180
PL. Tyack and c.w. Clark
different computer techniques, including that used by McCowan (1995), in
order to test the effect of different categorization methods on interpretations about signature whistles. Janik (1999) tested all of these different
categorization methods on one data set of whistles recorded from captive
dolphins that was reported in an earlier paper (Janik and Slater 1998). An
important feature of this data set was that Janik and Slater (1998) found
that the dolphins' usage of whistles helped to validate their categorization.
They categorized whistles by visual examination of spectrograms before
analyzing usage. They found four very stereotyped whistles, along with
a variety of other whistle types. The four stereotyped whistles showed
individual-specific usage. Each stereotyped whistle was used almost exclusively by only one individual dolphin and only when it was isolated from
other members of its group. Since each of these whistles was used by a
different individual when isolated,Janik and Slater (1998) categorized these
as signature whistles. This validation of signature whistles allowed them to
test how well the different categorization methods can identify these whistle
types. Five humans visually categorized spectrograms of the whistle sample,
presented in random order. No information on context was presented. All
five humans agreed on a classification of whistles that matched the individual usage of signature whistles by the dolphins. However, none of the
three computer methods was capable of identifying signature whistles reliably. The critical issue for resolving which categorization is correct is to
study how dolphins themselves categorize whistles. Previous research has
shown that dolphins can identify and categorize signature whistles, even if
they hear as little as 0.5s of the whistle (Caldwell et al. 1969). Playback
experiments with wild dolphins also show that dolphins respond more
strongly to the signature whistles of animals with whom they have shared
a strong bond (Sayigh et al. 1999). Future work with synthetic whistles is
required to determine which features dolphins use to identify signature
whistles.
Dolphins have extraordinarily good abilities of discriminating different
frequencies compared to other mammals, and can detect a change of as little
as 0.2% in frequency (Thompson and Herman 1975). The range in which
dolphins can discriminate frequencies the best in terms of percentage
change, or the Weber ratio, is near 10 kHz. This matches the frequency range
of dolphin whistles. Since dolphin whistles are frequency-modulated tonal
sounds, most researchers who have studied whistles have concluded that
frequency cues are critical elements of these communication signals. The
importance of discriminating among subtle frequency cues in whistles may
have created a selection pressure for such sensitive frequency discrimination. Bottlenose dolphins can detect faint signals best in a different
frequency range than that of their best hearing to detect differences in
frequency. Bottlenose dolphins can detect the faintest sounds at frequencies from 50 to 75kHz, much higher than the 10kHz frequency at which
dolphins can best discriminate differences in frequency. As just discussed
PL. Tyack and c.w. Clark
different computer techniques, including that used by McCowan (1995), in
order to test the effect of different categorization methods on interpretations about signature whistles. Janik (1999) tested all of these different
categorization methods on one data set of whistles recorded from captive
dolphins that was reported in an earlier paper (Janik and Slater 1998). An
important feature of this data set was that Janik and Slater (1998) found
that the dolphins' usage of whistles helped to validate their categorization.
They categorized whistles by visual examination of spectrograms before
analyzing usage. They found four very stereotyped whistles, along with
a variety of other whistle types. The four stereotyped whistles showed
individual-specific usage. Each stereotyped whistle was used almost exclusively by only one individual dolphin and only when it was isolated from
other members of its group. Since each of these whistles was used by a
different individual when isolated,Janik and Slater (1998) categorized these
as signature whistles. This validation of signature whistles allowed them to
test how well the different categorization methods can identify these whistle
types. Five humans visually categorized spectrograms of the whistle sample,
presented in random order. No information on context was presented. All
five humans agreed on a classification of whistles that matched the individual usage of signature whistles by the dolphins. However, none of the
three computer methods was capable of identifying signature whistles reliably. The critical issue for resolving which categorization is correct is to
study how dolphins themselves categorize whistles. Previous research has
shown that dolphins can identify and categorize signature whistles, even if
they hear as little as 0.5s of the whistle (Caldwell et al. 1969). Playback
experiments with wild dolphins also show that dolphins respond more
strongly to the signature whistles of animals with whom they have shared
a strong bond (Sayigh et al. 1999). Future work with synthetic whistles is
required to determine which features dolphins use to identify signature
whistles.
Dolphins have extraordinarily good abilities of discriminating different
frequencies compared to other mammals, and can detect a change of as little
as 0.2% in frequency (Thompson and Herman 1975). The range in which
dolphins can discriminate frequencies the best in terms of percentage
change, or the Weber ratio, is near 10 kHz. This matches the frequency range
of dolphin whistles. Since dolphin whistles are frequency-modulated tonal
sounds, most researchers who have studied whistles have concluded that
frequency cues are critical elements of these communication signals. The
importance of discriminating among subtle frequency cues in whistles may
have created a selection pressure for such sensitive frequency discrimination. Bottlenose dolphins can detect faint signals best in a different
frequency range than that of their best hearing to detect differences in
frequency. Bottlenose dolphins can detect the faintest sounds at frequencies from 50 to 75kHz, much higher than the 10kHz frequency at which
dolphins can best discriminate differences in frequency. As just discussed