Linguistic Elicitations

What are “Elicitations”?

I recently took a linguistics class called SLIYS from Ohio State for linguistics knowledge that is equivalent to an intro class in college. One of the activities we conducted was called “elicitations”, something I had not experienced previously. The idea of the activity is to learn about a language – the grammar, syntax, vocabulary, etc. – but the only data you’re allowed to have are translations of English to that target language. This process is similar to learning the language as a linguist would for a novel, undocumented language. The only difference is the lack of machinery to detect vocal sounds as there would be for a novel language, as knowing the phonetics of a language can help linguists discern patterns in other aspects of that language, but this difference is not essential for the level of precision we wanted in our elicitation, we only needed the experience. This process of obtaining information from only getting translations of words, sentences, and utterances (sentences but with context) is called elicitation in linguistics.

The Process

The first step was to establish a so called “sound inventory” of the language, according to the International Phonetic Alphabet (IPA). This alphabet consists of symbols to represent a specific sound a human could make based off of tongue placement and movement, as well as other muscles in or near the jaw. Vocal cords may also be considered for some consonants and all vowels. The language our group would be assigned to was Albanian, and our job would have been to find which sounds that were from the IPA appeared in Albanian, considering the IPA contains all sounds a jaw could make, so it was the best way to document the sounds, just like a real linguist would have. Once the sounds were documented, obtaining syntactic and grammatical rules would become easier.

After-words (Bad pun, I know), us elicitors would ask our language consultant, the person who contributed our translations, sentences with various number, person, tenses, and case use for nouns, to see if there was a difference in word sounds, which would signify that it may be a special rule to decline a noun or conjugate a verb. The process reminds me of how a neural network is trained, where input and output data are given, but not the rules for how that was obtained (because often times we don’t know what the rules are), so in a way we were all acting as neurons.

Speaking of neurons, these processes and the knowledge of linguistics, while perhaps not directly useful for neural networks, are helpful in understanding and applying neural networks to NLP and computational linguistics. In fact, from what I have learned, computer scientists working in this field will also consult linguists when working on NLP projects, to gain their expertise and advice, which is part of the reason I am invested in this linguistics program, in order to also gain a foothold in linguistics and better my own skills in the field.

Thanks for learning about linguistics with me, and taking a look to a part of an activity in a field that is not often called to thought when one mentions “linguistics”.