Muniche phonology article
September 4, 2013
While I was in the field this summer, an article that the team members of the Muniche Rapid Documentation Project wrote came out in the International Journal of American Linguistics. This article focuses on the segmental and prosodic phonology of Muniche, a remarkable linguistic isolate spoken in the Huallaga River basin of Peru, just where the foothills of the Andes start to become noticeable. This article (available here) has a somewhat methodologically interesting backstory.
Briefly, the project was organized around working with three rememberers of the language (no fully fluent speakers remain, alas) over the course of a single summer field season to document as much as we could. The most fluent rememberer, Alejandrina Chanchari, was probably in her early 90s at the time, and in very poor health, lending urgency to this work. As part of the project, we planned to create a number of works for interested community members, including an audio CD with recordings of all the headwords of the (modest) dictionary we prepared. With this community-oriented goal in mind, two of the team members, UC Berkeley graduate students Stephanie Farmer and Greg Finley, systematically obtained relatively clean recordings of each of the headwords, which they eventually used to make the planned CD.
Now, back to the article: it began as a quite modest paper that focused on the segmental phonology of the language, and the differences between what we had discovered and phonology as described in the previous major publication on Muniche. Given the language attrition with which we were faced, I initially held out little hope that we could do work on the prosodic system of the language, but we found ourselves willy-nilly having to deal with stress and other metrical phenomena in order to adequately discuss the segmental phonology. In particular, it turns out that glottal stops are associated with stress, and we needed to distinguish those glottal stops that could attributed to stress and weight requirements, from those that were attributable to the underlying segmental representation. To my pleasant surprise, it turned out that between the systematic dictionary recordings and the other tokens that the team had recorded, we were actually able to say a great deal about Muniche prosody: the recordings were sufficiently clear, and the rememberers speech exhibited great regularity in its stress patterns. And it was really the recordings that we had made for the community-oriented materials that provided the crucial empirical basis for our prosodic analysis.
I’ve worked on the prosodic systems of a number of languages, like that of Nanti (here) and Iquito (here), and based on this work, I had come to believe that work on stress systems of any reasonable complexity requires real-time face-to-face work with linguistic consultants to be able to explore analytical options sufficiently quickly and flexibly. But the Muniche case showed that this is not necessarily the case. You can do good work on prosodic systems with nothing more than a set of audio recordings: you just need a lot of them — like, say, 500-1,000 tokens. Of course, such work is a lot slower, and there is always the risk that you’ll be missing key data — my surprise was that the latter problem proved not to be anywhere near as great as I had expected (but we did have something like 1,000 recorded tokens).
In any case, this experience was a nice validation of the claim that community-oriented work and scientific work need not be in competition, and that the former can indeed support the latter. This work also suggested to me that recording a large number word tokens clearly, systematically, and where appropriate, in stress or tone frames, can be a valuable component of language documentation. Given the emphasis on texts in much language documentation work, the latter point might be worth keeping in mind.