Automatic detection of deception in child-produced speech using syntactic complexity features

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 944–953, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Automatic detection of deception in child-produced speech using syntactic complexity features Maria Yancheva Division of Engineering Science, University of Toronto Toronto Ontario Canada maria.yancheva@utoronto.ca Frank Rudzicz Toronto Rehabilitation Institute; and Department of Computer Science, University of Toronto Toronto Ontario Canada frank@cs.toronto.edu Abstract It is important that the testimony of chil- dren be admissible in court, especially given allegations of abuse. Unfortunately, children can be misled by interrogators or might offer false information, with dire consequences. In this work, we evalu- ate various parameterizations of five clas- sifiers (including support vector machines, neural networks, and random forests) in deciphering truth from lies given tran- scripts of interviews with 198 victims of abuse between the ages of 4 and 7. These evaluations are performed using a novel set of syntactic features, including mea- sures of complexity. Our results show that sentence length, the mean number of clauses per utterance, and the Stajner- Mitkov measure of complexity are highly informative syntactic features, that classi- fication accuracy varies greatly by the age of the speaker, and that accuracy up to 91.7% can be achieved by support vec- tor machines given a sufficient amount of data. 1 Introduction The challenge of disambiguating between truth and deception is critical in determining the ad- missibility of court testimony. Unfortunately, the testimony of maltreated children is often not ad- mitted in court due to concerns about truthfulness since children can be instructed to deny transgres- sions or misled to elicit false accusations (Lyon and Dorado, 2008). However, the child is often the only witness of the transgression (Undeutsch, 2008); automatically determining truthfulness in such situations is therefore a paramount goal so that justice may be served effectively. 2 Related Work Research in the detection of deception in adult speech has included analyses of verbal and non- verbal cues such as behavioral changes, facial ex- pression, speech dysfluencies, and cognitive com- plexity (DePaulo et al., 2003). Despite statistically significant predictors of deception such as shorter talking time, fewer semantic details, and less co- herent statements, DePaulo et al. (2003) found that the median effect size is very small. Deception without special motivation (e.g., everyday ‘white lies’) exhibited almost no discernible cues of de- ception. However, analysis of moderating factors showed that cues were significantly more numer- ous and salient when lies were about transgres- sions. Literature on deception in children is relatively limited. In one study, Lewis et al. (1989) studied 3-year-olds and measured behavioral cues, such as facial expression and nervous body movement, be- fore and after the elicitation of a lie. Verbal re- sponses consisted of yes/no answers. Results sug- gested that 3-year-old children are capable of de- ception, and that non-verbal behaviors during de- ception include increases in ‘positive’ behaviors (e.g., smiling). However, verbal cues of deception were not analyzed. Crucially, Lewis et al. (1989) showed that humans are no more accurate …

Content truncated to 500 words