Artificiality
Posted on June 15, 2008
I guess it’s kind of funny that I saw this article over at Yahoo, because I’ve been thinking about artificial intelligence and language lately.
From what I understand, AI experiments with language have had problems getting computers to figure out the difference between “time flies like an arrow” and “fruit flies like a banana,” and similar puzzles. Giving them a vocabulary isn’t so hard, but getting them to realize the idiosyncratic rules is an issue. I wonder if having them diagram sentences and do conjugations would be helpful.
In thinking about formulaic language, some formulas are pretty much set in stone, while others are flexible. Actually, set in stone is a good example of a formula that’s set in stone. It doesn’t literally mean what it says, but most people will recognize it as a metaphor for something rigid. The phrase isn’t set in rock or set in concrete it’s set in stone. Other formulas are more basic. The basic sentence in English is subject+verb, such as Jane swims. However, it’s very rare for us to use sentences so simple, because they don’t convey a lot of information. We use articles, adjectives, prepositions, adverbs, clauses, and so on to get our ideas across. Still, many of those are defined by some basic patterns. Nouns can be bare, but are often preceded by adjectives or articles, so we have adjective+noun and article+adjective+noun as formulas. It’s elementary grammar, but it’s formulaic language nonetheless.
With the language problems for AI, I wonder if they don’t gve them the rules of grammar, or if they don’t tell them the functions of each word. This would be a problem, since some words can function as verbs, nouns, or adjectives. For example:
- I like to surf.
- The surf was great.
- It’s a surf shop.
In the first example, surf is an infinitive verb, a noun in the second, and an adjective in the last. It’s easy to see where a computer might get confused.
Another possibility is that the computers are given the rules and told the various functions of words, but they are bad at predicting – they’re not able to apply previous knowledge to new words. I don’t know for sure, but it seems reasonable.
Related posts:
Leave a Comment
If you would like to make a comment, please fill out the form below.
