When I was studying for my MSc (in psychological research methods), I was fascinated by how irregular natural language is. Grammar based models of language didn't seem to me to give an elegant, sufficient-and-no-more-so account of how the brain handles language.
Plaut & Shallice's 1993 paper gives an excellent connectionist account of irregular noun learning, and how it may relate to forms of dyslexia. As part of my studies I recreated their model (with the kind help of Prof. Plaut and my long suffering project supervisor Dr M LeVoi) using different source data, derived directly from English word lists rather than synthetically created. The difference is functionally rather trivial, but I thought it made it more obvious that the effect was valid. More than anything else it gave me chance to get very deeply engrossed in their findings and the surrounding literature.
I've yet to retrieve (or rewrite) my original 'basic introduction to connectionist modelling' text, but I'll link that to this post when I do. In the meantime, here's an article by Prof. Plaut that explains it.
Back to language learning.
Anecdotally, people often claim that the best way of learning a language is by 'immersion'. If you're surrounded by it constantly, you pick it up very rapidly. Realistically, that's how children learn their milk language - we don't teach 2 year olds formal grammar, they just learn it gradually. Connectionist models give a really good account of that. I've written about how phrases lend themselves to that process here.
So far then:
- The models suggest that immersion/repeated presentation of linguistic units will lead to 'spontaneously' learning grammar.
- Learning grammar rules by rote and trying to apply them is tedious and demotivating.
- Attempting to translate every utterance from your native language into a foreign language with perfect grammar on the fly is going to be far too slow for meaningful conversation.
- Most of us speak informally with shortcuts, slang and incorrect grammar in our native languages.
This led me to think that the best way is just to be exposed to as much language as possible. Effectively, as many presentations of as wide a set of sample data as possible. So my approach to learning French is:
- Read as many books as possible
- Watch French TV
- Listen to French pop music
- Read the news in French (especially specialist news, to build up the technical vocabulary. I subscribe to various archery, beekeeping & computing websites)
- Read cartoons in French. I've learned some really quite startling phrases from Phiip. (Warning, frequently NSFW language)
- We also have formal lessons, but they point out how words I've already encountered fit together, rather than being an exercise in 'memorising rules'.
This seems to me a more natural way of learning, conforming more to how our brains actually work.