Correction software does not correct on its own!

To read the comments on my last three columns — “Writing in the XXI^and century”, “Who’s afraid of Antidote in CEGEP? » and « Let’s rethink the school to better teach French » — I find that many readers do not fully understand how writing assistance software works. Those who dispute their relevance as a pedagogical tool in the teaching of French mistakenly assume that these software programs do everything by themselves. However, they do not correct anything: they give the user the tools to do it well.

Moreover, these tools — various dictionaries and grammars — make these software programs much more than proofreaders. The potential of these tools is such that teachers who have mastered them (and who are adequately trained to do so) could use them to develop all sorts of exercises, games and methods for teaching the language.

In fact, what is most fascinating about writing assistance software is their real multiplier potential: well integrated into teaching, they will not be a crutch for students, but on the contrary a springboard for a whole generation of young Quebecers by allowing them to do certain things with the language that are rather difficult with traditional tools.

I have to admit here that the two software I know best are Word’s proofreader and Antidote. I use them daily, but there are others like ProLexis, Cordial and Le Robert Correcteur, each of which has its own particularities. Antidote offers some modules in English, while ProLexis is appropriate for long texts.

There are also other online resources — I am thinking in particular of the dictionary Usito and translation services like Reverso or DeepL.

What follows is very much based on my experience of the only Quebec product (Antidote), which is widely considered to be one of the best in the Francophonie. It is also the one whose design I know best, having done a major report on this subject in 2016. I therefore confess my Quebecois bias here, while recognizing that other software can do the job or be superior depending on the context.

And the proofreader was

The designers of Antidote were among the first to understand, 30 years ago, that writing assistance software should not “correct”, but “report”, because the language is too subtle. Even today, when artificial intelligence is much more developed, it remains futile to try to correct. However, it is precisely because the computer signals instead of correcting that it becomes useful for teaching, since it does not replace the person who writes.

The designer of Antidote, Éric Brunelle, explained to me recently that his inspiration was his French teacher in Tunisia, who marked the texts with different colors. Antidote does the same: errors are underlined with thick red lines, while other colors and types of underlining note what is inappropriate, to be checked or illogical. The user must make the right choice.

If I write “horses”, a big red line appears under the word. When I point to the word, a box suggests « horses », but sometimes two or three possibilities for more complex cases. The system can explain the rule to me immediately and I have the opportunity to correct or ignore the recommendation (like in this column, where I am forced to maintain « horses », because I include this fault intentionally). If I go wrong, I can go back and finally opt for « horses ».

We therefore understand here that a student who has no idea of ​​the rules of French will not be able to make the right choices. Conversely, if he has already learned them, he will be able to make informed decisions.

This is exactly what we observed during the uniform French test for CEGEPs in June 2021. Normally, the passing rate for this exam is 80% and this is largely due to the fact that the performance of students in spelling and grammar is very low. However, in June 2021, they had for the first and only time the right to do so with their computer (and therefore the self-correction software). And there, the success rate rose to 96%. This proves that the students know the rules better than you think, otherwise they would have been unable to improve.

A good capacity for analysis is all the more essential to make good use of such software as it occasionally signals errors that are not errors (so-called « false positives »). ), or mistakes in taste that are intended. For example, if I quote any sentence from sisters-in-law of Michel Tremblay, the software will find a fault. It is a case of intention that a software cannot know and which requires the judgment of the user. Sometimes, too, he will give me a whole series of warnings to tell me that if I write « lac », I must not confuse it with homonyms like « lacquer ».

This is a setting issue. To work properly, writing aids need you to introduce yourself and explain what you need. Antidote, among other things, first wants to know your mother tongue. If you write « English », for example, he will pay particular attention to the mistakes typical of an English speaker writing in French.

The system settings include a series of pre-established preferences that you can adjust: your ability in French, your linguistic region, your tolerance for regionalisms, the language level. You must also specify your gender (to allow the system to properly match your « I, me, me, us »). He also wants to know how to deal with inappropriate or offensive terms, anglicisms, lexical confusion (homonyms, inappropriateness, rare words). So if the disambiguation warnings annoy you, you just have to tell him not to show any more.

In addition to language, you can direct the corrector to typography (spaces, types of quotes, etc.) and style. This last aspect concerns the repetitions of words, impersonal turns of phrase, dull verbs, the degree of readability of the text, its inclusiveness.

