on the other, the eventualities and periods of time they represent and their relative temporal locations. The theory is formulated in logic and is a practical implementation of the concepts described in Ness Schelkens et al. We will show how an abductive resolution procedure can be used on this representation to extract temporal information from texts.

Computation and Language

Abductive reasoning with temporal information

Texts in natural language contain a lot of temporal information, both explicit and implicit. Verbs and temporal adjuncts carry most of the explicit information, but for a full understanding general world knowledge and default assumptions have to be taken into account. We will present a theory for describing the relation between, on the one hand, verbs, their tenses and adjuncts and, on the other, the eventualities and periods of time they represent and their relative temporal locations, while allowing interaction with general world knowledge. The theory is formulated in an extension of first order logic and is a practical implementation of the concepts described in Van Eynde 2001 and Schelkens et al. 2000. We will show how an abductive resolution procedure can be used on this representation to extract temporal information from texts. The theory presented here is an extension of that in Verdoolaege et al. 2000, adapted to VanEynde 2001, with a simplified and extended analysis of adjuncts and with more emphasis on how a model can be constructed.

Computation and Language

Easy and Hard Constraint Ranking in OT: Algorithms and Complexity

We consider the problem of ranking a set of OT constraints in a manner consistent with data. We speed up Tesar and Smolensky's RCD algorithm to be linear on the number of constraints. This finds a ranking so each attested form x_i beats or ties a particular competitor y_i. We also generalize RCD so each x_i beats or ties all possible competitors. Alas, this more realistic version of learning has no polynomial algorithm unless P=NP! Indeed, not even generation does. So one cannot improve qualitatively upon brute force: Merely checking that a single (given) ranking is consistent with given forms is coNP-complete if the surface forms are fully observed and Delta_2^p-complete if not. Indeed, OT generation is OptP-complete. As for ranking, determining whether any consistent ranking exists is coNP-hard (but in Delta_2^p) if the forms are fully observed, and Sigma_2^p-complete if not. Finally, we show that generation and ranking are easier in derivational theories: in P, and NP-complete.

Computation and Language

Multi-Syllable Phonotactic Modelling

This paper describes a novel approach to constructing phonotactic models. The underlying theoretical approach to phonological description is the multisyllable approach in which multiple syllable classes are defined that reflect phonotactically idiosyncratic syllable subcategories. A new finite-state formalism, OFS Modelling, is used as a tool for encoding, automatically constructing and generalising phonotactic descriptions. Language-independent prototype models are constructed which are instantiated on the basis of data sets of phonological strings, and generalised with a clustering algorithm. The resulting approach enables the automatic construction of phonotactic models that encode arbitrarily close approximations of a language's set of attested phonological forms. The approach is applied to the construction of multi-syllable word-level phonotactic models for German, English and Dutch.

Computation and Language

Taking Primitive Optimality Theory Beyond the Finite State

Primitive Optimality Theory (OTP) (Eisner, 1997a; Albro, 1998), a computational model of Optimality Theory (Prince and Smolensky, 1993), employs a finite state machine to represent the set of active candidates at each stage of an Optimality Theoretic derivation, as well as weighted finite state machines to represent the constraints themselves. For some purposes, however, it would be convenient if the set of candidates were limited by some set of criteria capable of being described only in a higher-level grammar formalism, such as a Context Free Grammar, a Context Sensitive Grammar, or a Multiple Context Free Grammar (Seki et al., 1991). Examples include reduplication and phrasal stress models. Here we introduce a mechanism for OTP-like Optimality Theory in which the constraints remain weighted finite state machines, but sets of candidates are represented by higher-level grammars. In particular, we use multiple context-free grammars to model reduplication in the manner of Correspondence Theory (McCarthy and Prince, 1995), and develop an extended version of the Earley Algorithm (Earley, 1970) to apply the constraints to a reduplicating candidate set.

Computation and Language

Finite-State Phonology: Proceedings of the 5th Workshop of the ACL Special Interest Group in Computational Phonology (SIGPHON)

Home page of the workshop proceedings, with pointers to the individually archived papers. Includes front matter from the printed version of the proceedings.

Computation and Language

Mathematical Model of Word Length on the Basis of the Cebanov-Fucks Distribution with Uniform Parameter Distribution

The data on 13 typologically different languages have been processed using a two-parameter word length model, based on 1-displaced uniform Poisson distribution. Statistical dependencies of the 2nd parameter on the 1st one are revealed for the German texts and genre of letters.

Computation and Language

Quantitative Neural Network Model of the Tip-of-the-Tongue Phenomenon Based on Synthesized Memory-Psycholinguistic-Metacognitive Approach

A new three-stage computer artificial neural network model of the tip-of-the-tongue phenomenon is proposed. Each word's node is build from some interconnected learned auto-associative two-layer neural networks each of which represents separate word's semantic, lexical, or phonological components. The model synthesizes memory, psycholinguistic, and metamemory approaches, bridges speech errors and naming chronometry research traditions, and can explain quantitatively many tip-of-the-tongue effects.

Computation and Language

Two-parameter Model of Word Length "Language - Genre"

A two-parameter model of word length measured by the number of syllables comprising it is proposed. The first parameter is dependent on language type, the second one - on text genre and reflects the degree of completion of synergetic processes of language optimization.

Computation and Language

Magical Number Seven Plus or Minus Two: Syntactic Structure Recognition in Japanese and English Sentences

George A. Miller said that human beings have only seven chunks in short-term memory, plus or minus two. We counted the number of bunsetsus (phrases) whose modifiees are undetermined in each step of an analysis of the dependency structure of Japanese sentences, and which therefore must be stored in short-term memory. The number was roughly less than nine, the upper bound of seven plus or minus two. We also obtained similar results with English sentences under the assumption that human beings recognize a series of words, such as a noun phrase (NP), as a unit. This indicates that if we assume that the human cognitive units in Japanese and English are bunsetsu and NP respectively, analysis will support Miller's $7 \pm 2$ theory.

Computation and Language

A Machine-Learning Approach to Estimating the Referential Properties of Japanese Noun Phrases

The referential properties of noun phrases in the Japanese language, which has no articles, are useful for article generation in Japanese-English machine translation and for anaphora resolution in Japanese noun phrases. They are generally classified as generic noun phrases, definite noun phrases, and indefinite noun phrases. In the previous work, referential properties were estimated by developing rules that used clue words. If two or more rules were in conflict with each other, the category having the maximum total score given by the rules was selected as the desired category. The score given by each rule was established by hand, so the manpower cost was high. In this work, we automatically adjusted these scores by using a machine-learning method and succeeded in reducing the amount of manpower needed to adjust these scores.

Computation and Language

Meaning Sort - Three examples: dictionary construction, tagged corpus construction, and information presentation system

It is often useful to sort words into an order that reflects relations among their meanings as obtained by using a thesaurus. In this paper, we introduce a method of arranging words semantically by using several types of `{\sf is-a}' thesauri and a multi-dimensional thesaurus. We also describe three major applications where a meaning sort is useful and show the effectiveness of a meaning sort. Since there is no doubt that a word list in meaning-order is easier to use than a word list in some random order, a meaning sort, which can easily produce a word list in meaning-order, must be useful and effective.

