Graphics Reference
In-Depth Information
6.4 Measurements of prosody and paralinguistic parameters
Verbal communication can be analyzed in a hierarchically ordered
manner. After processing the acoustic information with a signal-
theoretical approach, prosodic and linguistic language information
is processed separately. The prosodic information contains references
about the emotional and motivational content of the act of
communication. These correspond with or differ from the linguistic
emotion- and cognition-related information from the semantics. On the
highest level, the prosody and semantics of the spoken information
can be interpreted as possible intentions. Within the context of this
framework, reference is made only to emotion-related information.
The analysis of emotional language produces a multitude of
prosodic characteristics, which differ significantly from modal and/or
unemotional language. Among others, features such as fundamental
frequency progressions, sound pitch progressions, energy progressions,
speech pause frequency and pattern, the stretching of words and
syllables, frequency changes and voice qualities were identified
(Scherer, 2003a; Yanushevskaya et al., 2007). Researchers working in
the area of automatic emotion recognition in speech use these features
(Borst et al., 2004). In addition, there are many para-linguistic events
that transmit emotionally colored information, such as laughing
(Scherer et al., 2011). As already mentioned above, semantic and
speech-content phenomena are also researched in addition to the
prosodic and para-linguistic information hidden in speech (Schuller
et al., 2003; Scherer et al., 2012b).
6.5 Semantics
Content-analytical processes, for some time now, have been based on
key words or phrases. More complex processes carry out syntactic
and semantic analyses. In certain areas of application with a limited
speech scope, the results are robust and can be applied in practice.
The text-based detection of emotional information is a proven method:
Harvard Dictionaries, LIWC (Pennebaker and Francis, 1996) and
LEAS (Kessler et al., 2010) and is easy to use, once words have
been identified. Since natural speech rarely follows grammar rules,
computer-linguistic methods often face significant problems in the
identification of meanings. The automated content analysis can be
rendered less ambiguous with decision-theoretical methods.
So far, however, there is an insufficient number of language
databases with natural emotionality in the language. First analyses,
however, show promising emotion classification methods (Gnjatovic
'
Search WWH ::
Custom Search