Terry Copeck

Systematic construction of a versatile case system

Case systems abound in natural language processing. Almost any attempt to recognize and uniformly represent relationships within a clause – a unit at the centre of any linguistic system that goes beyond word level statistics – must be based on semantic roles drawn from a small, closed set. The set of roles describing relationships between a verb and its arguments within a clause is a case system. What is required of such a case system? How does a natural language practitioner build a system that is complete and detailed yet practical and natural? This paper chronicles the construction of a case system from its origin in English marker words to its successful application in the analysis of English text.

What is technical text

Abstract Beyond labeling it easier to process than other types, few researchers who use technical text in their work try to define what it is. This paper describes a study that investigates the character of texts typically considered technical. We identify 42 features of a text considered likely to correlate with its degree of technicality. These include both objectively verifiable measures like marked presence of interrogative or imperative sentences which are akin to the criteria used by Biber in Variation Across Speech and Writing , and subjective measures such as presence of hierarchical organization . All are less ambiguous than technicality, so our inventory may be suited to use in a procedure that classifies text as technical or non-technical. An inventory organizing and describing these lexical, syntactic, semantic and discourse features was used to rate nine varied sample texts. Analysis of 22 ratings of each text indicated that 31 features in the inventory were meaningful predictors of text technicality when considered independently. The inventory has been revised and a formula to compute technicality has been developed in the light of these findings.

Text Summarization as Controlled Search

We present a framework for text summarization based on the generate-and-test model. A large set of summaries is generated for all plausible values of six parameters that control a three-stage process that includes segmentation and keyphrase extraction, and a number of features that characterize the document. Quality is assessed by measuring the summaries against the abstract of the summarized document. The large number of summaries produced for our corpus dictates automated validation and fine-tuning of the summary generator. We use supervised machine learning to detect good and bad parameters. In particular, we identify parameters and ranges of their values within which the summary generator might be used with high reliability on documents for which no authors abstract exists.

Parsing and case analysis in TANKA

The TANKA project seeks to build a model of a technical domain by semi-automatically processing unedited English text that describes this domain. Each sentence is parsed and conceptual elements are extracted from the parse. Concepts are derived from the Case structure of a sentence, and added to a conceptual network that represents knowledge about the domain. The DIPETT parser has a particularly broad coverage of English syntax; its newest version can also process sentence fragments. The HAIKU subsystem is responsible for user-assisted semantic interpretation. It contains a Case Analyzer module that extracts phrases marking concepts from the parse and uses its past processing experience to derive the most likely Case realizations of each with almost no a priori semantic knowledge. The user must validate these selections. A key issue in our research is minimizing the number of interactions with the user by intelligently generating the alternatives offered.

A hybrid diagnostic-advisory system for small and medium-sized enterprises: A successful AI application

We describe a hybrid expert diagnosis-advisory system developed for small and medium enterprises. The Performance, Development, Growth (PDG) system is completely implemented and fully operational, and has been successfully used on real-world data from SMEs for several years. Although our system contains a great deal of the domain knowledge and expertise that is a hallmark of AI systems, it was not designed using the symbolic techniques traditionally used to implement such systems. We explain why this is so and discuss how the PDG system relates to expert systems, decision support systems, and general applications in AI. We also present an experimental evaluation of the system and identify developments currently under way and our plans for the future.

An Automatic Evaluation Framework for Improving a Configurable Text Summarizer

CALLISTO is a text summarization system that depends on machine learning techniques and is therefore sensitive to pre-established biases that may not be wholly appropriate. We set out to test whether other biases, modifying the space that CALLISTO explores, lead to improvements in the overall quality of the summaries produced. We present an automatic evaluation framework that relies on a summary quality measure proposed by Lin and Hovy. It appears to be the first evaluation of a text summarization system conducted automatically on a large corpus of news stories. We show the practicality of our methodology on a few experiments with the Machine Learning module of CALLISTO. We conclude that this framework gives reliable hints on the adequacy of a bias and could be useful in developing automatic text summarization systems that work with Machine Learning techniques.

Surface-marker-based dialog modelling: A progress report on the MAREDI project

Most information systems that deal with natural language texts do not tolerate much deviation from their idealized and simplified model of language. Spoken dialog is notoriously ungrammatical, however. Because the MAREDI project focuses in particular on the automatic analysis of scripted dialogs, we needed to develop a robust capacity to analyze transcribed spoken language. This paper summarizes the current state of our work. It presents the main elements of our approach, which is based on exploiting surface markers as the best route to the semantics of the conversation modelled. We highlight the foundations of our particular conversational model, and give an overview of the MAREDI system. We then discuss its three key modules, a connectionist network to recognise speech acts, a robust syntactic analyzer, and a semantic analyzer.