From the lab: how artificial grammar systems can help identify speech and language impairments

Dr. Diego Gabriel Krivochen works at the University of Reading on artificial grammars, as part of a group led by Prof. Douglas Saddy. We chatted to him about how they are currently used and how their future development could revolutionise diagnosis of speech and language impairments.

1.      What is an artificial grammar system?

Well, in the simplest sense it is just a set of symbols accompanied by rules that tell you what to do with those symbols; these rules have the form ‘if you find symbol X, replace it with Y’ (which can be a sequence of symbols). The result of applying the rules over and over again is a sequence of symbols, what we call a ‘string’.

2.       How could artificial grammar systems help practitioners identify speech and language impairments?

Although they don’t look the same at a glance, artificial grammar systems are developed to have key features in common with natural grammar in languages, that are processed in similar ways. So, for example, the ‘rules’ we observe in grammar that allow us to express common meaning and have common understanding, have common features with our artificial grammar systems. However, it’s important to remember that there are many factors involved in natural language comprehension (intonation, word meaning, etc.) so it’s not easy to base it entirely on structure.

A huge problem in the development of tools to identify language disorders is they are almost always specific to one language. As a result, multilingual children can sometimes appear to be language delayed compared to their monolingual peers. But given that language disorders are as common in monolingual as in multilingual children, there is a double jeopardy: a typically developing multilingual child may be incorrectly diagnosed due to perceived language delays and sent for treatment, but also a multilingual child with a developmental condition may be overlooked, assuming that the delay relates to their normal multilingual development.

Incorrectly diagnosing children as having or not having a speech and language impairment is both costly and can be detrimental to the child’s development. This is where artificial grammars become crucial: they don’t tap into the child’s understanding of one specific spoken language, but rather they allow us to probe into the underlying properties as they are processed in the brain.

3.       What research projects have you been involved in that have used them?

As part of the EU-funded Advancing the European Multilingual Experience (AThEME) project, we worked with a team led by Prof. Denis Delfitto at the University of Verona in a series of experiments which investigated when tasks involving language (natural or artificial) reach the point where speech and language impairments hinder performance. As far as artificial grammars go, we modified a task based on blue and red dots appearing on the left and right sides of a screen in a sequence which depends on the rules of an artificial grammar. We asked the participant to press a different button (left-right) depending on whether the circle they see is red or blue. So far, so good. However, within this ‘string’ of red and blue dots there were several hidden layers of regularities which arise due to the grammar we used: a red is always followed by a blue, and two blues are always followed by a red. If the participant is learning as they carry out the task, they will progressively get faster and more accurate in their responses, as they unconsciously learn the structure ruling these hidden regularities.

The research with Verona involved testing monolingual and bilingual children, with and without a diagnosis of developmental dyslexia. The results were fascinating. Bilingual dyslexics performed consistently better than monolingual dyslexics and that in some cases they reached the speed and accuracy of unimpaired children on all but the most complex conditions. On the other hand, monolingual children with developmental dyslexia, were less accurate and in some cases also slower when picking the regularities in the artificial grammar systems. These results suggest not only that bilingualism does not cause any additional problems for children with developmental dyslexia, but that there may be some underlying benefits in terms of processing for children who speak more than one language who have developmental dyslexia.

 4.       What do you think the future holds for artificial grammar systems?

We are currently working with researchers at the University of Geneva to expand on all this and improve accuracy. We want to explore just how complex the tasks can get and how people react. We’d also like to combine on-screen visual testing with some kind of auditory aspect. The more we can get artificial grammar systems to mimic the processing of natural grammar systems in the brain, the easier it will be to develop tools for professionals to identify and diagnose speech and language impairments in bilinguals – regardless of the languages spoken.

If you enjoyed this post and would like to read about artificial grammar in more technical detail, please click here for the full interview (pdf). You can also watch videos below of the researchers discussing their findings as part of the AThEME project, and read the full EU policy brief from the AThEME project on multilingualism and communicative impairment.