Computation and Language

CRL at Ntcir2

We have developed systems of two types for NTCIR2. One is an enhenced version of the system we developed for NTCIR1 and IREX. It submitted retrieval results for JJ and CC tasks. A variety of parameters were tried with the system. It used such characteristics of newspapers as locational information in the CC tasks. The system got good results for both of the tasks. The other system is a portable system which avoids free parameters as much as possible. The system submitted retrieval results for JJ, JE, EE, EJ, and CC tasks. The system automatically determined the number of top documents and the weight of the original query used in automatic-feedback retrieval. It also determined relevant terms quite robustly. For EJ and JE tasks, it used document expansion to augment the initial queries. It achieved good results, except on the CC tasks.

Computation and Language

A Decision Tree of Bigrams is an Accurate Predictor of Word Sense

This paper presents a corpus-based approach to word sense disambiguation where a decision tree assigns a sense to an ambiguous word based on the bigrams that occur nearby. This approach is evaluated using the sense-tagged corpora from the 1998 SENSEVAL word sense disambiguation exercise. It is more accurate than the average results reported for 30 of 36 words, and is more accurate than the best results for 19 of 36 words.

Computation and Language

Type Arithmetics: Computation based on the theory of types

The present paper shows meta-programming turn programming, which is rich enough to express arbitrary arithmetic computations. We demonstrate a type system that implements Peano arithmetics, slightly generalized to negative numbers. Certain types in this system denote numerals. Arithmetic operations on such types-numerals - addition, subtraction, and even division - are expressed as type reduction rules executed by a compiler. A remarkable trait is that division by zero becomes a type error - and reported as such by a compiler.

Computation and Language

Coaxing Confidences from an Old Friend: Probabilistic Classifications from Transformation Rule Lists

Transformation-based learning has been successfully employed to solve many natural language processing problems. It has many positive features, but one drawback is that it does not provide estimates of class membership probabilities. In this paper, we present a novel method for obtaining class membership probabilities from a transformation-based rule list classifier. Three experiments are presented which measure the modeling accuracy and cross-entropy of the probabilistic classifier on unseen data and the degree to which the output probabilities from the classifier can be used to estimate confidences in its classification decisions. The results of these experiments show that, for the task of text chunking, the estimates produced by this technique are more informative than those generated by a state-of-the-art decision tree.

Computation and Language

Microplanning with Communicative Intentions: The SPUD System

The process of microplanning encompasses a range of problems in Natural Language Generation (NLG), such as referring expression generation, lexical choice, and aggregation, problems in which a generator must bridge underlying domain-specific representations and general linguistic representations. In this paper, we describe a uniform approach to microplanning based on declarative representations of a generator's communicative intent. These representations describe the results of NLG: communicative intent associates the concrete linguistic structure planned by the generator with inferences that show how the meaning of that structure communicates needed information about some application domain in the current discourse context. Our approach, implemented in the SPUD (sentence planning using description) microplanner, uses the lexicalized tree-adjoining grammar formalism (LTAG) to connect structure to meaning and uses modal logic programming to connect meaning to context. At the same time, communicative intent representations provide a resource for the process of NLG. Using representations of communicative intent, a generator can augment the syntax, semantics and pragmatics of an incomplete sentence simultaneously, and can assess its progress on the various problems of microplanning incrementally. The declarative formulation of communicative intent translates into a well-defined methodology for designing grammatical and conceptual resources which the generator can use to achieve desired microplanning behavior in a specified domain.

Computation and Language

Correction of Errors in a Modality Corpus Used for Machine Translation by Using Machine-learning Method

We performed corpus correction on a modality corpus for machine translation by using such machine-learning methods as the maximum-entropy method. We thus constructed a high-quality modality corpus based on corpus correction. We compared several kinds of methods for corpus correction in our experiments and developed a good method for corpus correction.

Computation and Language

Man [and Woman] vs. Machine: A Case Study in Base Noun Phrase Learning

A great deal of work has been done demonstrating the ability of machine learning algorithms to automatically extract linguistic knowledge from annotated corpora. Very little work has gone into quantifying the difference in ability at this task between a person and a machine. This paper is a first step in that direction.

Computation and Language

Rule Writing or Annotation: Cost-efficient Resource Usage for Base Noun Phrase Chunking

This paper presents a comprehensive empirical comparison between two approaches for developing a base noun phrase chunker: human rule writing and active learning using interactive real-time human annotation. Several novel variations on active learning are investigated, and underlying cost models for cross-modal machine learning comparison are presented and explored. Results show that it is more efficient and more successful by several measures to train a system using active learning annotation rather than hand-crafted rule writing at a comparable level of human labor investment.

Computation and Language

A Complete WordNet1.5 to WordNet1.6 Mapping

We describe a robust approach for linking already existing lexical/semantic hierarchies. We use a constraint satisfaction algorithm (relaxation labelling) to select --among a set of candidates-- the node in a target taxonomy that bests matches each node in a source taxonomy. In this paper we present the complete mapping of the nominal, verbal, adjectival and adverbial parts of WordNet 1.5 onto WordNet 1.6.

Computation and Language

Joint and conditional estimation of tagging and parsing models

This paper compares two different ways of estimating statistical language models. Many statistical NLP tagging and parsing models are estimated by maximizing the (joint) likelihood of the fully-observed training data. However, since these applications only require the conditional probability distributions, these distributions can in principle be learnt by maximizing the conditional likelihood of the training data. Perhaps somewhat surprisingly, models estimated by maximizing the joint were superior to models estimated by maximizing the conditional, even though some of the latter models intuitively had access to ``more information''.

Computation and Language

Probabilistic top-down parsing and language modeling

This paper describes the functioning of a broad-coverage probabilistic top-down parser, and its application to the problem of language modeling for speech recognition. The paper first introduces key notions in language modeling and probabilistic parsing, and briefly reviews some previous approaches to using syntactic structure for language modeling. A lexicalized probabilistic top-down parser is then presented, which performs very well, in terms of both the accuracy of returned parses and the efficiency with which they are found, relative to the best broad-coverage statistical parsers. A new language model which utilizes probabilistic top-down parsing is then outlined, and empirical results show that it improves upon previous work in test corpus perplexity. Interpolation with a trigram model yields an exceptional improvement relative to the improvement observed by other models, demonstrating the degree to which the information captured by our parsing model is orthogonal to that captured by a trigram model. A small recognition experiment also demonstrates the utility of the model.

