Wednesday, December 14, 2005

The Problem With Meaning

A few years ago, during one of my rabid "embodiment" phases, I sat in on an AI discussion group comprised mostly of people who actually worked on AI. Being a mere cognitive psychologist, without a much knowledge about computers, I learned a lot, though I think most of the points and questions I raised were pretty stupid (I imagine some of the people in the group cringed every time I opened my mouth). However, there was one problem that I noticed when the AI people talked about "meaning," and no one in the group gave me what I felt was a satisfactory answer to it, so it's bugged me off and on ever since. The way I saw the problem back then was that in certain cases, it's necessary to understand the affordances of objects in order to understand sentences, and understanding affordances requires having a body. I'm no longer so certain that you need a body to understand the affordances of objects, but what you would need in place of that would be such a large body of interconnected knowledge, along with a wealth of "theories" about how to use that knowledge, that building an artificial system capable of comprehending some very simple sentences would be near impossible.

To illustrate this problem, consider two sentences.
(1) John kicked the ball to Mary.
(2) John crutched the ball to Mary.
The syntactic structures are the same, and it would be easy to teach the syntax to a computer. Thus you could get the machine to understand the outcome of the event described in the sentence: Mary ends up with the ball, as a result of an action performed by John. But what if you asked the machine to describe how John got the ball to Mary? What if the computer's task was to describe John's action? It could probably do OK with sentence (1), because the action isn't a novel one, and having been programmed with the meaning of the word "kicked," all it would have to do is spit that meaning out. But sentence (2) contains a novel verb, one which the machine is unlikely to have in its lexicon (unless the programmer has read the paper from which I stole the verb, and is trying to get one past me). You and I shouldn't have much trouble figuring out John's action in (2), even without context (if we added a little context, such as a sentence preceding both (1) and (2) that read, "John was standing across the table from Mary, and the ball was on the table," figuring out (2) would be even easier for us). What would the computer need to describe John's action in (2)?

The answer is that the computer would have to understand what sorts of actions are possible with a crutch -- the affordances of crutches -- as well as being able to reason about which of those actions would be effective in performing the action described in the sentence. As I said, I used to think that this required having a body (because affordances are organism-specific, and even body-specific), and that may still be the case. But if it doesn't require having a body, and even if it does, you've got to have a whole heck of a lot of background knowledge and folk theories to get the affordances of a crutch for John, and pick the relevant ones for a given context. For example, you'd have to know that the hard end of a crutch can be used to apply force to other solid objects (i.e., to push them); you'd have to know that balls are generally light enough to be pushed by a crutch; you'd have to know that some ways of applying force will work in some situations, but not in others (e.g., John and Mary may be close to each other, or indoors, making swinging the crutch like a baseball bat to hit the ball over to Mary impractical, and even dangerous); if you knew that the ball was far enough away from John that he would be forced to fully extend his arm and utilize the full length of the crutch to get to the ball, you'd have to know that the mechanics of the situation would probably require John to hold onto the bottom of the crutch and use the top (the part that goes under your arm when walking with a crutch) to hit the ball, while if the ball were closer to John, it might be easier to use the crutch the other way around. The list of things could go on and on. And the machine would have to know things like this for every novel verb it came across (and novel verbs, particularly denominal verbs like "crutched," are pretty common in everyday language).

So that's the problem I saw back then, and still see today. In order to write a program that can understand the meaning of sentences with which you and I would have no trouble, you basically have to program in most or all of the knowledge of at least a well-developed human child, if not an adult. And I don't see how that's really possible. It certainly doesn't seem to be possible today, since we don't have a firm understanding of how people reason about the mechanics of situations like the one in (2), or how they activate the relevant background knowledge (the paper linked above gives one potential answer, in the form of the "Indexical Hypothesis"). If a machine doesn't have that level of knowledge, every time it gets a novel verb, it's going to be lost.

14 comments:

Speedkill said...

That point was made in the undergrad AI class I just took. The problem with a program understanding, say, a sentence, isn't parsing it out, but storing all the background knowledge we take for granted. Which of course leads to much amusement when playing with the chatbots on the web or reading transcripts of the Loebner Prize competition.

Anonymous said...

In order to write a program that can understand the meaning of sentences with which you and I would have no trouble, you basically have to program in most or all of the knowledge of at least a well-developed human child, if not an adult. And I don't see how that's really possible.

Of course this is not possible.
But this is the wrong approach and this is why you are doomed just like all AI researchers in the last 50 years!
Can you see where and how you err?

1) You assume to think/speak on behalf of a "perfect" listener who will always "get it right", neither you, me or anyone else does that, we do have lots of misundertanding and still succeed in communicating nonetheless.
2) You unconsciously assume some kind of "design" about the required internals of a speech understanding software. There must be "somewhere" some definition of the meaning of words and the software will have to "retrieve" those meanings and accordingly...
And "accordingly" what?

