Ildikó Pilán

Rule-based and machine learning approaches for second language sentence-level readability

We present approaches for the identification of sentences understandable by second language learners of Swedish, which can be used in automatically generated exercises based on corpora. In this work we merged methods and knowledge from machine learning-based readability research, from rule-based studies of Good Dictionary Examples and from second language learning syllabuses. The proposed selection methods have also been implemented as a module in a free web-based language learning platform. Users can use different parameters and linguistic filters to personalize their sentence search with or without a machine learning component assessing readability. The sentences selected have already found practical use as multiple-choice exercise items within the same platform. Out of a number of deep linguistic indicators explored, we found mainly lexical-morphological and semantic features informative for second language sentence-level readability. We obtained a readability classification accuracy result of 71%, which approaches the performance of other models used in similar tasks. Furthermore, during an empirical evaluation with teachers and students, about seven out of ten sentences selected were considered understandable, the rulebased approach slightly outperforming the method incorporating the machine learning model.

Detecting Context Dependence in Exercise Item Candidates Selected from Corpora.

We explore the factors influencing the dependence of single sentences on their larger textual context in order to automatically identify candidate sentences for language learning exercises from corpora which are presentable in isolation. An in-depth investigation of this question has not been previously carried out. Understanding this aspect can contribute to a more efficient selection of candidate sentences which, besides reducing the time required for item writing, can also ensure a higher degree of variability and authenticity. We present a set of relevant aspects collected based on the qualitative analysis of a smaller set of context-dependent corpus example sentences. Furthermore, we implemented a rule-based algorithm using these criteria which achieved an average precision of 0.76 for the identification of different issues related to context dependence. The method has also been evaluated empirically where 80% of the sentences in which our system did not detect context-dependent elements were also considered context-independent by human raters.