Single Review on past events

How do children learn their language? – With this question Ariel Gutman started his Jour Fixe talk on November 22. And he gave a quick and short answer: “We don´t know”. Of course he didn´t want to leave it at that, but tried to answer the question by talking about “Prosody as means for acquisition of the syntactic skeleton and word categorization”. He explained the difficulties in testing the learning mechanism systematically, but that instead he could test certain aspects of language acquisition by computer simulation. Fortunately Ariel Gutman studied, besides linguistics, computer science as well, so it is easier for him to understand the algorithms that he is using for his experiments. His tests are based on the syntactic skeleton model (developed by Christope et al, 2008) that makes use of prosodic phrases as approximative syntactic units. He illustrated this on the basis of an example sentence: “The little boy is running fast” which consists of the nominal phrase “The little boy” and the verbal phrase “is running fast”, both being prosodic phrases as well. Whereas “the” and “is” are function words, “boy” and “running” are content words. The linguist´s assumptions are 1. children use the function word of the phrase in order to classify it, 2. children know already some of the most frequent content words and can identify them as objects (=nouns) or actions (=verbs)  and 3. based on this knowledge they can identify the class of the current prosodic phrase.

For his computer simulations Ariel Gutman used transcriptions of speech directed to 1- to 3-year old children. The computer algorithm groups different prosodic phrases into different clusters and then compares them to the correct syntactic categories. In order to simulate learning he uses the well-known Expectation-Maximization algorithm that consists of several phases: Initialization (initially assign the prosodic phrases to a possibly wrong  “syntactic” cluster), Maximization (estimate the parameters of the model), and Expectation (re-assign the prosodic phrases to new “syntactic” clusters). The two last steps are repeated until you achieve convergence. The initialization phase relies either on the function words or the content words appearing in the prosodic phrase. He admitted that clearly the experiment is not cognitively plausible, but it can serve as an approximation of true incremental learning. Finally, “we can evaluate the ‘quality’ of each resulting cluster by examining the largest syntactic category which it has captured”.

And the experiments seem to confirm his assumptions: The language learner can rely on prosodic boundaries, function words and content words in order to construct a syntactic skeleton. Ariel Gutman described learning as iterative process, as the semantic knowledge allows the discovery of the importance of function words in syntactic classification, and the knowledge of function words in turn may help in the classification of novel words.

But – and that was a really good question from the audience after Ariel Gutman´s presentation – what´s the practical information that he would give parents in order to teach their children a language? – As in the beginning the linguist had a short answer ready: I think the parents are doing fine!