Computation and Language

Robust Probabilistic Predictive Syntactic Processing

This thesis presents a broad-coverage probabilistic top-down parser, and its application to the problem of language modeling for speech recognition. The parser builds fully connected derivations incrementally, in a single pass from left-to-right across the string. We argue that the parsing approach that we have adopted is well-motivated from a psycholinguistic perspective, as a model that captures probabilistic dependencies between lexical items, as part of the process of building connected syntactic structures. The basic parser and conditional probability models are presented, and empirical results are provided for its parsing accuracy on both newspaper text and spontaneous telephone conversations. Modifications to the probability model are presented that lead to improved performance. A new language model which uses the output of the parser is then defined. Perplexity and word error rate reduction are demonstrated over trigram models, even when the trigram is trained on significantly more data. Interpolation on a word-by-word basis with a trigram model yields additional improvements.

Computation and Language

Generating a 3D Simulation of a Car Accident from a Written Description in Natural Language: the CarSim System

This paper describes a prototype system to visualize and animate 3D scenes from car accident reports, written in French. The problem of generating such a 3D simulation can be divided into two subtasks: the linguistic analysis and the virtual scene generation. As a means of communication between these two modules, we first designed a template formalism to represent a written accident report. The CarSim system first processes written reports, gathers relevant information, and converts it into a formal description. Then, it creates the corresponding 3D scene and animates the vehicles.

Computation and Language

The OLAC Metadata Set and Controlled Vocabularies

As language data and associated technologies proliferate and as the language resources community rapidly expands, it has become difficult to locate and reuse existing resources. Are there any lexical resources for such-and-such a language? What tool can work with transcripts in this particular format? What is a good format to use for linguistic data of this type? Questions like these dominate many mailing lists, since web search engines are an unreliable way to find language resources. This paper describes a new digital infrastructure for language resource discovery, based on the Open Archives Initiative, and called OLAC -- the Open Language Archives Community. The OLAC Metadata Set and the associated controlled vocabularies facilitate consistent description and focussed searching. We report progress on the metadata set and controlled vocabularies, describing current issues and soliciting input from the language resources community.

Computation and Language

Historical Dynamics of Lexical System as Random Walk Process

It is offered to consider word meanings changes in diachrony as semicontinuous random walk with reflecting and swallowing screens. The basic characteristics of word life cycle are defined. Verification of the model has been realized on the data of Russian words distribution on various age periods.

Computation and Language

Integrating Prosodic and Lexical Cues for Automatic Topic Segmentation

We present a probabilistic model that uses both prosodic and lexical cues for the automatic segmentation of speech into topically coherent units. We propose two methods for combining lexical and prosodic information using hidden Markov models and decision trees. Lexical information is obtained from a speech recognizer, and prosodic features are extracted automatically from speech waveforms. We evaluate our approach on the Broadcast News corpus, using the DARPA-TDT evaluation metrics. Results show that the prosodic model alone is competitive with word-based segmentation methods. Furthermore, we achieve a significant reduction in error by combining the prosodic and word-based knowledge sources.

Computation and Language

Computational properties of environment-based disambiguation

The standard pipeline approach to semantic processing, in which sentences are morphologically and syntactically resolved to a single tree before they are interpreted, is a poor fit for applications such as natural language interfaces. This is because the environment information, in the form of the objects and events in the application's run-time environment, cannot be used to inform parsing decisions unless the input sentence is semantically analyzed, but this does not occur until after parsing in the single-tree semantic architecture. This paper describes the computational properties of an alternative architecture, in which semantic analysis is performed on all possible interpretations during parsing, in polynomial time.

Computation and Language

Organizing Encyclopedic Knowledge based on the Web and its Application to Question Answering

We propose a method to generate large-scale encyclopedic knowledge, which is valuable for much NLP research, based on the Web. We first search the Web for pages containing a term in question. Then we use linguistic patterns and HTML structures to extract text fragments describing the term. Finally, we organize extracted term descriptions based on word senses and domains. In addition, we apply an automatically generated encyclopedia to a question answering system targeting the Japanese Information-Technology Engineers Examination.

Computation and Language

File mapping Rule-based DBMS and Natural Language Processing

This paper describes the system of storage, extract and processing of information structured similarly to the natural language. For recursive inference the system uses the rules having the same representation, as the data. The environment of storage of information is provided with the File Mapping (SHM) mechanism of operating system. In the paper the main principles of construction of dynamic data structure and language for record of the inference rules are stated; the features of available implementation are considered and the description of the application realizing semantic information retrieval on the natural language is given.

Computation and Language

Iterative Residual Rescaling: An Analysis and Generalization of LSI

We consider the problem of creating document representations in which inter-document similarity measurements correspond to semantic similarity. We first present a novel subspace-based framework for formalizing this task. Using this framework, we derive a new analysis of Latent Semantic Indexing (LSI), showing a precise relationship between its performance and the uniformity of the underlying distribution of documents over topics. This analysis helps explain the improvements gained by Ando's (2000) Iterative Residual Rescaling (IRR) algorithm: IRR can compensate for distributional non-uniformity. A further benefit of our framework is that it provides a well-motivated, effective method for automatically determining the rescaling factor IRR depends on, leading to further improvements. A series of experiments over various settings and with several evaluation metrics validates our claims.

Computation and Language

Stacking classifiers for anti-spam filtering of e-mail

We evaluate empirically a scheme for combining classifiers, known as stacked generalization, in the context of anti-spam filtering, a novel cost-sensitive application of text categorization. Unsolicited commercial e-mail, or "spam", floods mailboxes, causing frustration, wasting bandwidth, and exposing minors to unsuitable content. Using a public corpus, we show that stacking can improve the efficiency of automatically induced anti-spam filters, and that such filters can be used in real-life applications.

Computation and Language

Using the Distribution of Performance for Studying Statistical NLP Systems and Corpora

Statistical NLP systems are frequently evaluated and compared on the basis of their performances on a single split of training and test data. Results obtained using a single split are, however, subject to sampling noise. In this paper we argue in favour of reporting a distribution of performance figures, obtained by resampling the training data, rather than a single number. The additional information from distributions can be used to make statistically quantified statements about differences across parameter settings, systems, and corpora.

Computation and Language

Modeling informational novelty in a conversational system with a hybrid statistical and grammar-based approach to natural language generation

We present a hybrid statistical and grammar-based system for surface natural language generation (NLG) that uses grammar rules, conditions on using those grammar rules, and corpus statistics to determine the word order. We also describe how this surface NLG module is implemented in a prototype conversational system, and how it attempts to model informational novelty by varying the word order. Using a combination of rules and statistical information, the conversational system expresses the novel information differently than the given information, based on the run-time dialog state. We also discuss our plans for evaluating the generation strategy.

Computation and Language

The Role of Conceptual Relations in Word Sense Disambiguation

We explore many ways of using conceptual distance measures in Word Sense Disambiguation, starting with the Agirre-Rigau conceptual density measure. We use a generalized form of this measure, introducing many (parameterized) refinements and performing an exhaustive evaluation of all meaningful combinations. We finally obtain a 42% improvement over the original algorithm, and show that measures of conceptual distance are not worse indicators for sense disambiguation than measures based on word-coocurrence (exemplified by the Lesk algorithm). Our results, however, reinforce the idea that only a combination of different sources of knowledge might eventually lead to accurate word sense disambiguation.

