File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/concl/04/w04-1704_concl.xml
Size: 4,074 bytes
Last Modified: 2025-10-06 13:54:19
<?xml version="1.0" standalone="yes"?> <Paper uid="W04-1704"> <Title>Generated Narratives for Computer-aided Language Teaching</Title> <Section position="3" start_page="0" end_page="0" type="concl"> <SectionTitle> 2 Enhancements and Future Work </SectionTitle> <Paragraph position="0"> VINCI's use of semantic input by means of functional expressions is designed to allow it to function either as an autonomous narrative generation system (cf. (Callaway and Lester, 2001a), (Bringsjord and Ferrucci, 2000) for examples) or as a story authoring environment (cf. (Umaschi and Cassell, 1997)) in which a language teacher may select or construct high-level utterance speci cations, or alternatively, a learner may play with the order of a set of semantic speci cations, or even add new characters with their own traits, examining in each case the texts produced. Two kinds of enhancements can be used to improve output.</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.1 Encyclopedic Enrichment </SectionTitle> <Paragraph position="0"> In examples shown above, descriptions are based on simple static attributes (beauty, morality, etc.). In fact, VINCI's compound attribute mechanism also allows for the expression of actions by characters. Thus, the attribute kill.monsters in the lexical entry for Prince Braveheart might cause exists(hero) to lead to: Nearby there lived a prince called Braveheart, who was renowned for killing monsters. This mechanism is also applicable to the expression of a character's thoughts and attitudes, and past background information, both narrative desiderata (Bringsjord and Ferrucci, 2000), (Robin, 1993) as well as the generation more or less complex versions of the same text (cf. (Chali, 1998)). Work is underway on mechanisms for the dynamic temporal tagging of attributes, as a story develops. For example, a learner given $50 and instructed to purchase groceries in a textual supermarket would have his or her remaining money reduced by each purchase he or she describes.</Paragraph> <Paragraph position="1"> It should also be noted that the micro-world de ned by means of Vinci may be ctional, as in the cases above, or based on real people and events. For example, we have performed experiments based on a database of French authors, their works, and their biographical details such as date of birth, death, etc.</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.2 Narrative Enrichment </SectionTitle> <Paragraph position="0"> Appropriate use of anaphoric pronous and aggregation of sentences both have a signi cant e ect on perceptions of text quality (Callaway and Lester, 2001b). In a number of systems, both processes occur after sentences have been realized, at the level of revision, which often requires that utterances be reformulated. We propose to perform comparable operations at the level of semantic expressions. Suppose two functions: exists(X), which generates There was an X, and describe(X) which generates X was brave and handsome. The fact that both expressions share a common argument allows for their replacement by another, say exdesc(X), which aggregates the two functions to produce There was a brave and handsome X. Similarly, in the case of anaphoric relations, shared arguments allow for replacement of full names by pronouns. We are currently researching this.</Paragraph> <Paragraph position="1"> Finally, taking account of work by (Karamanis and Manurung, 2002) which shows that sharing of at least one argument characterizes a high percentage of successive sentences in a text, it is possible to use the sequence of arguments to order a sequence of semantic expressions. Perhaps more interestingly, it may be that one of the criteria of a new paragraph is a break in the chain of shared arguments from one semantic expression to the next. The paragraph breaks in the English and French texts above, while human-constructed, respect this constraint.</Paragraph> </Section> </Section> class="xml-element"></Paper>