We describe a computational model for planning phrases like “more than a quarter” and “25.9 per cent” which describe proportions at different levels of precision. The model lays out the key choices in planning a numerical description, using formal definitions of mathematical form (e.g., the distinction between fractions and percentages) and roundness adapted from earlier studies. The task is modeled as a constraint satisfaction problem, with solutions subsequently ranked by preferences (e.g., for roundness). Detailed constraints are based on a corpus of numerical expressions collected in the NumGen project, 1 1 NumGen: Generating intelligent descriptions of numerical quantities for people with different levels of numeracy ( http://mcs.open.ac.uk/sw6629/numgen ). NumGen was funded by the Economic and Social Research Council under Grant Ref. RES-000-22-2760. and evaluated through empirical studies in which subjects were asked to produce (or complete) numerical expressions in specified contexts.

Abstract We describe SkillSum, a Natural Language Generation (NLG) system that generates a personalised feedback report for someone who has just completed a screening assessment of their basic literacy and numeracy skills. Because many SkillSum users have limited literacy, the generated reports must be easily comprehended by people with limited reading skills; this is the most novel aspect of SkillSum, and the focus of this paper. We used two approaches to maximise readability. First, for determining content and structure (document planning), we did not explicitly model readability, but rather followed a pragmatic approach of repeatedly revising content and structure following pilot experiments and interviews with domain experts. Second, for choosing linguistic expressions (microplanning), we attempted to formulate explicitly the choices that enhanced readability, using a constraints approach and preference rules; our constraints were based on corpus analysis and our preference rules were based on psycholinguistic findings. Evaluation of the SkillSum system was twofold: it compared the usefulness of NLG technology to that of canned text output, and it assessed the effectiveness of the readability model. Results showed that NLG was more effective than canned text at enhancing users' knowledge of their skills, and also suggested that the empirical ‘revise based on experiments and interviews’ approach made a substantial contribution to readability as well as our explicit psycholinguistically inspired models of readability choices.

The problem of how to estimate variance parameters in client models from scarce data is addressed in the context of text-dependent, HMM-based, automatic speaker verification. Variance flooring and variance scaling are investigated as two alternative estimation techniques and are used with or without variance tying on the state level to reduce the number of parameters to estimate. The best results are achieved with no tying and a variance flooring method where the floor to a variance vector in a client model is proportional to the corresponding variance vector in a gender-dependent, multispeaker, non-client model. Further, variance tying reduces storage requirements considerably without much loss in recognition accuracy. It is also confirmed from a previous study that re-using non-client variances has comparable performance to variance flooring and is much simpler. Comparisons are made on three large telephone quality speech corpora.

This paper describes a concept-to-speech system for generating spoken descriptions of routes between places within Macquarie University Computing Department. The Natural Language Generation (NLG) component of the system generates a textual route description marked with intonational information. The discourse structure of the route description is related closely to the knowledge representation of the route. The NLG component includes a pitch accenting algorithm which places appropriate pitch accents on elements of the utterance requiring particular emphasis or stress. Our pitch accenting algorithm uses a domain knowledge base and a discourse history. From these it determines whether information selected to form the content of the utterance is shared mutual domain knowledge, given information, or new information. It can then assign an appropriate pitch accent to one word in each prosodic phrase. The text-to-speech component then determines the appropriate syllable to be accented in the w...