Proceedings of the 13th Conference on Natural Language Processing, KONVENS 2016, Bochum, Germany, September 19-21, 2016

The most important reasons for examining “non-standard data” with CL methods are the facts that this data represents a great deal of language behavior and that it serves as an object of scientific study in linguistics as a whole. This is true of the syntax of non-native second-language learners, the accents of non-native speakers, and the vocabularies of different dialect speakers. Computational linguists have a good deal to offer to the various subfields of linguistics studying non-standard data. By automating steps in analysis we make the analyses replicable and also modifiable, we improve opportunities for comparing similar analyses, and perhaps most importantly, we enable the analyses of large amounts of data, providing more comprehensive views. The data itself can be tricky to work with, however, as scientists in other fields are often specialized in a single language or language pair, which means that their data will not be varied enough to support all the research questions one would like to ask, e.g., the question of the generality of the techniques for a particular purpose. In other cases, the data simply won’t have been collected with an eye to answering some interesting questions, which may mean that important parameters haven’t been recorded. Finally, we note that non-automated analyses do not impose expectations that data be commensurate to the same strict degree (as automated ones), meaning that surprises can be in store even in well-studied data sets. This paper provides some concrete examples and discussion of these potential pitfalls. One can protect oneself from some of these risks by seeking collaboration with domain experts, which is to be recommended in any case, as a way of making the work richer and better informed. Further, it makes sense to approach novels sorts of data — and even novel sources of data of a sort one suspects is familiar — with a broad range of potential research questions. There is an awful lot of interesting work still to be done!