On the Contribution of MWE-based Features to a Readability Formula for French as a Foreign Language

Proceedings of Recent Advances in Natural Language Processing, pages 441–447, Hissar, Bulgaria, 12-14 September 2011. On the Contribution of MWE-based Features to a Readability Formula for French as a Foreign Language Thomas Franc¸ois Aspirant F.N.R.S. Centre for Natural Language Processing Institut Langage et Communication UCLouvain thomas.francois@uclouvain.be Patrick Watrin Centre for Natural Language Processing Institut Langage et Communication UCLouvain patrick.watrin@uclouvain.be Abstract This study aims to assess the usefulness of multi-word expressions (MWEs) as fea- tures for a readability formula that pre- dicts the difficulty of texts for French as a foreign language. Using a MWE extrac- tor combining a statistical approach with a linguistic filter, we define 11 predictors. These take into account the density and the probability of MWEs, but also their inter- nal structure. Our experiments show that the predictive power of these 11 variables is low and that a simple approach based on the average probability of n-grams is more effective. 1 Introduction With the success of the communicative and action- oriented approach in the teaching of a second or foreign language (L2), teachers are encouraged to work on authentic texts in order to bring their stu- dents in contact with real linguistic data. The web is a valuable source for such documents, but the search for a document tailored to the level of the students may sometimes be tedious. In this con- text, readability studies may help. They aim to de- velop tools capable of assessing the difficulty of texts for a given population through textual fea- tures only (such as the number of letters per word, the number of words per sentence, etc.). However, while many studies have examined the readability of English L1 (Chall and Dale, 1995), there are far fewer studies on readability in an L2, especially in French as a foreign language (FFL). In most cases, formulas for native speakers have been applied to L2 texts. However, the va- lidity of such an approach is far from established, because it relies on three suspect assumptions : (1) the understanding of readers in the L2 is compara- ble to that of native speakers, (2) the textual fea- tures considered in L1 formulas are relevant to L2 reading, and (3) the weighting of these variables may be the same in a formula for L1 and L2. If some work by Greenfield (2004) supports this vision, other authors disagree and consider that the peculiarities of the reading process in the L2, described by Koda (2005) among others, must be taken into account by designers of read- ability formulas. Of these dimensions, the in- terferences between the L1 and L2 of the learn- ers are certainly among the most studied topics (Uitdenbogerd, 2005; Laroche, 1979). Moreover, Franc¸ois (2009) has shown that considering verb modes and tenses leads to the significant improve- ment of a L2 formula. However, there is another textual aspect that is likely to be a good predictor of lexical difficulty for L2 readers: collocations and idioms. A good knowledge of these items is indeed …

Content truncated to 500 words