Grammar Verification

One of the most common exercises for a language student is the written composition (where practice of correct written expression is more important than the given topic). With the decline of hand-written submissions (even in primary schools), word processors have become a key location of language production and learning. Yet surprisingly, the multilingual orthographic and grammatical capabilities of common word processors (like MS Word or WordPerfect^?) are seldom used, particularly for languages other than English (open-source and online editors, such as OpenOffice^? or Writely.com, usually lack grammar checking capabilities). Rarer still is the use of stand-alone grammar checkers (such as Correcteur 101 or Anectode for French). This situation may be explained in part by the additional cost of some language modules and in part by the additional steps needed to install those modules, assuming the user is even aware that they exist). But mostly, the underuse of word processors is explained by the fact that the relevant technologies are almost never integrated directly into the instructional context. This is a shame, since the moment of production is so pivotal in effective language acquisition (catching mistakes as they happen rather than expecting students to thoroughly examine a teacher-corrected text).

“LePatron: French Writing Assistant” is a free, online tool (see LePatron^?.ca) designed specifically for learners of French as a second language. Its initial objectives were twofold: 1) to provide friendly, accessible explanations to students of common linguistic pitfalls (feedback from grammar checkers is usually aimed at native speakers and can be linguistically complex and difficult to comprehend), and 2) to help the instructor avoid repetitive corrections for common mistakes – in other words, a first line of defence for both instructor and student.

Co-developed by Terry Nadasdi and one of the authors (Sinclair), “LePatron” differs from most grammar tools in its pedagogical design: potential errors are flagged and explained, but not automatically or even easily corrected. The student must become an active participant in learning by manually correcting the text (rather than, say, right-clicking on a suggested edit, thereby circumventing the need to actually write the correct form). The feedback provided is intended to clearly explain a grammatical point, it is left to the student to apply the principle. In some cases, when the explanation is insufficient, an additional built-in page of explanation can be invoked, including some interactive exercises, and in other cases external resources are suggested (in particular Martin Beaudoin’s Pomme site (http://www.pomme.ualberta.ca/). These linked resources are an exemplary way of making the most of the web-based learning context.

As is LePatron^? can be a useful tool for the French language learner and instructor (the site currently receives over 20,000 hits per day from almost 100 different countries). We keep improving the site based on user feedback and by consulting the logs of over 190 million words in nearly one million texts (as of September 2006). One of the more interesting activities that can be done with students is to examine closely some of the strengths and weaknesses of the tool, and to speculate, linguistically, on why that might be. This emphasizes for students a fundamental reality: no grammar checker on the market today is perfect (far from it), for a variety of reasons, including the potential for syntactic complexity and semantic ambiguity. For instance, one of LePatron^?’s greatest weaknesses currently is its inability to deal intelligently with proper nouns. As such, almost all capitalized words are ignored (as potential proper nouns), even at the beginning of a sentence. Though this is a weakness of the tool, it is also an opportunity for discussion in the classroom about some characteristics of the French language.

Fill-in-the-Blanks for Dynamic Texts

A common drill in the toolbox of the language instructor is the fill-in-the-blank exercise. This type of exercise has easily made the transition from paper to the screen, often to great benefit. Creators of electronic fill-in-the-blank exercises are able to anticipate a variety of mistakes and have the computer provide immediate and meaningful feedback to the student – an obvious improvement over the print counterpart. However, despite excellent tools for creating electronic exercises (such as Hot Potatoes, see http://hotpot.uvic.ca/), electronic exercises take considerable time to develop, and involve hard-coded content (the same way that hypertextual links are almost always hard-coded in a document).

Since tools exist to perform syntactic and morphological analysis of texts (identifying parts of speech, like nouns and verbs, and canonical forms like the singular of a plural word), it should be possible to automatically generate fill-in-the-blank exercises, using any text on the web (in the appropriate language). This is precisely what the “HyperPoet: Linguistic Fill-in-the-Blanks” tool does, developed by one of the authors (Sinclair), see http://hyperpo.mcmaster.ca/LinguisticFillBlanks/ [this tool is ready, it will be available by publication time]. Users can point to a web address, upload a file, or paste content into a box, and the tool will use the specified options to dynamically create a fill-in-the-blanks exercise (in English, French, Italian, German, or Spanish). For instance, one page could be returned where the infinitive of all verbs is provided and the student must fill in the box with the correctly conjugated form. Alternatively, all prepositions could be replaced by blanks, and the student would need to provide the appropriate form.

One of the remarkable features about tools of this type is that students are able to generate useful exercises for themselves, they are not dependent solely on the instructor. Students can decide which type of text corresponds best to their level (e.g. newspaper articles, forum discussions, poetry), and generate as many exercises as they wish from authentic texts. Of course, such automatically generated exercises can also have disadvantages, such as misleading errors in the morphological analysis engine or the inability for instructors to provide contextual feedback (based on a specific incorrect answer). Still, much is gained in the use of text analysis on the rich corpus of authentic texts on the web.

Examine the Language Level of Texts

Several techniques have been developed by text analysis scholars for algorithmically quantifying the level of language of a text. Naturally, each technique has its particular strengths and weaknesses, and each one may be more applicable to different genres. Potentially more revealing than the results of any of these techniques, however, is an examination of the techniques themselves and the linguistic principles underlying them.

One readily accessible text analysis tool that can provide a variety of data on a text is Textalyzer (see http://textalyser.net/; many other tools exist online, including those found at http://tapor.ca/). Each step in the use of the Textalyzer tool is a pedagogical opportunity to better understand language analysis and methodologies. For instance, the interface on the first screen, as relatively simple as it is, contains terms such as “stoplist” and “polyword phrases”. Similarly, the results screen contains several terms that are worth examining closely, such as “lexical density” and the “Gunning-Fox Index”; a valuable exercise might be to have students research these terms on the web and compare their definitions. This article is not the appropriate venue to consider each one of the concepts and merits underlying the results of the Textalyzer [see…], suffice it to say that there is plenty of fodder for discussion of issues of language and its analysis.

We have found that students enjoy submitting their own texts to these types of analysis tools, where they discover aspects of their writing of which they were not aware (like a propensity for repeating a given phrase). An engaging activity can be to have students try to find texts on the web that most closely resemble the data profile of their own texts. Doing so can provoke interesting results and awaken the curiosity of the students for the relationship between text analysis and linguistic proficiency.