Introduction to Corpus Linguistics. Sandrine Zufferey

These tools have been developed for simplifying searches within a corpus. Otherwise, it would be very inconvenient to use the standard tools that are present in a word processor, for example. In particular, concordancers make it very easy to extract all the occurrences of a word or an expression with its left and right context, as well as to determine its main collocations. These tools also help us create a list of all the words in the corpus, sorted by frequency. While one corpus can be compared to another reference corpus, these tools also make it possible to extract a list of keywords that are specific to the corpus studied. In the field of multilingual corpora, aligners make it possible to align parallel corpora sentence by sentence, and then to extract a sentence and its translation by means of a bilingual concordancer.

      5) A quantitative study on this question could focus on the creation of categories for classifying spelling mistakes, for example, agreement errors, redoubling of consonants, dumb letters, etc., and then counting all the occurrences of errors belonging to each category. By applying a statistical test, this study would then make it possible to know whether students tend to make certain types of mistakes more often (e.g. grammatical errors) rather than other mistakes (e.g. lexicon errors). A qualitative study on this same question would identify some examples of spelling mistakes and analyze in detail the contexts in which they occur, for example the grammatical category of the words concerned, whether they are rare or frequent words, occurring in a long or a short sentence, what type of phonemes is poorly transcribed, etc. This study would make it possible to identify linguistic contexts that tend to be conducive to spelling mistakes.

6) The results of the quantitative corpus study summarized above, namely the quantification of the different types of spelling mistakes, could be considered as a kick off for an experimental study. For example, the corpus study could help identify one type of common error, and one type of rare mistake. An experiment could then help to determine whether being in a stressful situation or not has a different impact on the two types of error.

      7) a) In order to study a phonological phenomenon like this, a spoken corpus is essential. This corpus should be specific to the population of French-speaking Switzerland. A large type of corpus comprising a large number of different speakers would be desirable. Finally, this corpus should contain a synchronic type of data, corresponding to the current pronunciation, rather than to its diachronic evolution.

      b) In order to study the evolution of a language, a diachronic corpus is essential. This constraint implies the use of written data, since oral data only go back to the middle of the 20th Century. Finally, the chosen corpus should include productions made by adult native speakers.

      c) In order to study translation, a parallel corpus is required. This corpus should contain original texts in French and their translation in English. It should be a synchronic corpus, corresponding to current uses of the language.

1.10. Further reading

For a discussion regarding the main defining elements of corpus linguistics, the works by Habert et al. (1997) are an excellent introduction in French, even if tools and methods have evolved considerably since they were published. On Chomsky’s objections to corpus linguistics, refer to the book by McEnery and Wilson (2001, Chapter 1). A more detailed discussion with possible responses of corpus linguistics to these objections can be found in Aarts (2001). As regards the role of computer science in the evolution of corpus linguistics, as well as a typology of corpora, see Tognini-Bonelli (2010). On the differences between qualitative and quantitative methods in linguistics and the ways to combine them (refer to Litosseliti 2018). For the basic principles of the experimental methodology see Gillioz and Zufferey (forthcoming).