It is often assumed as self-evidential that speakers express their emotion, evaluation, or attitude by their speech to achieve their purposes. In this paper I shall show that this common view, apart from its seeming plausibility, does not always capture successfully the very nature of speaker's behavior in everyday communication, and suggest an alternative view for understanding the correlation between speaker's speech and their emotion, evaluation, and attitudes.
Close observation on everyday Japanese conversation data, especially focused on disfluent phenomena such as fillers, stuttering, and pressed voice (a kind of creaky voice), by using native speaker introspection reveals that the so-called common view has three defects. The idea that the speaker uses various fillers, various ways of stuttering, various voice qualities to express his/her emotion, evaluation, and attitudes cannot explain the detail of these disfluent phenomena because (i) it does not accept unintended speech because of its teleological nature, (ii) it does not really touch speaker's psychology because it is based on outward perspective, and (iii) it regards language as a thing to express something rather than an expressing action itself.

View full abstract

The “-desu/-masu” form not only functions as a formal language marker, it also indicates emotions/attitudes and characterizes some roles in conversation. These attribute of “-desu/-masu” indicate “psychological distance between speakers and addressees”, the relationship between speakers and addressees in communication and its change. This paper defines the language forms that require addressees, such as “-desu/-masu”, as “copresence markers” that create copresent space independently of contexts. We model the structure of communication with the notion of a copresence marker and the degree of copresence determined by contexts. In copresence, “-desu/-masu” indicates psychological distance. In nonpresence, however “-desu/-masu” does not typically appear, but a quasi-copresence sort of virtual space can be created by a copresence marker. In this case, the function of “-desu/-masu” as a copresence marker is foregrounded and an addressee is elicited. This relationship between a speaker and an addressee results in intimate emotion. Intimate emotions/attitudes can be explained by the function of a copresence marker that changes “nonpresence” into “copresence”. Distance can be explained by the change of mental relationship between speakers and addressees with psychological distance operating in “copresence”.

View full abstract

We examined the effects of familiarity, modality, and task difficulty on the use of fillers when describing a figure. A total of 56 adults (aged 18-38) participated in an experiment designed to elicit examples of disfluency words, such as fillers, affective interjections, and speech discontinuities. They were asked to solve a problem in same sex pairs. One was instructed to describe the figure, and the other to identify the correct figure from a choice of six. This experiment was done in various conditions when there was variation in how familiar the participants were with the figure, i.e., variation in familiarity; when the pairs could and not see each other, i.e., variation in modality; and when the task was both easy and difficult, i.e., variation in difficulty. The results showed two things. First, the average rates of filler, affective interjections, and speech discontinuities differed in relation to situational differences. Second, the filler rates varied with the type of filler. These results are discussed in terms of the relationship between mental processing and mental markers.

View full abstract

The choice behaviour of alternatives can be expressed as a two-stage process: a stage to recognize the character of each alternative and a stage of decision making based on the character, i.e., ‘recognize’ and ‘choose’ stages. Many of the existing studies on reputation and opinion analysis focus on the information extraction of the ‘recognize’ stage, whereas this paper describes the entire choice behaviour including the ‘choose’ stage and tries to capture unknown elements. This paper reports the selection process throughout the ‘choose’ stage, which can be described as conforming to Elimination-By-Aspects (EBA). EBA performs the selection by eliminating according to whether the alternative has the feature (aspect) in question or not. This paper achieves this process with choice strategies that eliminate or retain the alternatives. Moreover, it is insufficient to simply handle the appeared information, hence, it is necessary to read between the lines. Finally, we can know the merits and demerits of the alternatives by analyzing the reason for the triggering of selection or elimination.

