You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
45 lines
2.8 KiB
45 lines
2.8 KiB
Some measures of intelligibility and comprehension
|
|
|
|
us with additional information that could be used to modify or improve several of
|
|
the modules of the system. Whether such additional improvements at these
|
|
various levels of the system will actually contribute to improved intelligibility and
|
|
comprehension is difficult to assess at this time, since performance with meaning-
|
|
ful sentences is already quite high to begin with, as shown by the present results
|
|
obtained with the Harvard sentences.
|
|
|
|
In summary, the results of tests designed to measure word recognition in two
|
|
types of sentential context showed moderate to excellent levels of performance
|
|
with synthetic speech output from the current version of the text-to-speech system.
|
|
As in the previous section dealing with the evaluation of the intelligibility of iso-
|
|
lated words, the present results, particularly with rather diverse meaningful sen-
|
|
tences, suggest that the quality of the speech output at the present time is probably
|
|
quite satisfactory for a relatively wide range of applications requiring the process-
|
|
ing of unrestricted text. While there is room for improvement in the quality of the
|
|
output from various modules of the system, as suggested by the results of the Has-
|
|
kins anomalous sentences, it is not apparent whether the allocation of resources to
|
|
effect such changes in the system would produce any detectable differences. Dif-
|
|
ferences that might be detected, if any, might well require a very restricted listen-
|
|
ing environment in which all of the higher-level syntactic and semantic infor-
|
|
mation is eliminated, a situation that is unlikely to occur when the system is imple-
|
|
mented in an applied setting. Given these results on word recognition, however, it
|
|
still remains to be determined how well listeners can understand and comprehend
|
|
continuous speech produced by the system, a problem we turn to in the next sec-
|
|
tion of this chapter.
|
|
|
|
13.4 Comprehension
|
|
Research on comprehension and understanding of spoken language has received a
|
|
great deal of attention by numerous investigators in recent years. It is generally
|
|
agreed that comprehension is a complex cognitive process, initially involving the
|
|
input and subsequent encoding of sensory information, the retrieval of previously
|
|
stored knowledge from long-term memory, and the subsequent interpretation, in-
|
|
tegration or assimilation of various sources of knowledge that might be available
|
|
to a listener at the time. Comprehension, therefore, depends on a relatively large
|
|
number of diverse factors, some of which are still only poorly understood at the
|
|
present time. Measuring comprehension is difficult because of the interaction of
|
|
many of these factors and the absence of any coherent model that is broad enough
|
|
to deal with the diverse nature of language understanding.
|
|
|
|
One of the factors that obviously plays an important role in listening com-
|
|
|
|
161
|