Henry S. Thompson

Using genetic algorithms to create meaningful poetic text

This article presents a series of experiments in automatically generating poetic texts. We confined our attention to the generation of texts which are syntactically well-formed, meet certain pre-specified patterns of metre and broadly convey some given meaning. Such aspects can be formally defined, thus avoiding the complications of imagery and interpretation that are central to assessing more free forms of verse. Our implemented system, McGONAGALL, applies the genetic algorithm to construct such texts. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress. The conducted experiments broadly indicated that relatively meaningful text could be produced if the constraints on metre were relaxed, and precise metric text was possible with loose semantic constraints, but it was difficult to produce text which was both semantically coherent and of high quality metrically.

The HCRC Map Task corpus: natural dialogue for speech recognition

The HCRC Map Task corpus has been collected and transcribed in Glasgow and Edinburgh, and recently published on CD-ROM. This effort was made possible by funding from the British Economic and Social Research Council.The corpus is composed of 128 two-person conversations in both high-quality digital audio and orthographic transcriptions, amounting to 18 hours and 150,000 words respectively.The experimental design is quite detailed and complex, allowing a number of different phonemic, syntactico-semantic and pragmatic contrasts to be explored in a controlled way.The corpus is a uniquely valuable resource for speech recognition research in particular, as we move from developing systems intended for controlled use by familiar users to systems intended for less constrained circumstances and naive or occasional users. Examples supporting this claim are given, including preliminary evidence of the phonetic consequences of second mention and the impact of different styles of referent negotiation on communicative efficacy.

Using SGML as a Basis for Data-Intensive NLP

This paper describes the LT NSL system (McKelvie et al, 1996), an architecture for writing corpus processing tools. This system is then compared with two other systems which address similar issues, the GATE system (Cunningham et al, 1995) and the IMS Corpus Workbench (Christ, 1994). In particular we address the advantages and disadvantages of an SGML approach compared with a non-SGML database approach.

Using SGML as a Basis for Data-Intensive Natural Language Processing

This paper describes the LT NSL system (McKelvie et al., 1996), an architecture for writing corpus processing tools. This system is then compared with two other systems which address similar issues, the GATE system (Cunningham et al., 1995) and the IMS Corpus Workbench (Christ, 1994). In particular we address the advantages and disadvantages of an SGML approach compared with a non-sgml database approach.

Structural non-correspondence in translation

Kaplan et al (1989) present an approach to machine translation based on co-description. In this paper we show that the notation is not as natural and expressive as it appears. We first show that the most natural analysis proposed in Kaplan et al (1989) cannot in fact cover the range of data for the important translational phenomenon in question. This contribution extends the work reported on in Sadler et al (1989) and Sadler et al (1990). We then go on to discuss alternatives which depart from or extend the formalism proposed in Kaplan et al (1989) in various respects, pointing out some directions for further research. The strategies discussed have been implemented.

COMPOSE-REDUCE PARSING

Two new parsing algorithms for context-free phrase structure grammars are presented which perform a bounded amount of processing per word per analysis path, independently of sentence length. They are thus capable of parsing in real-time in a parallel implementation which forks processors in response to non-deterministic choice points.

Best-first enumeration of paths through a lattice—an active chart parsing solution

Abstract Active chart parsing offers a clean and flexible way of implementing dynamic programming to find the best path through a-cylic directed lattices. This paper describes how this comes about and what the general form of a chart parsing realization of dynamic programming takes. Advantage is taken of the resulting flexibility to produce a system which not only finds the best path, but enumerates paths in order. Finally, we exemplify the process for finding paths through a word lattice given bi-class probabilities, and give a few results from some experiments.

Chart Parsing for Loosely Coupled Parallel Systems

Of the parallel systems currently available, far and away the most common are loosely coupled collections of conventional processors, and this is likely to remain true for some time. By loosely coupled I mean that the processors do not share memory, so that some form of stream or message-passing protocol is required for processor-processor communication. It follows that in most cases the programmer must make explicit appeal to communication primitives in the construction of software which exploits the available parallelism. Even in shared-memory systems, the absence of parallel constructs from available programming languages may mean that appeal to a similar communication model may be necessary, at least in the short term.

Parallel intersection and serial composition of finite state transducers

We describe a linguistically expressive and easy to implement parallel semantics for quasi-deterministic finite state transducers(FSTS) used as acceptors. Algorithms are given for determining acceptance of pairs of phoneme strings given a parallel suite of such transducers and for constructing the equivalent single transducer by parallel intersection. An algorithm for constructing the serial composition of a sequence of such transducers is also given. This algorithm can produce generally nondeterministic FSTS and an algorithm is presented for eliminating the unacceptable nondeterminism. Finally, the work is discussed in the context of other work on finite state transducers.

Automatic evaluation of computer generated text: a progress report on the TextEval project

We present results of experiments designed to assess the usefulness of a new technique for the evaluation of translation quality, comparing human rankings with automatic measures. The basis of our approach is the use of a standard set and the adoption of a statistical view of translation quality. This approach has the ability to provide evaluations which avoid dependence on any particular theory of translation, which are therefore potentially more objective than previous techniques. The work presented here was supported by the Science and Engineering and the Social and Economic Research Councils of Great Britain, and would not have been possible without the gracious assistance of Ian Mason of Heriot Watt University, Edinburgh.