In this regard, a teacher wrote to me recently to tell me that she had begun to introduce writing assistance software to her secondary three students. She explained to me that this software was an excellent tool for analyzing how one can make a text less negative, or remove its passive turn. This is possible thanks to the analysis capabilities of the algorithms.

For my part, I rarely had to use the “style” function. Except that this year, I’m going to step out of my comfort zone and write a script for a documentary series project. As this will be producing texts intended to be spoken rather than read, I’m likely to find the stylistic analysis function (words per sentence, word length) devilishly useful. But that will be for me to judge. Because the software decides nothing.

The potential of dictionaries and guides

I have only touched on the tape the interest of the other tools of a writing assistance software, which is in the guides and dictionaries of Antidote. On this level, I specify that these tools can also be used independently of the correction software.

I have been using Antidote for 20 years as an autodidact and I only received a two-hour training as part of a congress of journalists, in 2018. During this training, one of the participants had made a little laugh of her, because she used Antidote strictly for her dictionaries, she was unaware of even its correction functions. I took care not to make fun of her, because in 20 years, I myself had never considered dictionaries!

Writing assistance software like Antidote — just like the other main software of its kind — includes several reference works.

Let’s go beyond the classics such as the dictionaries of definitions, synonyms, antonyms, conjugation, the Visual, the historical dictionary, the dictionary of quotations and that of word families. Lately, I discovered two other types of dictionaries that I didn’t know at all (co-occurrences and lexical field), which are of great interest.

A co-occurrence is one word associated with another. For example, if I say « dog », it is associated with « rabid, mad, shepherd, tracker ». Antidote’s designers sampled tens of thousands of texts to build a gigantic dictionary of 900,000 combinations.

Is a “short” horse? Can it be « mangy »? For the word « horse », the dictionary of co-occurrences will give you 134 associated adjectives (small, fiery, saddled, speckled, crippled), 15 classifying adjectives (chestnut, gelding, thoroughbred), 56 complement nouns (of battle, of wood , rocking), 58 verbs with “horse” as subject (galloping, rearing up, kicking, dragging) and hundreds with the word as a complement (harnessing the horse, riding on its high horses).

The other dictionary that I find remarkable is that of the lexical field. It is in this work that we will find that the adjective corresponding to « horse » can be « equestrian », « equestrian » or « equine » depending on the context. But there are 60 others, and Antidote also gives 377 nouns associated with the horse (jockey, hoof, dung, harness, bucephalus, appaloosa), 37 verbs (piaffe, spur, strip) and 16 proper nouns (Troie, Jolly Jumper, Gericault).

Good writing assistance software also allows you to build your own dictionary to include neologisms. The software already knows a bunch of proper names, first names, place names, but not all of them, so we can add more. It is also possible to enter nouns, adjectives and all classes of words. For my book on the French, I made him learn the interjection “râlâlâ”. For other texts, I added the verbs “renmieuter” and “aplaventrir”. I told him that they were conjugated like « love » and « finish » respectively, and since then he organizes himself. If I write « je renmieutent » or « il s’aplaventris », he will point out the error (« je renmieute » and « il s’aplaventrit »).

I am not a pedagogue, but the didactic interest of this kind of software is obvious. Consider only the query function. If I look for a rhyme with « horse », I get 2,228 words in -al, 82 words in -val and 34 in -eval (Perceval, medieval). But he can find alliteration, that is, words that start like « horse ». There are 129 (chevronner, goat, and even the proper name Chevènement).

In my opinion, the main pedagogical interest of the dictionary function of software is to open all the reference works on the same page. When I was young, if I wanted to know the definition of « horse », its synonyms, its antonyms, the quotations with this word, its etymology, its translation, its grammatical use, or even to see illustrations, I had to consult four , six or eight works. There, I click on « horse » and everything appears. And wherever you are, just click on the word that interests you to move on to the next one. For example, for « horse », the lexical field gives me the adjective « isabelle ». Sorry ? I click on “Isabelle” and the definition tells me “pale yellow”. It opens up perspectives of knowledge, but also didactics.

This is why I think that all those who are opposed to the idea of ​​integrating this kind of technology into the teaching of French in CEGEP would benefit from trying it before getting on their high horse.