Biology session

8-month-old infants are able to distinguish words based only on the statistical information of the speech stream. In this study, we tried to answer the question if family dogs (Canis familiaris) that live in a similar familial environment could possess the ability of word segmentation based only on the statistical regularities. We studied two parts of the statistical information of the speech stream with fully non-invasive electroencephalographic (EEG) method. We tested whether the brain activity of the dogs (Event-related potential, ERP) differ between: 1) syllable sequences that occur more or less frequently 2) the processing of syllable sequences that usually co-occur with each other (words with high transitional probability) against the combinations that could combine with other syllables (words with low transitional probability).

In the first phase of the experiment – so-called familiarization – we presented artificially synthesized human speech stream through speakers over c.12 minutes to seven family dogs (n=7). We made the stimuli from 12 different syllables (e.g. da, ro, pi) and played in a semi-randomized order without pauses. In the second test phase we presented three-syllable sequences (‘words’) from the familiarization with pauses (e.g. daropi, tibudo): 1) words with high frequency and high transitional probability 2) words with low frequency and high transitional probability 3) words with low frequency and low transitional probability – so-called part-words – (e.g. daropi+golatu=pigola) which contained the word boundaries of the high frequency words 4) random words which weremade from the syllables of the low frequency words but never appeared in that order during familiarization. All conditions included two words and every word was presented 40 times while we measured the ERP-s of the dogs.

We compared the ERP-s of the high frequency and the random words with permutation t-test after the beginnings of the words and found significant differences in the time windows among e.g. 200 and 300 ms, 350 and 450 ms and 530 and 560 ms. on the Fz centreline electrode. We also found significant difference between the conditions of low frequency words and part-words in the time window among 470 and 475 ms. However, the ERP-s of the part-word condition were more similar to the ERP-s of the high frequency words and the ERP-s of the low frequency words were more similar to the ERP-s of the random words. Because of these reasons our results could be more explainable with the appearance probability of the syllables that form words and their combinations instead of the frequency and transition probability of the syllable sequences. According to our experiment, dogs seem sensitive to the syllable frequencies of the human speech stream. However, our ongoing research does not prove that dogs can reliably segment the syllable sequences into words based on the distribution of statistical information.