Trainable Generation of Big-Five Personality Styles through Data-Driven Parameter Estimation

Proceedings of ACL-08: HLT, pages 165–173, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Trainable Generation of Big-Five Personality Styles through Data-driven Parameter Estimation Franc¸ois Mairesse Cambridge University Engineering Department Trumpington Street Cambridge, CB2 1PZ, United Kingdom farm2@eng.cam.ac.uk Marilyn Walker Department of Computer Science University of Sheffield Sheffield, S1 4DP, United Kingdom lynwalker@gmail.com Abstract Previous work on statistical language gen- eration has primarily focused on grammat- icality and naturalness, scoring generation possibilities according to a language model or user feedback. More recent work has investigated data-driven techniques for con- trolling linguistic style without overgenera- tion, by reproducing variation dimensions ex- tracted from corpora. Another line of work has produced handcrafted rule-based systems to control specific stylistic dimensions, such as politeness and personality. This paper describes a novel approach that automati- cally learns to produce recognisable varia- tion along a meaningful stylistic dimension— personality—without the computational cost incurred by overgeneration techniques. We present the first evaluation of a data-driven generation method that projects multiple per- sonality traits simultaneously and on a contin- uous scale. We compare our performance to a rule-based generator in the same domain. 1 Introduction Over the last 20 years, statistical language models (SLMs) have been used successfully in many tasks in natural language processing, and the data avail- able for modeling has steadily grown (Lapata and Keller, 2005). Langkilde and Knight (1998) first applied SLMs to statistical natural language genera- tion (SNLG), showing that high quality paraphrases can be generated from an underspecified representa- tion of meaning, by first applying a very undercon- strained, rule-based overgeneration phase, whose outputs are then ranked by an SLM scoring phase. Since then, research in SNLG has explored a range of models for both dialogue and text generation. One line of work has primarily focused on gram- maticality and naturalness, scoring the overgener- ation phase with a SLM, and evaluating against a gold-standard corpus, using string or tree-match metrics (Langkilde-Geary, 2002; Bangalore and Rambow, 2000; Chambers and Allen, 2004; Belz, 2005; Isard et al., 2006). Another thread investigates SNLG scoring mod- els trained using higher-level linguistic features to replicate human judgments of utterance quality (Rambow et al., 2001; Nakatsu and White, 2006; Stent and Guo, 2005). The error of these scoring models approaches the gold-standard human rank- ing with a relatively small training set. A third SNLG approach eliminates the overgen- eration phase (Paiva and Evans, 2005). It applies factor analysis to a corpus exhibiting stylistic vari- ation, and then learns which generation parameters to manipulate to correlate with factor measurements. The generator was shown to reproduce intended fac- tor levels across several factors, thus modelling the stylistic variation as measured in the original corpus. Our goal is a generation technique that can tar- get multiple stylistic effects simultaneously and over a continuous scale, controlling stylistic di- mensions that are commonly understood and thus meaningful to users and application developers. Our intended applications are output utterances for intelligent training or intervention systems, video game characters, or virtual environment avatars. …

Content truncated to 500 words