Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm UniversitySomewhat surprisingly, although the emotion/vocalcue interface in speech has been investigatedextensively, there is no widely accepteddefinition or taxonomy of emotion. Apparently,there is no standard psychologicaltheory of emotion that could decide the issueonce and for all: the number of basic (and secondary)emotions is still a moot point. Nevertheless,certain emotions are often considered torepresent “basic emotions”: at least fear, anger,happiness, sadness, surprise and disgust areamong the basic emotions (Cornelius, 1996).Research on the vocal expression of emotionhas been largely based on scripted noninteractionalmaterial; a typical scenario involvesa group of actors simulating emotionswhile reading out an emotionally neutral sentenceor text. There are now also databases containingnatural emotional speech, but these corpora(necessarily) tend to containblended/uncertain and mixed emotions ratherthan “pure” basic emotions (see Scherer, 2003,for a review).Emotions in speech: clinical investigationsThe vocal cues of affect have been investigatedalso in clinical settings, i.e. with a view tocharting the acoustic/prosodic features of certainemotional states (or states of emotionaldisorders or mental disorders). For example, itis generally assumed that clinical depressionmanifests itself in speech in a way which issimilar to sadness (a general, “non-morbid”emotional state). Thus, a decreased average f0,a decreased f0 minimum, and a flattened f0range are common, along with decreased intensityand a lower rate of articulation (Scherer,2000). Voiced high frequency spectral energygenerally decreases. Intonationally, sadness/depressionmay typically be associatedwith downward directed f0 contours.Psychiatric interest in prosody has recentlyshed light on the interrelationship betweenschizophrenia and (deficient or aberrant) prosody.Several investigators have argued that schizophrenicsrecognize emotion in speech considerablyworse than members of the normalpopulation. Productively, the situation appearsquite similar, i.e. schizophrenics cannot conveyaffect through vocal cues as consistently andeffectively as normal subjects (Murphy & Cutting,1990). In the investigation by Murphy &Cutting (1990), a group of schizophrenics wereto express basic emotions (neutral, angry, surprise,sad) while reading out a number of sentences.The raters (normal subjects) had significantdifficulty recognizing the simulated emotions(as opposed to portrayals of the sameemotions by a group representing the normalpopulation).In general, it has been found out that speechand communication problems typically precedethe onset of psychosis; dysarthria and dysprosodyappear to be common. Affective flatteningis indeed a diagnostic component of psychosis(along with, for example, grossly disorganizedspeech), and anomalous prosody (e.g. a lack ofany observable speech melody) may thus be anessential part of the dysprosody evident in psychosis(Golfarb & Bekker, 2009). Moreover,schizophrenics’ speech seems to contain morepauses and hesitation features than normalspeech (Covington et al., 2005). Interestingly,although depressed persons’ speech also typicallycontains a decreased amount of speech perthe speech situation, the distribution of pausesappears to be different from schizophrenicspeech: schizophrenics typically pause in“wrong” (syntactically/semantically) unmotivatedplaces, while the pausing is more logicaland grammatical in depressed speech. Schizophrenicspeech thus seems to reflect the erraticsemantic structure of what is said (Clemmer,1980).It would be fascinating to think that certainprosodic features (or their absence) could be ofhelp for the general practitioner when diagnosingmental disorders. Needless to say, such featurescould never be the only diagnostic toolbut, in the best scenario, they would providesome assistive means for distinguishing betweensome alternative diagnostic possibilities.Emotions in speech: an interactionalclinical approachIn the following sections we outline a preliminaryapproach to investigating emotionalspeech and interaction within a clinical context.What follows is, at this stage, a proposal ratherthan a definitive research agenda.Prosodic analysis: 4-Tone EVoOur first proposal concerns the prosodic annotationprocedure for a speech material producedin a (clinical) setting inducing emotionally ladenspeech. As is well known, ToBI labeling(Beckman & Ayers, 1993) is commonly used inthe prosodic transcription of (British and Amer-177