We present a data-to-speech system called D2S, which can be used for the creation of datato -speech systems in different languages and domains. The most important characteristic of a data-to-speech system is that it combines language and speech generation: language generation is used to produce a natural language text expressing the system's input data, and speech generation is used to make this text audible. In D2S, this combination is exploited by using linguistic information available in the language generation module for the computation of prosody. This allows us to achieve a better prosodic output quality than can be achieved in a plain text-to-speech system. For language generation in D2S, the use of syntactically enriched templates is guided by knowledge of the discourse context, while for speech generation pre-recorded phrases are combined in a prosodically sophisticated manner. This combination of techniques makes it possible to create linguistically sound but efficient system...

We added a sentence planning component to an existing ITS that teaches students how to troubleshoot mechanical systems.

A number of generation systems have chosen hypertext as a target media and study ways of adapting its content and links to the presentation context and the particular user.

Abstract not found

Abstract not found

Although there is an increasing shift towards evaluating Natural Language Generation (NLG) systems, there are still many NLG-specific open issues that hinder effective comparative and quantitative evaluation in this field. The paper starts off by describing a task-based, i.e., black-box evaluation of a hypertext NLG system. Then we examine the problem of glass-box, i.e., module specific, evaluation in language generation, with focus on evaluating machine learning methods for text planning.

In this paper we describe content determination issues involved in the Atlas.txt project, which aims to automatically describe georeferenced information such as census data as text for the visually-impaired (VI). Texts communicating geo-referenced census information contain census data abstractions and their corresponding geographic references. Because visually impaired users find interpreting geographic references hard, we hypothesized that an introduction message about the underlying geography should help the users to interpret the geographic references easily. We performed user studies to design and evaluate the introduction message. An initial evaluation study with several sighted users and one partially sighted user showed that an introduction message is certainly preferred by most participants. Many of them used an introduction message themselves when they described maps textually. But the study also showed that the introduction message made no difference when the participants were asked to draw maps using the information in the textual descriptions. 1

Abstract. This paper describes the Mountain language generation system, which is designed as a domain-independent, machine translationbased approach to generate utterances for use within a spoken dialog system. We describe the method used to train the generation engine from a corpus of in-domain human responses, and show typical output of the Mountain system. The results of our initial evaluation suggest this approach can be a viable method of language generation. We also discuss potential applications where this approach is likely to be most useful, as well as planned improvements to the system and future research using Mountain. 1

One approach to verification and validation of language processing systems includes the verification of system resources. In general, the grammar is a key resource in such systems. In this paper we discuss verification of lexicalized tree adjoining grammars (LTAGs) (Joshi and Schabes, 1997) as one instance of a system resource, and as one phase of a larger verification effort.

One of our general goals is to answer the question of what is the "added value" of a Natural Language interaction for a learner that interacts with an ITS. To do so, we applied simple Natural Language Generation techniques to improve the feedback provided by intelligent tutoring systems built within the DIAG framework (Towne 1997a). We have evaluated the original version of the system and the enhanced one with a between subjects experiment. The results are mixed. Although differences rarely achieve statistical significance, there appears to be slight improvement on performance measures in the group interacting with the enhanced system; however, subjects using the original system have better recall of the actions they took.