Computation and Language

Looking Under the Hood : Tools for Diagnosing your Question Answering Engine

In this paper we analyze two question answering tasks : the TREC-8 question answering task and a set of reading comprehension exams. First, we show that Q/A systems perform better when there are multiple answer opportunities per question. Next, we analyze common approaches to two subproblems: term overlap for answer sentence identification, and answer typing for short answer extraction. We present general tools for analyzing the strengths and limitations of techniques for these subproblems. Our results quantify the limitations of both term overlap and answer typing to distinguish between competing answer candidates.

Computation and Language

Three-Stage Quantitative Neural Network Model of the Tip-of-the-Tongue Phenomenon

A new three-stage computer artificial neural network model of the tip-of-the-tongue phenomenon is shortly described, and its stochastic nature was demonstrated. A way to calculate strength and appearance probability of tip-of-the-tongue states, neural network mechanism of feeling-of-knowing phenomenon are proposed. The model synthesizes memory, psycholinguistic, and metamemory approaches, bridges speech errors and naming chronometry research traditions. A model analysis of a tip-of-the-tongue case from Anton Chekhov's short story 'A Horsey Name' is performed. A new 'throw-up-one's-arms effect' is defined.

Computation and Language

Introduction to the CoNLL-2001 Shared Task: Clause Identification

We describe the CoNLL-2001 shared task: dividing text into clauses. We give background information on the data sets, present a general overview of the systems that have taken part in the shared task and briefly discuss their performance.

Computation and Language

Learning Computational Grammars

This paper reports on the "Learning Computational Grammars" (LCG) project, a postdoc network devoted to studying the application of machine learning techniques to grammars suitable for computational use. We were interested in a more systematic survey to understand the relevance of many factors to the success of learning, esp. the availability of annotated data, the kind of dependencies in the data, and the availability of knowledge bases (grammars). We focused on syntax, esp. noun phrase (NP) syntax.

Computation and Language

Combining a self-organising map with memory-based learning

Memory-based learning (MBL) has enjoyed considerable success in corpus-based natural language processing (NLP) tasks and is thus a reliable method of getting a high-level of performance when building corpus-based NLP systems. However there is a bottleneck in MBL whereby any novel testing item has to be compared against all the training items in memory base. For this reason there has been some interest in various forms of memory editing whereby some method of selecting a subset of the memory base is employed to reduce the number of comparisons. This paper investigates the use of a modified self-organising map (SOM) to select a subset of the memory items for comparison. This method involves reducing the number of comparisons to a value proportional to the square root of the number of training items. The method is tested on the identification of base noun-phrases in the Wall Street Journal corpus, using sections 15 to 18 for training and section 20 for testing.

Computation and Language

Applying Natural Language Generation to Indicative Summarization

The task of creating indicative summaries that help a searcher decide whether to read a particular document is a difficult task. This paper examines the indicative summarization task from a generation perspective, by first analyzing its required content via published guidelines and corpus analysis. We show how these summaries can be factored into a set of document features, and how an implemented content planner uses the topicality document feature to create indicative multidocument query-based summaries.

Computation and Language

Multidimensional Transformation-Based Learning

This paper presents a novel method that allows a machine learning algorithm following the transformation-based learning paradigm \cite{brill95:tagging} to be applied to multiple classification tasks by training jointly and simultaneously on all fields. The motivation for constructing such a system stems from the observation that many tasks in natural language processing are naturally composed of multiple subtasks which need to be resolved simultaneously; also tasks usually learned in isolation can possibly benefit from being learned in a joint framework, as the signals for the extra tasks usually constitute inductive bias. The proposed algorithm is evaluated in two experiments: in one, the system is used to jointly predict the part-of-speech and text chunks/baseNP chunks of an English corpus; and in the second it is used to learn the joint prediction of word segment boundaries and part-of-speech tagging for Chinese. The results show that the simultaneous learning of multiple tasks does achieve an improvement in each task upon training the same tasks sequentially. The part-of-speech tagging result of 96.63% is state-of-the-art for individual systems on the particular train/test split.

Computation and Language

A Bit of Progress in Language Modeling

In the past several years, a number of different language modeling improvements over simple trigram models have been found, including caching, higher-order n-grams, skipping, interpolated Kneser-Ney smoothing, and clustering. We present explorations of variations on, or of the limits of, each of these techniques, including showing that sentence mixture models may have more potential. While all of these techniques have been studied separately, they have rarely been studied in combination. We find some significant interactions, especially with smoothing and clustering techniques. We compare a combination of all techniques together to a Katz smoothed trigram model with no count cutoffs. We achieve perplexity reductions between 38% and 50% (1 bit of entropy), depending on training data size, as well as a word error rate reduction of 8.9%. Our perplexity reductions are perhaps the highest reported compared to a fair baseline. This is the extended version of the paper; it contains additional details and proofs, and is designed to be a good introduction to the state of the art in language modeling.

Computation and Language

Classes for Fast Maximum Entropy Training

Maximum entropy models are considered by many to be one of the most promising avenues of language modeling research. Unfortunately, long training times make maximum entropy research difficult. We present a novel speedup technique: we change the form of the model to use classes. Our speedup works by creating two maximum entropy models, the first of which predicts the class of each word, and the second of which predicts the word itself. This factoring of the model leads to fewer non-zero indicator functions, and faster normalization, achieving speedups of up to a factor of 35 over one of the best previous techniques. It also results in typically slightly lower perplexities. The same trick can be used to speed training of other machine learning techniques, e.g. neural networks, applied to any problem with a large number of outputs, such as language modeling.

Computation and Language

Portability of Syntactic Structure for Language Modeling

The paper presents a study on the portability of statistical syntactic knowledge in the framework of the structured language model (SLM). We investigate the impact of porting SLM statistics from the Wall Street Journal (WSJ) to the Air Travel Information System (ATIS) domain. We compare this approach to applying the Microsoft rule-based parser (NLPwin) for the ATIS data and to using a small amount of data manually parsed at UPenn for gathering the intial SLM statistics. Surprisingly, despite the fact that it performs modestly in perplexity (PPL), the model initialized on WSJ parses outperforms the other initialization methods based on in-domain annotated data, achieving a significant 0.4% absolute and 7% relative reduction in word error rate (WER) over a baseline system whose word error rate is 5.8%; the improvement measured relative to the minimum WER achievable on the N-best lists we worked with is 12%.

Computation and Language

Information Extraction Using the Structured Language Model

The paper presents a data-driven approach to information extraction (viewed as template filling) using the structured language model (SLM) as a statistical parser. The task of template filling is cast as constrained parsing using the SLM. The model is automatically trained from a set of sentences annotated with frame/slot labels and spans. Training proceeds in stages: first a constrained syntactic parser is trained such that the parses on training data meet the specified semantic spans, then the non-terminal labels are enriched to contain semantic information and finally a constrained syntactic+semantic parser is trained on the parse trees resulting from the previous stage. Despite the small amount of training data used, the model is shown to outperform the slot level accuracy of a simple semantic grammar authored manually for the MiPad --- personal information management --- task.

