tv.um.es

Reliable readability assessment systems are fundamental to produce accessible information. In the 1970s, the first readability formulae started to appear, such as the Gunning Fog Index and the Flesch Reading Ease Test. They take into account two superficial aspects of complexity, mean sentence and word length, to predict the readability level of a text. Over the last decades, clarity has become a matter of increasing concern for legislators and institutions (Eichhoff-Cyrus and Antos, 2008). As a matter of fact, normative acts are a classical example of texts drafted with a view of striking an optimal balance of precision and ease of reading, two objectives that often work in opposite directions. Indeed, while the former often calls for, among other things, lexical and syntactic sophistication, the latter is ore easily achieved by resorting to basic vocabulary and less convoluted phrasing.
A formula to assess specifically the readability of texts drafted in Italian was created in the late 1980s, the Gulpease index (Lucisano and Piemontese, 1988). It models readability as a function of three variables: the number of sentences, the number of words and the number of letters. This index was generally welcomed with much enthusiasm in that it provides a relatively easy-to-interpret measure of the overall difficulty of a text. It is not a case that it is still used today as the starting point of different readability studies (Felici and Griebel, 2019; Venturi, 2012), in association with more sophisticated analyses on the lexical and syntactic level. Today, “second generation readability tools”, such as READ-IT for Italian (Dell’Orletta, Montemagni, and Venturi, 2011), allow for an automatic investigation of a wide range of linguistic features.
Moving from this background, this study aims to investigate whether the Gulpease index can be used as a reliable predictor of lexical and syntactic readability in the legal domain. To this end, we relied on LEX.CH.IT, a corpus of Swiss legislation in Italian (Canavese, 2019). More precisely, we worked on a subcorpus of approximately 100 texts covering the timespan 2000-2018 with a similar length. For each text, we extracted some lexical (basic vocabulary, MATTR) and syntactic features (such as noun-verb ratio, lexical density, use of subjunctive mood, subordination, passive voice, nominal modification) through READ-IT. We then moved on to compare these results with the readability level predicted by the Gulpease index. We resorted to a number of quantitative analysis tools to detect the presence of correlation between the Gulpease index and other indicators traditionally associated with a higher or a lower level of readability. The results suggest that the Gulpease index is not able to capture lexical and syntactic complexity in all its facets. Therefore, ad hoc tests are necessary to assess the actual level of readability of a legal text.