Ted Caldwell's scientific contributions

Publications (3)

Article
Full-text available
In this paper, we present EXEMPLARS, an object-oriented, rule-base d framework designed to support practical, dynamic text generation, emphasizing its novel features compared to existing hybrid systems that mix template-style and more sophisticated techniques. These features-.include an extensible classification~based text planning mechanism, a def...
Article
Full-text available
In this paper, we present E XEMPLARS, an object-oriented, rule-based framework designed to support practical, dynamic text generation, emphasizing its novel features compared to existing hybrid systems that mix template-style and more sophisticated techniques. These features include an extensible classification-based text planning mechanism, a defi...
Article
Full-text available
In this paper, we present E XEMPLARS, a practical, object-oriented framework for generating XML- annotated texts, and compare its capabilities with XSL, an emerging W3C standard for transforming XML documents which can be used for simple text generation applications. We illustrate the framework's unique features—object-oriented specialization, abil...

Citations

... For the Language Generation in DEG we used the Exemplars framework, a template based NLG system. [WC98] ...
... The Exemplars framework requires Java objects as input. In principle, any XML data can serve as input after being read into Java application objects, or if they are accessed in Java via XML APIs. Figure 13 shows some example XML input data, taken from White and Caldwell [1999]. White and Caldwell [1999], p. 2. Figure 14 shows two of the exemplars that can be used to convert the data from Figure 13 to an HTML page containing a section entitled "People", which gives a short description of each person in the input data together with a list of his or her meetings. ...
... The idea of combining text generation with past experience has been explored before. White and Caldwell (1998) used a set of hand crafted examples in order to generate responses through templates. More recently, Song et al. (2016) also explored a hybrid system with an information retrieval component, but their system is very different: It uses a complex ranking system at a high computational cost, requires a post-reranking component to exploit previous dialogue turns (about half of which are copied over as predictions), and they only evaluate their system in a chit-chat set-up, reporting only BLEU scores. ...