Analysing deception in a psychopath ’ s speech : a quantitative approach

Psychopathy involves a series of specifi c cognitive, social and emotional features which make the psychopath different from the general population; the two most signifi cant characteristics are extreme selfi shness and deep emotional defi cit that is refl ected in apathy. Notably, psychopaths are skilled communicators who that use language to lie. As there has been little examination of the speech associated specifi cally with psychopaths, especially in the Spanish language, the present study aims to contrast different veracious excerpts to others which are deceptive. The text analysis is framed within forensic computational linguistics, and complemented with some information related to the stylometric profi le of the text. The investigation shows how the parameter mainly affected by the psychological condition of the psychopath subject is the distribution of grammatical persons; in addition, some further evidence includes the frequency of certainty adverbs and verbs related to cognitive processes. Key-words: Psychopathy; mental disorder; forensic linguistics; deception; computational analysis. http://dx.doi.org/10.1590/0102-445040702531513856 Ana Carolina Vilela-Ardenghi & Ana Raquel Motta


Introduction: Mental Diseases and Language
There has always been an interest in the study of mental diseases from different points of view.They have been normally studied from the clinical perspective.Nevertheless, some scholars have studied different psychological disorders by means of word analysis or linguistic patterns, most of them within the English speaking context.This type of studies are already found in the 50s such as the ones carried out by Lorenz and Cobb (1952;1953) which were mostly related to psychoneurotic and maniac patients.Maniac people's speech was also of interest for Andreasen and Pfohl (1976).These studies explored different syntactic and grammatical patterns revealed by the linguistic behaviour of mentally disordered people.Lindenfeld (1973) observed changes in people's syntactic patterns according to their affective state.Schizophrenia has also been explored from a linguistic approach.To this respect, Chaika (1974) identifi ed six linguistic aspects that were specifi c of schizophrenic speech.Thomas et al. (1987) focused on the syntactic structures used by schizophrenic patients while speaking.They could fi nd that those with negative symptoms used more simple syntactic constructions than those with positive symptoms, who, at the same time committed more syntactic and grammatical errors.
More recent examples use computational linguistics for their analysis.This is the case of Lott et al. (2002), who analyse the speech of three types of mental patients: schizophrenic, bipolar and depressive.Anderson et al. (2008) made a quantitative and qualitative analysis of the lexis used in narratives written by people who suffered from social phobia.The literary productions of some well-known writers have also been analysed given the fact that these writers suffered from some mental disease.Forgeard (2008) compared eminent writers diagnosed with unipolar disorder to those who were bipolar (Jane Austen, Charlotte Brönte, Henry James, Leon Tolstoy or Scott Fitzgerald).She found some differences between the two groups.Bipolar writers made more allusions to death, fewer references to other people rather than themselves and used fewer cognitive verbs in their speech.In a recent study Cantos (2014) focused his analysis on the controversial writer Poe.He observed how Poe's mental status was refl ected in his linguistic behavior, affecting both lexis and syntax.

Psychopathy
The current study explores psychopathy.This mental disorder has been conceived in different ways by different experts.We are going to follow Woodworth and Porter (2002) in their defi nition of this mental disease as a "Personality disorder characterized by a profound affective defi cit accompanied by a lack of respect for the rights of others and societal rules".
Psychopathy involves a series of specifi c neurobiological, social and emotional features that make the psychopath different from the general population.From a biological perspective, psychopaths' brain presents several structural and functional abnormalities such as grey matter reduction in frontal and temporal areas, as well as anomalies in the prefrontal cortex.
Ángela Almela, Gema Alcaraz-Mármol, Pascual Cantos In addition, the psychopath's behavior is characterized by specifi c features such as pathological lying, poor behavioral controls, failure to accept responsibility for own actions, grandiose estimation of the self, shallow affect and lack of remorse or guilt.These fi ve features can be considered to be the most relevant, although not the only ones.
As stated above, mental diseases such as Psychopathy can be explored from different viewpoints, among which Linguistics is found.However, limited research can be found in this regard.Scholars such as Cleckley (1976), Williamson (1993) and Brinkley et al. (1999) have focused on psychopath's discourse cohesion and coherence.In recent years, Hancock et al. have gone beyond by examining more specifi c cues such as lexis and morphology.
One of the main behavioural features of psychopaths is pathological lying.The procedures for the detection of deception have been divided into those based on physiological methods and those based on behavioural methods.Among the former, we can highlight the polygraph, brain activity analysis, thermal analysis or voice stress analysis.As for the latter, we can fi nd nonverbal or verbal assessment tools.The most interesting for us are the second ones, that is, those which are based on communication, namely, statement validity assessment, reality monitoring and linguistic analysis, which is the one that is followed here.

