Current research issues in ICALL
Oberseminar, Detmar Meurers, Summer Semester 2011
Each of us as learners had different language experiences, yet we have converged on broadly the same language system. How so? Some views hold that there are constraints upon learner’s induction of language pre- programmed in some innate language acquisition device. Others hold that the constraints are in the dynamics of language itself – that language form, language meaning, and language usage come together to promote robust induction by means of statistical learning over limited samples. The research described here explores this question with regard English verbs, their grammatical form, semantics, and patterns of usage. Analyses of a 100-million-word corpus show how Zipfian scale-free distributions of usage ensure robust learning of linguistic constructions as categories: constructions are (1) Zipfian in their type-token distributions in usage, (2) selective in their verb form occupancy, and (3) coherent in their semantics. Parallel psycholinguistic experiments demonstrate the psychological reality of these constructions in language users.
I present a two-dimensional annotation scheme for capturing the given-new distinction in corpora. While classical approaches to ”information status” are limited to referring expressions (DPs), information structure theory from early on has made reference to the givenness or novelty of natural language expressions in general. Therefore, in order to investigate, e.g. prosodic effects on verbs or adjectives, the notion of information status has to be generalised to such non-referential expressions (”lexical information status”). I will talk about on-going annotation work and present some empirical results from spontaneous and read speech. I will also address the question how this relates to the problem of capturing focus-background in corpus data.
Conventional spell checkers usually only perform a shallow string comparison with a lexicon. Real-word spelling errors where the misspelled word is orthographically identical with another word present in the lexicon can not be detected in this way. If for example one misspells ”dairy” as ”diary”, the error will remain unnoticed.
Budanitsky and Hirst in “Correcting real-word spelling errors by restoring lexical cohesion” (2005) present a system to tackle this problem. They use the semantic network WordNet for English to find out whether a word, semantically unrelated to the words in its context, would have a replacement word obtained by a minimal alternation of the original that would fit in the semantic context.
In my talk, I will present Budanitsky and Hirst’s algorithm and report on the transfer of their system to German.
BACKBONE is a European project that has created and piloted small, pedagogically enriched video corpora (www.uni-tuebingen.de/backbone). The corpora are based on video interviews with adult speakers who talk about specific topics linked to their private and professional lives. The pilot courses were run as blended-learning courses in Moodle (learning platform). In my talk I will present the results of one of the pilot courses in which I created several tasks that focus on learning topic-based multi-word units (MWUs). I will analyze the uptake of these topic-based multi-word expressions in student writing based on different kinds of input. I will first look at the overall composite results. Secondly, I will trace the lexical development of individual students and compare it to their intuitive evaluation of the learning effects.
_________________________________________________________________________________
Last updated: December 31, 2011