Computation and Language

Anaphora and Discourse Structure

We argue in this paper that many common adverbial phrases generally taken to signal a discourse relation between syntactically connected units within discourse structure, instead work anaphorically to contribute relational meaning, with only indirect dependence on discourse structure. This allows a simpler discourse structure to provide scaffolding for compositional semantics, and reveals multiple ways in which the relational meaning conveyed by adverbial connectives can interact with that associated with discourse structure. We conclude by sketching out a lexicalised grammar for discourse that facilitates discourse interpretation as a product of compositional rules, anaphor resolution and inference.

Computation and Language

Conceptual Analysis of Lexical Taxonomies: The Case of WordNet Top-Level

In this paper we propose an analysis and an upgrade of WordNet's top-level synset taxonomy. We briefly review WordNet and identify its main semantic limitations. Some principles from a forthcoming OntoClean methodology are applied to the ontological analysis of WordNet. A revised top-level taxonomy is proposed, which is meant to be more conceptually rigorous, cognitively transparent, and efficiently exploitable in several applications.

Computation and Language

Boosting Trees for Anti-Spam Email Filtering

This paper describes a set of comparative experiments for the problem of automatically filtering unwanted electronic mail messages. Several variants of the AdaBoost algorithm with confidence-rated predictions [Schapire & Singer, 99] have been applied, which differ in the complexity of the base learners considered. Two main conclusions can be drawn from our experiments: a) The boosting-based methods clearly outperform the baseline learning algorithms (Naive Bayes and Induction of Decision Trees) on the PU1 corpus, achieving very high levels of the F1 measure; b) Increasing the complexity of the base learners allows to obtain better ``high-precision'' classifiers, which is a very important issue when misclassification costs are considered.

Computation and Language

Modelling Semantic Association and Conceptual Inheritance for Semantic Analysis

Allowing users to interact through language borders is an interesting challenge for information technology. For the purpose of a computer assisted language learning system, we have chosen icons for representing meaning on the input interface, since icons do not depend on a particular language. However, a key limitation of this type of communication is the expression of articulated ideas instead of isolated concepts. We propose a method to interpret sequences of icons as complex messages by reconstructing the relations between concepts, so as to build conceptual graphs able to represent meaning and to be used for natural language sentence generation. This method is based on an electronic dictionary containing semantic information.

Computation and Language

Integrating Multiple Knowledge Sources for Robust Semantic Parsing

This work explores a new robust approach for Semantic Parsing of unrestricted texts. Our approach considers Semantic Parsing as a Consistent Labelling Problem (CLP), allowing the integration of several knowledge types (syntactic and semantic) obtained from different sources (linguistic and statistic). The current implementation obtains 95% accuracy in model identification and 72% in case-role filling.

Computation and Language

Learning class-to-class selectional preferences

Selectional preference learning methods have usually focused on word-to-class relations, e.g., a verb selects as its subject a given nominal class. This papers extends previous statistical models to class-to-class preferences, and presents a model that learns selectional preferences for classes of verbs. The motivation is twofold: different senses of a verb may have different preferences, and some classes of verbs can share preferences. The model is tested on a word sense disambiguation task which uses subject-verb and object-verb relationships extracted from a small sense-disambiguated corpus.

Computation and Language

Knowledge Sources for Word Sense Disambiguation

Two kinds of systems have been defined during the long history of WSD: principled systems that define which knowledge types are useful for WSD, and robust systems that use the information sources at hand, such as, dictionaries, light-weight ontologies or hand-tagged corpora. This paper tries to systematize the relation between desired knowledge types and actual information sources. We also compare the results for a wide range of algorithms that have been evaluated on a common test setting in our research group. We hope that this analysis will help change the shift from systems based on information sources to systems based on knowledge sources. This study might also shed some light on semi-automatic acquisition of desired knowledge types from existing resources.

Computation and Language

Enriching WordNet concepts with topic signatures

This paper explores the possibility of enriching the content of existing ontologies. The overall goal is to overcome the lack of topical links among concepts in WordNet. Each concept is to be associated to a topic signature, i.e., a set of related words with associated weights. The signatures can be automatically constructed from the WWW or from sense-tagged corpora. Both approaches are compared and evaluated on a word sense disambiguation task. The results show that it is possible to construct clean signatures from the WWW using some filtering techniques.

Computation and Language

Testing for Mathematical Lineation in Jim Crace's "Quarantine" and T. S. Eliot's "Four Quartets"

The mathematical distinction between prose and verse may be detected in writings that are not apparently lineated, for example in T. S. Eliot's "Burnt Norton", and Jim Crace's "Quarantine". In this paper we offer comments on appropriate statistical methods for such work, and also on the nature of formal innovation in these two texts. Additional remarks are made on the roots of lineation as a metrical form, and on the prose-verse continuum.

Computation and Language

The Open Language Archives Community and Asian Language Resources

The Open Language Archives Community (OLAC) is a new project to build a worldwide system of federated language archives based on the Open Archives Initiative and the Dublin Core Metadata Initiative. This paper aims to disseminate the OLAC vision to the language resources community in Asia, and to show language technologists and linguists how they can document their tools and data in such a way that others can easily discover them. We describe OLAC and the OLAC Metadata Set, then discuss two key issues in the Asian context: language classification and multilingual resource classification.

Computation and Language

Richer Syntactic Dependencies for Structured Language Modeling

The paper investigates the use of richer syntactic dependencies in the structured language model (SLM). We present two simple methods of enriching the dependencies in the syntactic parse trees used for intializing the SLM. We evaluate the impact of both methods on the perplexity (PPL) and word-error-rate(WER, N-best rescoring) performance of the SLM. We show that the new model achieves an improvement in PPL and WER over the baseline results reported using the SLM on the UPenn Treebank and Wall Street Journal (WSJ) corpora, respectively.

Computation and Language

Part-of-Speech Tagging with Two Sequential Transducers

We present a method of constructing and using a cascade consisting of a left- and a right-sequential finite-state transducer (FST), T1 and T2, for part-of-speech (POS) disambiguation. Compared to an HMM, this FST cascade has the advantage of significantly higher processing speed, but at the cost of slightly lower accuracy. Applications such as Information Retrieval, where the speed can be more important than accuracy, could benefit from this approach. In the process of tagging, we first assign every word a unique ambiguity class c_i that can be looked up in a lexicon encoded by a sequential FST. Every c_i is denoted by a single symbol, e.g. [ADJ_NOUN], although it represents a set of alternative tags that a given word can occur with. The sequence of the c_i of all words of one sentence is the input to our FST cascade. It is mapped by T1, from left to right, to a sequence of reduced ambiguity classes r_i. Every r_i is denoted by a single symbol, although it represents a set of alternative tags. Intuitively, T1 eliminates the less likely tags from c_i, thus creating r_i. Finally, T2 maps the sequence of r_i, from right to left, to a sequence of single POS tags t_i. Intuitively, T2 selects the most likely t_i from every r_i. The probabilities of all t_i, r_i, and c_i are used only at compile time, not at run time. They do not (directly) occur in the FSTs, but are "implicitly contained" in their structure.

