Barceló-Coblijn, L. 1 , Irurtzun, A. 2, 3 , Palmer-Riera, M. 1 & Gomila, A. 1
1 Universitat de les Illes Balears
2 IKER
3 CNRS
We present a new technique for the extraction of more refined data from the language source using a combination of linguistic, morpho-syntatic analysis and network science, able to extract information that cannot be observed by the unaided eye. Complexity patterns of word combination emerge from a sample of spontaneous speech under the form of a complex network made of words/morphemes (its nodes) and syntactic links (its edges). Objective indicators are extreacted reflecting the cohesion of the speaker's linguistic system. Hence, linguistic networks seem to be good biomarkers/endophenotypes for language impairment/atypical language development.
We applied this technique to 7 one-year-long corpora of typical children covering 6 different languages, 32 samples of Down syndrome, 20 cases of Specific language impairment and 20 cases of Hearing impariment. Moreover, we got the first data of a pilot study of Williams syndrome and networks. We observe that typical children develop their linguistic networks following a common developmental path, whereas the atypical cases differ from that developmental path. The structure of atypical linguistic networks is different in many ways and the formal indicators support the idea that these speakers have a qualitatively different ability to combine words and information management.