English grammar?

  • Thread starter tgt
  • Start date
  • #26
apeiron
Gold Member
2,013
1
If you are using set theoretic descriptions - all those brackets - then you are using nested hierarchies perhaps without realising it. And this is the general aspect of Chomsky I would find correct (he was rather weird and Platonic about things such as the evolution of language).

BTW the brain itself is hierarchically structured in its generation of actions and language just reflects that. What makes language different is that output is serially constrained (we have to make choices about which noise to make first). But discussion of that leads into truly complicated neurolinguistics and systems principles.
 
  • #27
2,123
79
If you are using set theoretic descriptions - all those brackets - then you are using nested hierarchies perhaps without realising it. And this is the general aspect of Chomsky I would find correct (he was rather weird and Platonic about things such as the evolution of language).

BTW the brain itself is hierarchically structured in its generation of actions and language just reflects that. What makes language different is that output is serially constrained (we have to make choices about which noise to make first). But discussion of that leads into truly complicated neurolinguistics and systems principles.
OK but I'm not claiming (nor do AI workers) that the predicate calculus is a model for how language is generated. Intuitively nouns and verbs would seem to be at the most basic level of language. Does that mean they are at the top or at the bottom of the hierarchy? The predicate calculus places them at the bottom and this is reflected in some nesting patterns (adverbs are outside the parenthesis). Howeover, our concepts of parts of speech don't always hold when analyzing languages that a very different from the IE model. In Western Shoshone dialects of the western US, "The chicken ran across the road" would be literally translated as "Running chicken goes to (other side by) the road." "Other side by" is indicated by particle which can be loosely translated "a place further from me" than some point of reference, in this case, the road. It can be seen that here "running chicken" is nested inside the parenthesis whereas in English, "run" would be outside the parenthesis. GOES TO (RUNNING(CHICKEN), PLACE1 BY(ROAD1) PLACE1 ACROSS TO (ROAD1)). Note, the Shoshone contains more information than the English since the necessary construction will tell us if the chicken is still by the road (so it can still be caught and served for dinner) or has moved on to freedom.
 
Last edited:
  • #28
apeiron
Gold Member
2,013
1
If you are coming at this from a computer science perspective then hierarchies will be a problem because computers do hierarchies very badly. They are anti-hierarchy systems pretty much!

You can find a thread of research which runs through Ashby, McKay and Grossberg which would be as good as it gets when comes to getting hierarchical "computation".

But take your question about nouns and verbs. First I would point out that this is a divide that results from a standard brain dichotomy (and dichotomies are the path to hierarchies).

So we have a world that is divided into objects and actions, locations and motions. We have a whole - some situation - and first must come the simple divide. There is a chicken. There is its running. We can now construct a proto-hierarchy out of this. Something vaguely specified (the whole that is "an object acted") becomes specifiable as two sub-components (what object? what kind of action?)

Then through hierarchies (via further dichotomies) we can increasingly specify (or qualify) this initial thought.

So the chicken is placed within the realm of objects as (for instance), [material [living [feathered [farmbird [chicken]]]]. Note the dichotomous distinctions (and the more dichotomous, the more meaningful we find them).

We are also saying [not-immaterial [not abiotic [not furred [not wild [not a duck, goose, etc].

This is one of the major things absent in most computational approaches of course. Only positive information is represented (what something is) and not at the same time all the things it is not (which is how brains do it - the idea of chicken involves not just rousing chicken circuits so to speak, but also suppressing goose and duck circuits).

As to the Shoshone example, there is no grammar that contradicts the general Chomskian story. Some languages may be better at sharply expressing some kinds of thought, but like the numbers of names Eskimoes have for snow, that tends to get exaggerated.

The wide variety of serial utterance arrangements is why serial computers struggle with human language comprehension. But hierarchically operating human brains don't.
 
  • #29
2,123
79
As to the Shoshone example, there is no grammar that contradicts the general Chomskian story. Some languages may be better at sharply expressing some kinds of thought, but like the numbers of names Eskimoes have for snow, that tends to get exaggerated.

The wide variety of serial utterance arrangements is why serial computers struggle with human language comprehension. But hierarchically operating human brains don't.
I certainly agree that there is something like a universal grammar for human language. I've worked with AI people in developing "intelligent" programs for medical applications. The idea was to get the computer to "think" like a physician. These programs are moderately successful and are improving all the time. However, we all understood that that the use of the predicate calculus was justified based on its utility, not because it was model for how the human brain actually processes language. In other words, we wanted to be clear that we were not making claims that the predicate calculus is the best possible model for a universal grammar. The predicate calculus can deal with the "shifting" hierarchies that occur between specific grammars and semantic structures, but the individual programs are language specific.
 
Last edited:
  • #30
2,123
79
The wide variety of serial utterance arrangements is why serial computers struggle with human language comprehension. But hierarchically operating human brains don't.
I missed this comment. It would that seem that grammars that make heavy use of morphology would be more amenable to serial processing. There is considerable freedom in how to arrange words in a sentence in highly inflected languages which would suggest that all languages are not equal in terms of the modeling grammars in computer programs. The difficulties with semantic structures also depend on how well developed classification trees are in a given language. Some languages are 'flatter' than others.
 
Last edited:
  • #31
apeiron
Gold Member
2,013
1
You mean humans could make it easier for computers by adopting grammar with built-in pointers that direct back to the hierarchy that generated them - and then stuck to those rules :-)

I spent a few years with AI types in the early 70s. Speech recognition was one of the best funded areas (all those military intelligence applications!). That was when the gap between AI and real I became clear to me.
 
  • #32
mgb_phys
Science Advisor
Homework Helper
7,774
12
You mean humans could make it easier for computers by adopting grammar with built-in pointers that direct back to the hierarchy that generated them - and then stuck to those rules :-)
You mean like Germans?
 
  • #33
2,123
79
I spent a few years with AI types in the early 70s. Speech recognition was one of the best funded areas (all those military intelligence applications!). That was when the gap between AI and real I became clear to me.
Like I said, the AI types I worked with clearly recognized that AI was not intended to model the human brain. Do you know what "real I" is?
 
  • #34
apeiron
Gold Member
2,013
1
Do you know what "real I" is?
Yes. And that's why I switched to hierarchy theory, dissipative structure and systems science approaches to the questions that interested me.

And the GOFAI dreamers still exist. Check out the enthusiasm for Kurzweil's singularity for instance.
 

Related Threads on English grammar?

  • Last Post
Replies
21
Views
4K
  • Last Post
Replies
17
Views
4K
  • Last Post
2
Replies
28
Views
4K
  • Last Post
2
Replies
45
Views
17K
  • Last Post
Replies
6
Views
1K
  • Last Post
3
Replies
50
Views
7K
  • Last Post
Replies
9
Views
3K
Replies
34
Views
4K
Replies
35
Views
6K
  • Last Post
Replies
11
Views
2K
Top