Computation and Language

What is the minimal set of fragments that achieves maximal parse accuracy?

We aim at finding the minimal set of fragments which achieves maximal parse accuracy in Data Oriented Parsing. Experiments with the Penn Wall Street Journal treebank show that counts of almost arbitrary fragments within parse trees are important, leading to improved parse accuracy over previous models tested on this treebank (a precision of 90.8% and a recall of 90.6%). We isolate some dependency relations which previous models neglect but which contribute to higher parse accuracy.

Computation and Language

Combining semantic and syntactic structure for language modeling

Structured language models for speech recognition have been shown to remedy the weaknesses of n-gram models. All current structured language models are, however, limited in that they do not take into account dependencies between non-headwords. We show that non-headword dependencies contribute to significantly improved word error rate, and that a data-oriented parsing model trained on semantically and syntactically annotated data can exploit these dependencies. This paper also contains the first DOP model trained by means of a maximum likelihood reestimation procedure, which solves some of the theoretical shortcomings of previous DOP models.

Computation and Language

Generating Multilingual Personalized Descriptions of Museum Exhibits - The M-PIRO Project

This paper provides an overall presentation of the M-PIRO project. M-PIRO is developing technology that will allow museums to generate automatically textual or spoken descriptions of exhibits for collections available over the Web or in virtual reality environments. The descriptions are generated in several languages from information in a language-independent database and small fragments of text, and they can be tailored according to the backgrounds of the users, their ages, and their previous interaction with the system. An authoring tool allows museum curators to update the system's database and to control the language and content of the resulting descriptions. Although the project is still in progress, a Web-based demonstrator that supports English, Greek and Italian is already available, and it is used throughout the paper to highlight the capabilities of the emerging technology.

Computation and Language

A procedure for unsupervised lexicon learning

We describe an incremental unsupervised procedure to learn words from transcribed continuous speech. The algorithm is based on a conservative and traditional statistical model, and results of empirical tests show that it is competitive with other algorithms that have been proposed recently for this task.

Computation and Language

A Statistical Model for Word Discovery in Transcribed Speech

A statistical model for segmentation and word discovery in continuous speech is presented. An incremental unsupervised learning algorithm to infer word boundaries based on this model is described. Results of empirical tests showing that the algorithm is competitive with other models that have been used for similar tasks are also presented.

Computation and Language

Using a Support-Vector Machine for Japanese-to-English Translation of Tense, Aspect, and Modality

This paper describes experiments carried out using a variety of machine-learning methods, including the k-nearest neighborhood method that was used in a previous study, for the translation of tense, aspect, and modality. It was found that the support-vector machine method was the most precise of all the methods tested.

Computation and Language

Part of Speech Tagging in Thai Language Using Support Vector Machine

The elastic-input neuro tagger and hybrid tagger, combined with a neural network and Brill's error-driven learning, have already been proposed for the purpose of constructing a practical tagger using as little training data as possible. When a small Thai corpus is used for training, these taggers have tagging accuracies of 94.4% and 95.5% (accounting only for the ambiguous words in terms of the part of speech), respectively. In this study, in order to construct more accurate taggers we developed new tagging methods using three machine learning methods: the decision-list, maximum entropy, and support vector machine methods. We then performed tagging experiments by using these methods. Our results showed that the support vector machine method has the best precision (96.1%), and that it is capable of improving the accuracy of tagging in the Thai language. Finally, we theoretically examined all these methods and discussed how the improvements were achived.

Computation and Language

Universal Model for Paraphrasing -- Using Transformation Based on a Defined Criteria --

This paper describes a universal model for paraphrasing that transforms according to defined criteria. We showed that by using different criteria we could construct different kinds of paraphrasing systems including one for answering questions, one for compressing sentences, one for polishing up, and one for transforming written language to spoken language.

Computation and Language

A Straightforward Approach to Morphological Analysis and Synthesis

In this paper we present a lexicon-based approach to the problem of morphological processing. Full-form words, lemmas and grammatical tags are interconnected in a DAWG. Thus, the process of analysis/synthesis is reduced to a search in the graph, which is very fast and can be performed even if several pieces of information are missing from the input. The contents of the DAWG are updated using an on-line incremental process. The proposed approach is language independent and it does not utilize any morphophonetic rules or any other special linguistic information.

Computation and Language

Fast Context-Free Grammar Parsing Requires Fast Boolean Matrix Multiplication

In 1975, Valiant showed that Boolean matrix multiplication can be used for parsing context-free grammars (CFGs), yielding the asympotically fastest (although not practical) CFG parsing algorithm known. We prove a dual result: any CFG parser with time complexity $O(g n^{3 - \epsilson})$, where $g$ is the size of the grammar and $n$ is the length of the input string, can be efficiently converted into an algorithm to multiply $m \times m$ Boolean matrices in time $O(m^{3 - \epsilon/3})$. Given that practical, substantially sub-cubic Boolean matrix multiplication algorithms have been quite difficult to find, we thus explain why there has been little progress in developing practical, substantially sub-cubic general CFG parsers. In proving this result, we also develop a formalization of the notion of parsing.

Computation and Language

Using Tree Automata and Regular Expressions to Manipulate Hierarchically Structured Data

Information, stored or transmitted in digital form, is often structured. Individual data records are usually represented as hierarchies of their elements. Together, records form larger structures. Information processing applications have to take account of this structuring, which assigns different semantics to different data elements or records. Big variety of structural schemata in use today often requires much flexibility from applications--for example, to process information coming from different sources. To ensure application interoperability, translators are needed that can convert one structure into another. This paper puts forward a formal data model aimed at supporting hierarchical data processing in a simple and flexible way. The model is based on and extends results of two classical theories, studying finite string and tree automata. The concept of finite automata and regular languages is applied to the case of arbitrarily structured tree-like hierarchical data records, represented as "structured strings." These automata are compared with classical string and tree automata; the model is shown to be a superset of the classical models. Regular grammars and expressions over structured strings are introduced. Regular expression matching and substitution has been widely used for efficient unstructured text processing; the model described here brings the power of this proven technique to applications that deal with information trees. A simple generic alternative is offered to replace today's specialised ad-hoc approaches. The model unifies structural and content transformations, providing applications with a single data type. An example scenario of how to build applications based on this theory is discussed. Further research directions are outlined.

