Baugh’s three utterances were also recorded for the database. More than
four hundred students at Stanford took part in this experiment, in which
they listened to the recorded segments and made note of their best guesses
in terms of race, ethnicity, and sex.
The third and fourth experiments had to do with the measurement of
very fine phonetic and acoustic detail in order to identify how sensitive
listeners are to cues well below the level of consciousness. The
methodology here involved isolating the single word hello from Baugh’s
three recordings. The reasons for limiting the experiment this way are
important:
We have several reasons for examining one word. This allowed us to
hold external factors to a minimum. Second, it also illustrates how
little speech is needed for dialect identification. “Hello” is a self-
contained utterance, making perceptual studies more natural. By
focusing on one short word, we are able to hold utterance duration
well below one second (x = 414 msec.), making it comparable to
other studies already made.
Student participants were again asked to identify race, ethnicity and
sex on the basis of listening to that single variable.
(Purnell et al. 1999: 13)In each of these four experiments, the authors found that they had to reject
the null hypothesis. That is, they began with the expectation that there
would be no correlation between language features and discriminatory
experiences. However, 70 percent of the time, listeners were able to
identify race, ethnicity and sex on the basis of an utterance less than a
second long (ibid.: 14). After analysis of the four experiments, the authors
came to four clear, statistically significant conclusions:
1. Dialect-based discrimination takes place.
2. Ethnic group affiliation is recoverable from speech.
3. Very little speech is needed to discriminate between dialects.
4. Some phonetic correlates or markers of dialects are
recoverable from a very small amount of speech.(ibid.: 11)