View full abstract

Some wonderful experiences are expressed in Japanese by the word “Kandoh” which could be translated into English as “emotional-affect”. The “emotional-affect” is defined by the dictionary as “making people have strong feeling in facing beautiful or wonderful things”. According to the public-opinion poll by Mitsubishi Research Institute, Inc. in 2003, the mass media is one of the major subjects evoking emotionalaffect. But there have been no studies that tried to define a mental state of emotionalaffect.
The purpose of our study is to describe the emotional-affect for evaluation of broadcast programs. First, we addressed a questionnaire about emotional-affect and picked up words which expressed the situation of emotional-affect from the answers. Furthermore, we calculated a distance of the word by similarity measures based on the subjective evaluation. The obtained results are 1) the emotional affect could be classified into a few main groups, and 2) individual groups were classified by some factors including the object and kind of emotion, not the emotion itself. These results suggest that emotional-affect is the general term of the conditions of mind such as affirmative impressions and uncontrollable minds due to a very strong passive and compulsory stimulus.

View full abstract

We measured affective meaning of words by SD method on three dimensions (Evaluation, Potency, Activity) as EPA scores. Affect Control Theory is a mathematical and social-psychological theoretical frame in order to use these EPA scores how combinations of words generate affective meaning of sentences. First we demonstrate the validity of our frame by using EPA score to distinguish homonyms in the process of Kana-Kanji translation system. Second our frame has an ability to distinguish cross-cultural affective difference of meanings in the process of the translation from Japanese to English and reverse as well.

View full abstract

The purpose of this research is to look for the lesson improvement method by learner's feeling evaluation and learning attitude analysis using the comment mail on a lesson wrote by the university student in a mobile phone. The sentences of these comment mails were classified into four categories, interest, motivation, knowledge and consideration as the standard of the feeling evaluation. Significant differences were not found in the morpheme level. However as a result of classification of the meanings of the contents, there are many motivation and consideration in high score students, and many interest and knowledge in low score students. Each one student was selected from high and low score students, and their sentences were compared, and it was shown that high score students remake the contents of the lesson by their own terms, but low score student makes a copy of content the teacher taught. In conclusion, for the lesson improvement, lesson should be remade to learner's motivation and consideration increase and learner's remaking the contents of the lesson increase.

View full abstract

In order to make synthetic speech rich in its expression, fundamental frequency contours were analyzed for the utterances of several emotional degrees with joy and sadness based on a model for the process of generation. Changes in controlling parameters of the model with regard to degrees of emotion were examined in terms of linguistic factors of the utterances. As a result, the baseline frequency increases as emotional degree increases, especially for sadness utterances. About the phrase commands, the rate of occurrence increases as emotional degree increases at the right branch boundary in the grammatical structure for both joy and sadness, while the rate of occurrence at the left branch boundary for sadness is almost constant for emotional degrees. The change of the amplitude of phrase commands is depended on the kind of position of grammatical structure. About the accent commands, timings of their onsets and offsets are almost constant for emotional degrees. They are depended on the accent types of prosodic words. The magnitude of the accent commands changes as emotional degree increases depending on the positions of prosodic words from the beginning of the utterance.

View full abstract

This report describes a study on an estimation method of degree of speaker's emotion by using acoustic and linguistic features expressed in their anger utterances during a natural dialogue. We set two types of pseudo dialogues, the human-computer and the human-human, to induce anger utterances from 10 speakers. To make an emotional speech corpus with degree of emotion, a 5-scale subjective evaluation was conducted to grade each utterance on its emotional degree. The emotional speech corpus was examined to find acoustic and linguistic features which estimate the emotional degree of each utterance. Decision trees were adopted as classifiers for our estimation examination to find optimal sets of the acoustic and linguistic features for an anger degree estimation. As a result, we find specific tendencies of the tree acousitc features in strong anger utterances, the linguistic parameters' potential to estimate degree of anger emotion, and the capability of decision tree to estimate utterances with two kinds of acoustic features as the strong anger.

View full abstract

As a preparatory study for using emotion in advanced retrieval systems of books and movies, we analyzed expressions of emotion that appeared in film and book reviews, and defined the related components. First, through the manual analysis of 653 subjective expressions in 82 book reviews selected randomly from the web, we defined four major components related to expressions of emotion and their subcomponents: “Attitude”, “Subject”, “Object”, and “Reason” ; we analyzed the existence of each component and their combinations in the reviews. Second, we found that different writers stated different “Reasons” for the same “Emotion” about the same “Object” that tendency was confirmed in our additional analysis on emotion expressions, in which we focused on “Reasons” using different corpora consisting of film and book reviews and newspaper articles. Finally, we conducted experiments using 16 human subjects in two sessions of focused group interviews. The results showed that “Reasons” associated with emotion expressions appearing in film reviews are important and useful for users to select films relevant for them.