Computation and Language

Blind Normalization of Speech From Different Channels and Speakers

This paper describes representations of time-dependent signals that are invariant under any invertible time-independent transformation of the signal time series. Such a representation is created by rescaling the signal in a non-linear dynamic manner that is determined by recently encountered signal levels. This technique may make it possible to normalize signals that are related by channel-dependent and speaker-dependent transformations, without having to characterize the form of the signal transformations, which remain unknown. The technique is illustrated by applying it to the time-dependent spectra of speech that has been filtered to simulate the effects of different channels. The experimental results show that the rescaled speech representations are largely normalized (i.e., channel-independent), despite the channel-dependence of the raw (unrescaled) speech.

Computation and Language

Models and Tools for Collaborative Annotation

The Annotation Graph Toolkit (AGTK) is a collection of software which facilitates development of linguistic annotation tools. AGTK provides a database interface which allows applications to use a database server for persistent storage. This paper discusses various modes of collaborative annotation and how they can be supported with tools built using AGTK and its database interface. We describe the relational database schema and API, and describe a version of the TableTrans tool which supports collaborative annotation. The remainder of the paper discusses a high-level query language for annotation graphs, along with optimizations, in support of expressive and efficient access to the annotations held on a large central server. The paper demonstrates that it is straightforward to support a variety of different levels of collaborative annotation with existing AGTK-based tools, with a minimum of additional programming effort.

Computation and Language

Creating Annotation Tools with the Annotation Graph Toolkit

The Annotation Graph Toolkit is a collection of software supporting the development of annotation tools based on the annotation graph model. The toolkit includes application programming interfaces for manipulating annotation graph data and for importing data from other formats. There are interfaces for the scripting languages Tcl and Python, a database interface, specialized graphical user interfaces for a variety of annotation tasks, and several sample applications. This paper describes all the toolkit components for the benefit of would-be application developers.

Computation and Language

TableTrans, MultiTrans, InterTrans and TreeTrans: Diverse Tools Built on the Annotation Graph Toolkit

Four diverse tools built on the Annotation Graph Toolkit are described. Each tool associates linguistic codes and structures with time-series data. All are based on the same software library and tool architecture. TableTrans is for observational coding, using a spreadsheet whose rows are aligned to a signal. MultiTrans is for transcribing multi-party communicative interactions recorded using multi-channel signals. InterTrans is for creating interlinear text aligned to audio. TreeTrans is for creating and manipulating syntactic trees. This work demonstrates that the development of diverse tools and re-use of software components is greatly facilitated by a common high-level application programming interface for representing the data and managing input/output, together with a common architecture for managing the interaction of multiple components.

Computation and Language

An Integrated Framework for Treebanks and Multilayer Annotations

Treebank formats and associated software tools are proliferating rapidly, with little consideration for interoperability. We survey a wide variety of treebank structures and operations, and show how they can be mapped onto the annotation graph model, and leading to an integrated framework encompassing tree and non-tree annotations alike. This development opens up new possibilities for managing and exploiting multilayer annotations.

Computation and Language

The tip-of-the-tongue phenomenon: Irrelevant neural network localization or disruption of its interneuron links ?

On the base of recently proposed three-stage quantitative neural network model of the tip-of-the-tongue (TOT) phenomenon a possibility to occur of TOT states coursed by neural network interneuron links' disruption has been studied. Using a numerical example it was found that TOTs coursed by interneron links' disruption are in (1.5 + - 0.3)x1000 times less probable then those coursed by irrelevant (incomplete) neural network localization. It was shown that delayed TOT states' etiology cannot be related to neural network interneuron links' disruption.

Computation and Language

Seven Dimensions of Portability for Language Documentation and Description

The process of documenting and describing the world's languages is undergoing radical transformation with the rapid uptake of new digital technologies for capture, storage, annotation and dissemination. However, uncritical adoption of new tools and technologies is leading to resources that are difficult to reuse and which are less portable than the conventional printed resources they replace. We begin by reviewing current uses of software tools and digital technologies for language documentation and description. This sheds light on how digital language documentation and description are created and managed, leading to an analysis of seven portability problems under the following headings: content, format, discovery, access, citation, preservation and rights. After characterizing each problem we provide a series of value statements, and this provides the framework for a broad range of best practice recommendations.

Computation and Language

Annotation Graphs and Servers and Multi-Modal Resources: Infrastructure for Interdisciplinary Education, Research and Development

Annotation graphs and annotation servers offer infrastructure to support the analysis of human language resources in the form of time-series data such as text, audio and video. This paper outlines areas of common need among empirical linguists and computational linguists. After reviewing examples of data and tools used or under development for each of several areas, it proposes a common framework for future tool development, data annotation and resource sharing based upon annotation graphs and servers.

Computation and Language

Computational Phonology

Phonology, as it is practiced, is deeply computational. Phonological analysis is data-intensive and the resulting models are nothing other than specialized data structures and algorithms. In the past, phonological computation - managing data and developing analyses - was done manually with pencil and paper. Increasingly, with the proliferation of affordable computers, IPA fonts and drawing software, phonologists are seeking to move their computation work online. Computational Phonology provides the theoretical and technological framework for this migration, building on methodologies and tools from computational linguistics. This piece consists of an apology for computational phonology, a history, and an overview of current research.

Computation and Language

Phonology

Phonology is the systematic study of the sounds used in language, their internal structure, and their composition into syllables, words and phrases. Computational phonology is the application of formal and computational techniques to the representation and processing of phonological information. This chapter will present the fundamentals of descriptive phonology along with a brief overview of computational phonology.

Computation and Language

Querying Databases of Annotated Speech

Annotated speech corpora are databases consisting of signal data along with time-aligned symbolic `transcriptions'. Such databases are typically multidimensional, heterogeneous and dynamic. These properties present a number of tough challenges for representation and query. The temporal nature of the data adds an additional layer of complexity. This paper presents and harmonises two independent efforts to model annotated speech databases, one at Macquarie University and one at the University of Pennsylvania. Various query languages are described, along with illustrative applications to a variety of analytical problems. The research reported here forms a part of several ongoing projects to develop platform-independent open-source tools for creating, browsing, searching, querying and transforming linguistic databases, and to disseminate large linguistic databases over the internet.

Computation and Language

Integrating selectional preferences in WordNet

Selectional preference learning methods have usually focused on word-to-class relations, e.g., a verb selects as its subject a given nominal class. This paper extends previous statistical models to class-to-class preferences, and presents a model that learns selectional preferences for classes of verbs, together with an algorithm to integrate the learned preferences in WordNet. The theoretical motivation is twofold: different senses of a verb may have different preferences, and classes of verbs may share preferences. On the practical side, class-to-class selectional preferences can be learned from untagged corpora (the same as word-to-class), they provide selectional preferences for less frequent word senses via inheritance, and more important, they allow for easy integration in WordNet. The model is trained on subject-verb and object-verb relationships extracted from a small corpus disambiguated with WordNet senses. Examples are provided illustrating that the theoretical motivations are well founded, and showing that the approach is feasible. Experimental results on a word sense disambiguation task are also provided.

