Stratificational Grammar

Chris Palmer wrote:

Okay, okay. What is Stratificational Grammar?

Invented by Syd Lamb while at Yale. He's now at Rice. The standard textbook is by Lockwood. Current power proponents are Adam Makkai and Jim Copeland. Syd and his troop, especially Makkai, were the prime movers in founding LACUS back in the "no TG, no publish" days.

Formally, Strat is a set of 3, 4 or 5 linked context-sensitive production systems. There's a bottom stratum for sounds and a top stratum for meanings, and some number of middle strata for capturing the relationships between the two. So each stratum by itself captures generalizations at its particular level (phonotactics, lexotactics, semotactics). When input is parsed, each stratum constrains the operation of the next stratum up. When output is generated, each stratum constrains the operation of the next stratum down.

The relevant formal theory has hardly been dealt with in linguistics, but it is very mature in computer science under the rubric of "syntax-directed translation", something that is in very common use for the construction of compilers and interpreters.

Unique in Strat is a graphical notation used to capture the relationships within and between the strata -- referred to lovingly as "spaghetti" or "Lamb-o-grams". You'll see the resemblance to pasta when you look at Lockwood's intro.

So lexemes from one set demand an input from another set of lexemes?!?

Token 1 : a preposition
needs
Token 2 : a noun

for example?

Not exactly. More like the following, for the tokens

the cat sat on the mat .

the

Okay, we're talking about a referent the speaker thinks I can identify.

cat

Uh huh, so it's a feline the speaker thinks I can identify. Oh, I know, he means Tiddles, his old tabby. We haven't been talking about Tiddles or any other animals, so I wonder what he wants to tell me about Tiddles.

sat

Ah. There was an instance of sitting some time in the past, which Tiddles was engaged in. So what? Tiddles' sitting doesn't seem particularly eventful. I wonder if there is some special attribute of this instance of sitting he's going to tell me about.

on

Thought so. Tiddles engaged previously in a sedentary act, and the speaker wants to tell me more about what it was that was underneath the cat at the time.

the

Well, whatever it was that was underneath Tiddles during that sedentary act, I reckon the speaker thinks I'll be able to identify it.

mat

Well, imagine that. If it was the mat (the speaker's jute mat at the back door, I'm sure he means) Tiddles was sitting on, then I

bet

Tiddles wanted to be let in. Maybe it'd started to rain, or she was hungry, or she was scared, or ...

.

Hmm. Maybe he's not going to say anything more about Tiddles' sitting on the mat. Maybe this would be a good time to ask him why he thinks Tiddles did such a thing.

As I said, it's a computational model. Being formal is an occupational hazard of computational models -- they're good for some kinds of natural language processing on the computer. They're probably not very good models of cognitive processing in wetware.

Copyright © 1997, Don Blaheta
Last updated: 2 Dec 1997