You just started building some prototype architecture without giving any thoughts about what you are actually doing and going with all those assumptions.
Then you find problems with this "design", but that's your own faulty thinking which creates these problems.
Go back and think again, specially, watch for unheeded assumptions.
To highlight what a different approach could look like, see Jeffrey Elman's "An alternative view of the mental lexicon"
http://crl.ucsd.edu/~elman/Papers/elman_tics_opinion_2004.pdf
Not that this is likely to be the "one and true" answer either, but at least it is more thought out.
For a brain refreshener I would also recommend Alex Gross's "Is Evidence Based Linguistics the Solution? Is Voodoo Linguistics the Problem?"
http://languag2.home.sprynet.com/f/evidence.htm

Chris said...

Anon, I'm not sure I made either of those 2 assumptions. First, I used as an example two sentences that are quite easy for the vast majority of English speakers to understand (a fact empirically confirmed in the paper I cited). I'm not demanding a perfect listner, only one that can do the things that we do easily. Second, I spent the bulk of this post describing a situation in which neither the human listner nor the computer has a definition of the word in a lexicon. Instead, the human and the computer would have to retrieve many different concepts (I assume its uncontroversial that humans have concepts, and Fodor not withstanding, that they have content). But thanks for the references. I'll check them out.

Jim Davies said...

Nice Blog. I write a lot about Cognitive Science too, if you want to check it out.

http://jimdavies.blogspot.com/

Anonymous said...

This sounds similar to a conclusion I came to and wrote about in my blogpost, It Takes an Agent to be Semantic. And I also agree with you about the difficulty of this but I feel that progress in this area is on an exponential curve. Jeff Hawkins, with his company Numenta, may come up with something powerfull enough to come closer than ever before. I include some links to his work and express this idea further in The Semantic Web Awaits Real Semantic Machinery

Anonymous said...

I generally don't talk to environmental scientists to find out what it's like to be a astronaut. Why would I talk to a computer to find out what it's like to be a human child with a busted leg? Any computer worth talking to would have insights that a human wouldn't, because their 'body' is different. Hopefully, such a computer would be curious about how we humans crutch balls, as well, and not wholly concerned with our complete destruction.

Fido the Yak said...

The discussion of affordances is fascinating. I question the statement made in Zhang's paper, "Affordances only make sense from a system point of view." To me it implies a reification, or making "system" do more than it needs to do. That is, if we allow that systems are real, empircally detectable phenomena, then we should be careful to distinguish a reference to that kind of system from the idea of a system as an explanatory model or useful all-purpose descriptor for a patterned assemblage of interrelations. I wouldn't want to actually describe something like an array of perceptual affordables without invoking the idea of a system, but neither would I want to make a commitment to its status before I had examined the matter.

If Gibson had in mind a particular kind of system, say an ecology of living organisms, but we take his use of "system" to be more generalizable, what will we have to account for if we apply his concept to phenomena lacking certain properties of the elements of an ecology?

What role does embodiment play in Gibson's system? Motility? Agency? Chris, perhaps you are correct that a body wouldn't be needed to understand affordances, but I think you will need something besides a large body of interconnected knowledge to account for the dynamic interplay between an organism and its environment. Isn't that the dynamic that makes Gibson's work interesting?

Anonymous said...

Rare GHD
Babyliss
polo boots
polo shoes
cheap straighteners
ghd hair straighteners
hair straighteners
Purple GHD IV Styler
Pure White GHD
Pure Black GHD
Pink GHD IV Styler
Kiss GHD Styler
Gray GHD IV Styler
Gold GHD IV Styler
GHD Rare Styler
GHD IV Salon Styler
Black GHD IV Styler
Benefit GHD IV Styler

Anonymous said...

herve leger
herve leger dresses
chanel
chanel iman
chanel outlet
chanel handbag
Chanel Wallet
Chanel Watch
chanel purse
Chanel Sunglasses
chanel bags
chanel earrings
chanel jewelry
chanel shoes

Unknown said...

i want to show you the following products:
gucci boots
gucci shoes
gucci t-shirt
gucci bags
gucci handbags
gucci clothing
gucci watch
gucci sunglasses
gucci jewelry
gucci boots
gucci shoes
gucci t-shirt
gucci bags
gucci handbags
gucci clothing
gucci watch
gucci sunglasses
gucci jewelry
gucci boots
gucci shoes
gucci t-shirt
gucci bags
gucci handbags
gucci clothing
gucci watch
gucci sunglasses
gucci jewelry

Unknown said...

ankh royalty

ankhroyalty

ankh royalty clothing

ankh royalty sweats

ankh royalty tracksuits
babyliss
babyliss pro
babyliss hair
babyliss i trim
babyliss flat iron
babyliss you curl
babyliss hair straighteners
babyliss hair straightener
babyliss straightener
babyliss straighteners
babyliss portability
babyliss straightening irons
babyliss hair iron
babyliss straightening iron

Unknown said...

air max,Nike air max
air max 90
nike air max 90
air max 1
air max 95
nike air max 360
nike air max ltd
air max ltd
air max 360
air max classic,Air Max Huarache
air max light
air max bw
air max plus
air max 180
air max nike
air max TN
air max 87
timberland
timberland boots
timberland shoes
cheap timberland boots
timberland uk
timberlands
timberland 2010
timberland outlet

月月 said...

jordan 12
jordan 13
jordan 14
jordan 15
jordan 16
jordan 17 rings
jordan 18
jordan 19

janewangleilei said...

jordan 1
jordan 2
jordan 3
jordan 4
jordan 5
jordan 6
jordan 7