The Apple, Pear, Banana Theory

A Short Primer (aka Computational Linguistics) Finding linguistics too hard? Prefer the cosy comfort of cloudy equations? Prefer to let the computer do the work instead of trying to actually understand messy natural languages? Here is what you do -

1. Take one or several corpora.
2. Look for apples.
3. Look for bananas.
4. Look for pears.
5. Notice how apples aren't always followed by bananas or pears. Notice how there are many more apples than there are bananas, and how infrequent pears are.
6. Set this up as some fancy equation - not ANY old equation; it must be Bayesian.
7. Design some experiment to test the idea that people are sensitive to these statistics*.
8. Publish papers, give talks.
9. Move on to pineapples, artichokes and oranges. Relabel these apples, bananas and pears (abiding by the time-tested "what's-in-a-name" principle).

Example equation - p(apples|bananas) = [ p(bananas|apples) * p(apples) ] / p(bananas) Advantages - Easy to do. Both setting up the equations and the experiments can be quite complex and fulfilling. The Extended Version - Natura Morta (aka Fruits on tables) Not happy? Want to go further? Look at the table these fruits lie on. Is it wooden? Is it plastic? Recast your equations to take this 'context' into account. Look for context sensitive p(apples|bananas) equations - is this number larger on wooden as compared to plastic tables? Make the bold claim - given sufficient contexts, and sufficient fruit, everything will be understood.

---

NB: Remember, the ONLY job of linguists is to discover the fruit. Once that triviality is out of the way, the Apple, Pear, Banana Theory is what is important. *If you find that adult participants don't behave like the equation, either (a) you're missing some statistical property or (b) you're onto something interesting; please post as a comment to this post ;)