View full abstract

We annotated emotion tags to text-dialogs in comics with focusing facial expressions in order to construct reliable dialog corpus with emotion tags, and evaluated the reliability of the constructed corpus. Generally, the relationship between language expression and emotion of the speaker is ambiguous, so it is difficult to distinguish correct emotion existing inside of the speaker with referring only the language expression in dialog. To solve this problem, there exist many investigations using non-lingual information of acoustic data though it costs to collect much speech data. Therefore, in this paper, we focuse on the facial expression appearing in comics to gain the reliability of the emotional annotating. For instance, we used 10 comic books on “Chibimarukochan” containing 29, 538 sentences and constructed emotional corpus annotated by the scheme, where two annotators annotate facial tags and emotion tags temporally and then decide correct tags by their discussion together. The correct emotion tags were 16, 635. The evaluation results proved that the agreement ratio of the temporary emotion tags between the two annotators was 78% which was as good as the related work in speech (83%) and the correctness of the decided emotion tags was 97%. Next, since in our trial experiment to extract emotional suffix expression from the corpus we successed to extract 3, 164 ones, the usability for dialog analysis on emotion was clarified. Thus, we confirmed that the facial expression in comics is as effective for distinction of emotions as speech and the schema of corpus construction using facial expression in comics is reliable.

View full abstract

A human-like common sense and judgment is necessary to materialize a computer that can take communication with human. Because, when people talk to each other, they have the concept of emotion in our mind consciously or unconsciously. In the case, the ability to call concept in mind and to associate with many referred concepts will be an important matter. This paper proposes the method to systemize judgment concerning emotion, based on the mechanism to associate concept with many other referred concepts. As a result, the percentage of correct answers of the emotion judgment system is approximately 88.0%. Therefore, the emotion judgment system using the technique proposed in this paper is an effective system.

View full abstract

In recent years, approach which tries to process human's sensibility with computer has become active as information processing technology develops. It is necessary to recognize human emotions so that the anthropomorphic agent and the sensibility robot may behave like the person, and to express own emotions. We are researching the emotion recognition technology to apply it to the sensibility robot. The approach of most emotion estimation targeted only superficial emotion expression. In this paper, we propose an emotion estimation algorithm based on the emotion word (or emotion idiom) and the emotion occurrence event sentence pattern. A prototype system based on the proposed method has been constructed and an evaluation experiment has been carried out. The result shows that the proposed method is effective.

View full abstract

We propose a novel type of document classification task that quantifies how much a given document (review) appreciates the target object by using a continuous measure called sentiment polarity score (SP score) rather than binary polarity (good or bad). An SP score gives a concise summary of a review, and provides more information than binary classification. The difficulty of this task lies in the quantification of polarity. In this paper we use support vector regression (SVR) to tackle this problem. Experiments on book reviews using five-point scales show that SVR outperforms a multi-class classification method using support vector machines, and the results are close to human performance. We also examine the effect of sentence subjectivity detection using a Naive Bayes classifier, and show that this improves the robustness of the classifier.

View full abstract

This paper describes a method for classifying feedback documents into two polarities: positive and negative. In this method, we classify reputations into “Object Level Reputations” and “Attribute Level Reputations”. Object level reputations are the reputations concerning the object of the feedback document. Attribute level reputations are the reputations concerning one attribute of the object. Since we asuume that the polarity of the object level reputation corresponds to the polarity of the feedback document, feedback documents are first classified using the object level reputation. The feedback documents that do not contain object level reputations are classified using the attribute level reputations. In addition, this paper proposes a method for reliability evaluation of reputations considering two levels of reputations. In this method, we regard the attribute level reputations that has the opposite polarity of the feedback document as reliable reputations. The experimental results using movie reviews showed that the proposed method could classify feedback documents more correctly than the previous method, and that the proposed measure can be one of the reliability measures for reputations.

View full abstract