Automatic Grammar Checking for Second Language Learners – the Use of Prepositions

This paper presents a partial extension of a tool for automatic grammar checking of Swedish text. The work was carried out within a research project aiming at designing the grammar checker to meet the needs of second-language writers. The paper discusses the construction and implementation of a new set of matching rules for the grammar checking system, written with the purpose of detecting second language learners’ writing errors related to the use of prepositions. A minor evaluation indicated high precision for the performance of this set of rules and an Fscore of 40 per cent. The paper also discusses the conclusions drawn during the research process for the future treatment of problems with the use of prepositions. 1 A grammar checker for Swedish text GRANSKA is an automatic grammar checker developed at the Department of Numerical Analysis and Computer Science (Nada) at the Royal Institute of Technology in Stockholm (Domeij et al., 2000; Knutsson, 2001). Other grammar checkers for Swedish have been developed at the universities in Gothenburg (Andersson et al., 1999) and Uppsala (Sågvall Hein, 1998) and at Lingsoft Oy in Finland (Birn, 2000; Arppe, 2000). Like Lingsofts grammar checker, G RANSKA is a phenomenon-based system. This means that the program is provided with error detection rules which are designed to match expected writing errors in the input text. This also means that the program does not try to make a full syntactic analysis of the text and that it is not expected to react to a syntactically distorted but very unusual sequence like Arg gula Nicke på var mannen den hatten med (Angry the yellow Nicke with was the man the hat with ) whereas agreement violations at a morpho-syntactic level like Mannen med det gula hatten var arga på Nicke(The man with the yellow hat was angry with Nickethe Swedish noun phrase and predicative (dis)agreement markers disappear in the English translation) are detected, providing the user with diagnosis and suggestions for the correction of the errors. The underlying morpho-syntactic analysis is carried out by means of a statistical trigram tagger (Carlberger and Kann, 1999). The text that is to be checked is first analyzed by the tagger and the tagged text is the input to the rule-based grammar checking process. Since the fall of 2001 the main target for the research on grammar checking at Nada are writers who learn Swedish as a second language (Knutsson et al., 2003b). The aim of the research project named CrossCheck is to develop a useful tool for this particular group of writers. There is a fairly well-defined set of phenomena known to be particularly difficult to second language learners of Swedish, namely noun phrase agreement, word order, tense and the use of prepositions (Knutsson et al., 2003a; Staerner, 2001; Öhrman, 2000; PitkänenKoli, 1990). It is the latter category that we have paid attention to in our search for error patterns in second language learners’ texts for the subsequent formalization and eventually the construction of a new set of rules for the system.