Deception in Language
The linguistic analysis aims to distinguish fabricated messages from truthful ones; it operates independently of the message meaning, and it uses sophisticated statistical text analysis tools.One of these software applications is LIWC which stands for Linguistic Inquiry and Word Count.LIWC counts and classifi es words into psychologically meaningful categories.2200 words and word stems are grouped into 72 broad categories which are relevant to psychological processes.Some of these categories are used in other deception detection methods such as Reality Monitoring.LIWC lexicon has shown correlation with human ratings of a large number of written texts, which suggests that it is a valid tool for analysis.Indeed, we fi nd comprehensive accounts of LIWC dimensions such as Pennebaker et al. (2001) and the Spanish equivalent Ramírez-Esparza et al. (2007).Signifi cantly enough, LIWC was fi rst tested by Newman et al. (2003) on a corpus of university students' deceptive and truthful written and spoken language purposely produced.Furthermore, in this process of data collection the participants are not biased towards the concealment of the lies, which, according to Bull et al. (2006), is highly frequent among professional liars.The cost of the lies being detected would not be high in this case, opposite to what happens in high-stakes situations.
One of the key issues in psycholinguistics is the refl ection of the emotional and cognitive frames of humans on the oral and written language they produce.Early approaches to psycholinguistic concerns involved almost exclusively qualitative philosophical analyses.More recent research in this fi eld provides empirical evidence on the relation between language and the state of mind of subjects, or even their mental health (Rosenberg & Tucker, 1978).In this regard, further studies by Pennebaker and his team have dealt with the therapeutic effect of verbally expressing emotional experiences and memories.LIWC was developed precisely for providing an effi cient method for studying these psycholinguistic concerns, and has been considerably improved since its fi rst version (Francis & Pennebaker, 1993).
Within the fi rst dimension, namely standard linguistic processes, most categories involve function words and grammatical information; thus, the selection of words is straightforward, as in the case of articles, which are made up of three words in English -a, an and the-and of nine words in Spanish -el, la, los, las, uno, un, una, unos and unas.On the other hand, the second and fourth dimensions are more subjective, especially those denoting emotional processes within the second dimension.These categories indeed required human judges to make the lexical selection.For all subjective categories, an initial list of word candidates was compiled from dictionaries and thesauri for all subjective categories.Similar to the fi rst dimension, the third dimension, relativity, comprises a category concerning time, which is quite clear-cut: past, present, and future tense verbs.Within the same dimension, this is also the case of the category space, in which spatial prepositions and adverbs have been included.
Finally, the fourth dimension involves word categories related to personal concerns intrinsic to the human condition.As explained below, this dimension has often been excluded in deception detection studies, on the basis that it is too content-dependent (Hancook et al., 2011;Newman et al., 2003).

Description of the Case Study
Turning now to our case study, our subject is a 27-year-old Spanish single woman, who, at the moment the data were collected, was studying a Master's degree in Translation and Interpreting.One morning, she alleged she had been raped the previous night, but she commented she could not remember almost anything.She suspected that she was drugged with Scopolamine, which is a substance that destroys free will and may cause amnesia.The case becomes highly well-known, mainly due to the use of this drug in Spain as a new way to commit this type of crime.The police asked her to write everything she can remember as an attempt to help the presumed victim to offer more details.At this point we have to comment that it is this written description what we use as the material for the analysis. 1  After the allegation was formalized, the subject starts having problems with her attorney.He observes a strange behavior in his client.The subject attacks him verbally.She even threatens with accusing him of sexual harassment.That is why she is evaluated by an expert and she is diagnosed with psychopathic behavior.It is confi rmed that the subject suffers a dissocial personality disorder that is characterized, among other things, as psychopathic behavior.In fact, it was demonstrated that the subject had lied and she had not been raped.
1. Due to the non-disclosure agreement signed with the expert who provided the data, the original text cannot be included as an appendix.

