Table 10. The most common match patterns, which consist of 2 events. With these ideas in mind, a detailed note was made on the 222 recordings. The texts of the speakers and their agents represented about half a million common words, and each word was timed in the flow of the given language, a particularly useful characteristic of the body for associating non-verbal events with the corresponding text. The text has also been spoken for morphology and syntax with notes, another important feature that, because of its information on linguistic errors and incompleteness, can contribute, during a conversation, to learn more about the cognitive flow of linguistic behavior and to the development of more natural interactive systems. Video commentary included facial expression classes, gaze, eyebrows, head movement, hand shape, touch, posture, Deixis, emotions and emblem. Audio has been commented on for intonation sentence classes, emotions and speech, in addition to phonetic events in language, such as silence, hesitation, reboot, non-phonetic sounds and noise. Automatic methods have been applied to the annotation of the phonetic characteristics of the sound track: in addition to marking the absolute values of F0 and intensity, a special algorithm (Szekrényes, 2014, 2015) was used to comment on the stylized contours of intonation and intensity of language, to capture the contribution of linguistic prosody to the multimodal expression of pragmatic content of interaction. Pragmatic levels of note included class management, attention, agreement, deixis and information structure. Because each event in a class was designated as time-related, the diagram linked virtually all events of any kind to a different event. Due to computational constraints (which we will do more specifically in the next section), this work concerns only the physical classes of the gaze, the shape of the hand, the movement of the head and posture, and the pragmatic interpretive classes of the concordance and the emblem. This study uses data from this restricted series of HuComTech Corpus classes to identify multimodal models related to agreement/disagreement behavioral events. Among the many possible observations, we point out only the role of blinking: there is usually as an accompaniment agreement, either its beginning or its end (although with fewer incidents, the same goes for disagreements).

The beginning of the agreement is also strongly related to “e,down”, that is, the spokesperson stops looking down – look up effectively, it will most likely touch the eyes of the agent. If the speaker often begins to look down (“b,down”), as he begins to accept, this may suggest a moment of reflection and, ultimately, it is the turn to continue the conversation. “I say that with the respect it deserves, but… is a great way to explain a disagreement, especially in a professional or formal environment. As mentioned above, shaking your head is an essential element in the models back and forth.