“Could you make me a favour and do coffee, please?”: Implications for Automatic Error Correction in English and Dutch

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 49–53, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics “Could you make me a favour and do coffee, please?”: Implications for Automatic Error Correction in English and Dutch Sophia Katrenko UiL-OTS Utrecht University s.katrenko@uu.nl Abstract The correct choice of words has proven chal- lenging for learners of a second language and errors of this kind form a separate category in error typology. This paper focuses on one known example of two verbs that are often confused by non-native speakers of Germanic languages, to make and to do. We conduct ex- periments using syntactic information and im- mediate context for Dutch and English. Our results show that the methods exploiting syn- tactic information and distributional similarity yield the best results. 1 Introduction When learning a second language, non-native speak- ers make errors at all levels of linguistic analy- sis, from pronunciation and intonation to language use. Word choice errors form a substantial part of all errors made by learners and may also be observed in writing or speech of native speak- ers. This category of errors includes homophones. Some commonly known confusions in English are accept-except, advice-advise, buy-by-bye, ate-eight, to name but a few. Other errors can be explained by a non-native speaker’s inability to distinguish be- tween words because there exists only one corre- sponding word in their native language. For ex- ample, Portuguese and Spanish speakers have diffi- culties to differentiate between te doen (to do) and te maken (to make), and Turkish between kunnen (can), weten (to know) and kennen (to know) in Dutch (Coenen et al., 1979). Adopting terminol- ogy from Golding and Roth (1999) and Rozovskaya and Roth (2010), do/make and kunnen/kennen/weten form two confusion sets. However, unlike the case of kunnen/kennen/weten, where the correct choice is often determined by syntactic context 1, the choice between to make and to do can be motivated by semantic factors. It has been argued in the litera- ture that the correct use of these verbs depends on what is being expressed: to do is used to refer to daily routines and activities, while to make is used to describe constructing or creating something. Since word choice errors have different nature, we hypoth- esize that there may exist no uniform approach to correct them. State-of-the-art spell-checkers are able to detect spelling and agreement errors but fail to find words used incorrectly, e.g. to distinguish to make from to do. Motivated by the implications that the correct prediction of two verbs of interest may have for au- tomatic error correction, we model the problem of choosing the correct verb in a similar vein to selec- tional preferences. The latter has been considered for a variety of applications, e. g. semantic role la- beling (Zapirain et al., 2009). Words such as be or do have been often excluded from consideration be- cause they are highly polysemous and “do not select strongly for their arguments” (McCarthy and Car- roll, 2003). In …

Content truncated to 500 words