Research Aim
Having described the facts that lead to this study, we delve into the aim of the study.The present work goes beyond the analysis of a psychopath's speech: it focuses on the act of lying, by contrasting different veracious excerpts to others which are deceptive.In other words, it explores deception in a psychopath's speech through quantitative analysis by means of psycholinguistic categories in a descriptive fashion.This means that the results are not to be statistically projected because of the small size of the sample.

Data
As stated above, the material for the analysis was the statement that was written by the subject, and that was required by the Catalan Police (called Mossos d'Esquadra).This type of procedure is not usual in European Civil Law, but it was admitted as an expert proof.For the analysis, the text has been divided into two fi les: one comprising the deceptive excerpts in the statement, and another fi le with the rest of the text used as the control sample.

Results
Thus, the two text fragments: "true text" versus "deceptive text" have been the basis of our analysis, by means of contrasting various aspects and dimensions.The basic extension parameters of both texts are given below (Table 1): Although both texts diverge in extension, their STTR are very similar, and therefore the comparison becomes very consistent as the percentage of new types for every n tokens is virtually identical.Similarly, another contrastive parameter: mean word length (in characters) is also stable in both texts (Table 2 and Figure 1): Other, more subtle parameters (mean sentence length and fi gureuse), reveal the fi rst noticeable differences: true-text sentences are on average 2.5 words longer and a more prone to contain numbers: potential hints of using completely different sentence generation strategies when telling the truth or deceiving.
Focusing on the dimensions analysis (Table 3), we have discarded those with a relative frequency lower than 2 in the true text (Table 4).   2 show the most prominent dimension differences between the true text and deceptive text; negative percentage difference evidence a more solid usage among true statements, whereas positive values more prominent habit in deception.

Discussion and Final Remarks
It is worth noting that in the studies on statement veracity and validity, the psychological dimension is the most informative one.However, it has been observed that the data related to this dimension on some of the deceptive excerpts do not correspond to the prototypical parameters, probably because of the mental disorder of the subject.First, regarding cognitive processes, the linguistic elements categorized as 'insight' -e.g.think, consider-are most frequently found in truthful statements, and such is the case in the present analysis.However, there is a statistically signifi cant difference between the values of certainty words -e.g.always, never-, which are considerably more frequent in untruthful excerpts, which happens to be the norm, since this kind of words are normally used as a strategy for concealing lies (Newman et al., 2003;Almela, 2012).
As far as affective processes are concerned, positive emotions have been associated to truthful statements in Spanish (Almela, 2012), which is similar to the results in the English language (Mihalcea & Strapparava, 2009;Newman et al., 2003).On the contrary, negative emotions have traditionally been associated to deception in English, although in Spanish some of them are positively correlated with truthfulness, namely anxiety and sadness.Interestingly enough, in the psychopath's speech both groups of emotions are more frequent when she tells lies, probably in an attempt to empathize with her readers.The overall percentage of affection words in untruthful excerpts doubles the category in the truthful ones.Furthermore, words related to social processes are less frequent in untruthful excerpts as a global category.Since this category is typically associated to truthfulness (Newman et al., 2003;Zhou et al., 2004), it matches the normal parameter.
One of the defi ning cues to untruthfulness is the relative abundance of 2nd and 3rd person, whereas 1st person singular is used rather moderately, since the speaker prefers not to identify him/herself with the lies he/she is telling.However, this does not apply to subjects with a psychopathic personality; these are unable to feel any remorse or guilt, and have a marked egocentrism and a grandiose estimation of self.Linguistically, this is refl ected in a strong presence of the fi rst person singular along the whole statement, with a considerable increase in untruthful excerpts (more than 3 points).
The values of average sentence length are also indicative.Unlike the results from studies with spoken language corpora, in the written medium untruthful statements are characterized by a shorter sentence length, probably because of the speaker's fear to fall into contradiction (Almela, 2012;Zhou et al., 2004).This means in practice that there is a tendency to simplify sentences and to reduce the use of subordination and coordination.
Last but not least, it is worth noting that there is a decrease in the amount of words related to number in deceptive excerpts.This fi ts into the usual pattern of deceptive communication, since the speaker prefers not to add specifi city to their speech.

Table 1 -
Extension parameters of true text and deceptive text

Table 2 -
More extension parameters of true text and deceptive text

word length (in characters) Mean sentence length (in tokens) Use of numbers (total fi gure)
Figure 1 -Word selection regarding word length .

Table 3 -
Dimensions of true text and deceptive text

Table 4 and
Figure