Computation and Language

Decision Lists for English and Basque

In this paper we describe the systems we developed for the English (lexical and all-words) and Basque tasks. They were all supervised systems based on Yarowsky's Decision Lists. We used Semcor for training in the English all-words task. We defined different feature sets for each language. For Basque, in order to extract all the information from the text, we defined features that have not been used before in the literature, using a morphological analyzer. We also implemented systems that selected automatically good features and were able to obtain a prefixed precision (85%) at the cost of coverage. The systems that used all the features were identified as BCU-ehu-dlist-all and the systems that selected some features as BCU-ehu-dlist-best.

Computation and Language

The Basque task: did systems perform in the upperbound?

In this paper we describe the Senseval 2 Basque lexical-sample task. The task comprised 40 words (15 nouns, 15 verbs and 10 adjectives) selected from Euskal Hiztegia, the main Basque dictionary. Most examples were taken from the Egunkaria newspaper. The method used to hand-tag the examples produced low inter-tagger agreement (75%) before arbitration. The four competing systems attained results well above the most frequent baseline and the best system scored 75% precision at 100% coverage. The paper includes an analysis of the tagging procedure used, as well as the performance of the competing systems. In particular, we argue that inter-tagger agreement is not a real upperbound for the Basque WSD task.

Computation and Language

Memory-Based Shallow Parsing

We present memory-based learning approaches to shallow parsing and apply these to five tasks: base noun phrase identification, arbitrary base phrase recognition, clause detection, noun phrase parsing and full parsing. We use feature selection techniques and system combination methods for improving the performance of the memory-based learner. Our approach is evaluated on standard data sets and the results are compared with that of other systems. This reveals that our approach works well for base phrase identification while its application towards recognizing embedded structures leaves some room for improvement.

Computation and Language

Unsupervised discovery of morphologically related words based on orthographic and semantic similarity

We present an algorithm that takes an unannotated corpus as its input, and returns a ranked list of probable morphologically related pairs as its output. The algorithm tries to discover morphologically related pairs by looking for pairs that are both orthographically and semantically similar, where orthographic similarity is measured in terms of minimum edit distance, and semantic similarity is measured in terms of mutual information. The procedure does not rely on a morpheme concatenation model, nor on distributional properties of word substrings (such as affix frequency). Experiments with German and English input give encouraging results, both in terms of precision (proportion of good pairs found at various cutoff points of the ranked list), and in terms of a qualitative analysis of the types of morphological patterns discovered by the algorithm.

Computation and Language

Mostly-Unsupervised Statistical Segmentation of Japanese Kanji Sequences

Given the lack of word delimiters in written Japanese, word segmentation is generally considered a crucial first step in processing Japanese texts. Typical Japanese segmentation algorithms rely either on a lexicon and syntactic analysis or on pre-segmented data; but these are labor-intensive, and the lexico-syntactic techniques are vulnerable to the unknown word problem. In contrast, we introduce a novel, more robust statistical method utilizing unsegmented training data. Despite its simplicity, the algorithm yields performance on long kanji sequences comparable to and sometimes surpassing that of state-of-the-art morphological analyzers over a variety of error metrics. The algorithm also outperforms another mostly-unsupervised statistical algorithm previously proposed for Chinese. Additionally, we present a two-level annotation scheme for Japanese to incorporate multiple segmentation granularities, and introduce two novel evaluation metrics, both based on the notion of a compatible bracket, that can account for multiple granularities simultaneously.

Computation and Language

Ellogon: A New Text Engineering Platform

This paper presents Ellogon, a multi-lingual, cross-platform, general-purpose text engineering environment. Ellogon was designed in order to aid both researchers in natural language processing, as well as companies that produce language engineering systems for the end-user. Ellogon provides a powerful TIPSTER-based infrastructure for managing, storing and exchanging textual data, embedding and managing text processing components as well as visualising textual data and their associated linguistic information. Among its key features are full Unicode support, an extensive multi-lingual graphical user interface, its modular architecture and the reduced hardware requirements.

Computation and Language

Monads for natural language semantics

Accounts of semantic phenomena often involve extending types of meanings and revising composition rules at the same time. The concept of monads allows many such accounts -- for intensionality, variable binding, quantification and focus -- to be stated uniformly and compositionally.

Computation and Language

A variable-free dynamic semantics

I propose a variable-free treatment of dynamic semantics. By "dynamic semantics" I mean analyses of donkey sentences ("Every farmer who owns a donkey beats it") and other binding and anaphora phenomena in natural language where meanings of constituents are updates to information states, for instance as proposed by Groenendijk and Stokhof. By "variable-free" I mean denotational semantics in which functional combinators replace variable indices and assignment functions, for instance as advocated by Jacobson. The new theory presented here achieves a compositional treatment of dynamic anaphora that does not involve assignment functions, and separates the combinatorics of variable-free semantics from the particular linguistic phenomena it treats. Integrating variable-free semantics and dynamic semantics gives rise to interactions that make new empirical predictions, for example "donkey weak crossover" effects.

Computation and Language

NLTK: The Natural Language Toolkit

NLTK, the Natural Language Toolkit, is a suite of open source program modules, tutorials and problem sets, providing ready-to-use computational linguistics courseware. NLTK covers symbolic and statistical natural language processing, and is interfaced to annotated corpora. Students augment and replace existing components, learn structured programming by example, and manipulate sophisticated models from the outset.

Computation and Language

Unsupervised Discovery of Morphemes

We present two methods for unsupervised segmentation of words into morpheme-like units. The model utilized is especially suited for languages with a rich morphology, such as Finnish. The first method is based on the Minimum Description Length (MDL) principle and works online. In the second method, Maximum Likelihood (ML) optimization is used. The quality of the segmentations is measured using an evaluation method that compares the segmentations produced to an existing morphological analysis. Experiments on both Finnish and English corpora show that the presented methods perform well compared to a current state-of-the-art system.

Computation and Language

Bootstrapping Lexical Choice via Multiple-Sequence Alignment

An important component of any generation system is the mapping dictionary, a lexicon of elementary semantic expressions and corresponding natural language realizations. Typically, labor-intensive knowledge-based methods are used to construct the dictionary. We instead propose to acquire it automatically via a novel multiple-pass algorithm employing multiple-sequence alignment, a technique commonly used in bioinformatics. Crucially, our method leverages latent information contained in multi-parallel corpora -- datasets that supply several verbalizations of the corresponding semantics rather than just one. We used our techniques to generate natural language versions of computer-generated mathematical proofs, with good results on both a per-component and overall-output basis. For example, in evaluations involving a dozen human judges, our system produced output whose readability and faithfulness to the semantic input rivaled that of a traditional generation system.

Computation and Language

Evaluating the Effectiveness of Ensembles of Decision Trees in Disambiguating Senseval Lexical Samples

This paper presents an evaluation of an ensemble--based system that participated in the English and Spanish lexical sample tasks of Senseval-2. The system combines decision trees of unigrams, bigrams, and co--occurrences into a single classifier. The analysis is extended to include the Senseval-1 data.

Computation and Language