Behavior-Based Language Generation for Believable Agents,

Abstract

This work was supported in part by Fujitsu Laboratories and Mitsubishi Electric Research Laboratories. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of any other parties.

Towards personality-based user adaptation: psychologically informed stylistic language generation

“…Previous research on the generation of linguistic variation includes both rule-based and statistical approaches, as well as hybrid methods that combine rule-based linguistic knowledge with statistical methods (Langkilde and Knight, 1998;Langkilde-Geary, 2002). This includes work on variation using parameters based on pragmatic effects (Fleischman and Hovy, 2002;Hovy, 1988), stylistic factors such as formality, sentence length, and syntactic structure (Belz, 2005;Bouayad-Agha et al, 2000;DiMarco and Hirst, 1993;Green and DiMarco, 1996;Paiva and Evans, 2005;Paris and Scott, 1994;Power et al, 2003;, emotion (Cahn, 1990), lexical choice (Inkpen and Hirst, 2004), user expertise or confidence (DiMarco and Hirst, 1993;Forbes-Riley and Litman, 2007;Forbes-Riley et al, 2008;Porayska-Pomsta and Mellish, 2004;Wang et al, 2005), a theory of linguistic politeness (Brown and Levinson, 1987;Gupta et al, 2007Gupta et al, , 2008Porayska-Pomsta and Mellish, 2004;Walker et al, 1997;Wang et al, 2005;Wilkie et al, 2005), theories of personality (André et al, 2000;Ball and Breese, 1998;Isard et al, 2006;Loyall and Bates, 1995), and individual differences and preferences for both style and content (Belz, 2008;Lin, 2006;Reiter and Sripada, 2002;Stent et al, 2004;Walker et al, 2007). While there are strong relations between these different notions of style, and the types of linguistic variation associated with personality factors, here we limit our detailed discussion of prior work to personality generation.…”

“…In contrast to our approach, the work on ECAs has typically modeled the generation task using templates, which have been labeled as expressing a particular personality, rather than by manipulating parameters within modules of the NLG pipeline. Loyall and Bates (1995) is one of the first papers to suggest the use of personality models for language generation in ECAs. They present a model where personality factors are integrated with emotions, intentions and desires, and use template-based generation indexed by personality variables.…”

Individual Differences

“…This includes work on variation using parameters based on pragmatic effects (Hovy 1988;Fleischman and Hovy 2002), stylistic factors such as formality, sentence length, and syntactic structure (Power et al 2003;Bouayad-Agha et al 2000;DiMarco and Hirst 1993;Green and DiMarco 1996;Paiva and Evans 2005;Belz 2005;Walker et al 2002;Paris and Scott 1994), emotion (Cahn 1990), lexical choice (Inkpen and Hirst 2004), user expertise or confidence (Porayska-Pomsta and Mellish 2004;Wang et al 2005;DiMarco and Hirst 1993;Forbes-Riley and Litman 2007;Forbes-Riley et al 2008), a theory of linguistic politeness (Brown and Levinson 1987;Walker et al 1997;Gupta et al 2007Gupta et al , 2008Wilkie et al 2005;Porayska-Pomsta and Mellish 2004;Wang et al 2005), theories of personality (Ball and Breese 1998;André et al 2000;Loyall and Bates 1995;Isard et al 2006), and individual differences and preferences for both style and content Walker et al 2007;Reiter and Sripada 2002;Lin 2006;Belz 2008). While there are strong relations between these different notions of style, and the types of linguistic variation associated with personality factors, here we limit our detailed discussion of prior work to personality generation.…”

“…Loyall and Bates (1995) is one of the first papers to suggest the use of personality models for language generation in ECAs. They present a model where personality factors are integrated with emotions, intentions and desires, and use template-based generation indexed by personality variables.…”

Expressive AI: A Hybrid Art and Science Practice

“…Subjective Avatars provides an example of expressive AI's focus on specificity. The characters in Subjective Avatars were built using Hap, a language designed to facilitate the crafting of specific, unique characters [20]. This is in contrast to both ALife and topdown approaches to character which attempt to define universal character frameworks in which specific characters are "tuned-in" by adjusting parameters in the model [21].…”