 |
|
|
 |
|
|
|
|
MightyMaus |
| | | | | | posted 10/4/2001 21:28 |   | | | | | | | | | Alright, I was recently watching a marathon of Star Trek: The Next Generation, and one of hte episodes seemed like a good basis for a language discussion.
In this episode, entitled "Darmok", The crew of the Enterprise-D makes first contact with a species known as the Tamarians. Using the Universal Translator system, they're able to comprehend the words that the Tamarians are saying... but are unable to interpret them.
When they speak, the translators were able to take the words, such as "Darmok on the ocean", "Temba, his arms open." or "Shaka, when the walls fell", but were unable to scan the meaning behind these statements. Picard and the Tamarian captain are inexplicably beamed down to a planet by the Tamarians to fight some beast.
The crew of the Enterprise-D and Picard figure out at about the same time that the Tamarians communicate exclusively in metaphor, citing old legends, etc. They have little to no sense of Ego as humans know it.
The question is: would a Tamarian-based language system be a stepping stone to teach computers how to comprehend meaning? Or are we too far away from being able to do this? Could we load the text of, say, Beowulf into a computer, and expect it to know that "Beowulf and Grendel ar Heorod" means two adversaries in mortal combat or that "King Hrothgar, his cup full" means celebration? Are these abstractions a step between the parse-and-paste method computers use today and the analysis of meaning that computers of the future may be able to do? I'm not sure where you guys'll go with this; I don't even know where /I'm/ going with it. :) But it seemed an appropriate topic.
Well, until next time,
Raskolnikov, leaving St. Petersburg. (trans: Goodbye)
--Jay
| | |
| | |
Jean |
| | | | | | posted 10/6/2001 21:35 | | | | | | | | | | Jay, you may wish to check the thread "When gorillas go mad they go mad" in this forum (the link is at the bottom).
You are asking about a computer's ability to discern the meaning of metaphor through understanding the text on which that metaphor is based. But a prior requirement is for the computer to *understand* in the first place. And that is a problem that AI is struggling with, using different approaches and methods. Before we hand the computer the full text of Beowulf, we need to pass the stage where a computer can understand a simple sentence, like "the rain in Spain falls mainly in the plains". Now what would constitute understanding that? Perhaps the ability to conclude that less rain falls in the Spanish mountains, or the ability to use this information in a larger framework.
This is understanding in a basic level. Metaphors (as well as anaphoras, idioms, allusions etc.) stretch that ability even further. How do humans percieve metaphors? Metaphors are either learned (you read "Lolita" and you derive from the text the metaphorical meaning of that term) or you never read the book but someone tells you what the meaning is.
In the case of a computer, the first technique requires understanding and the ability to deduce conclusions based on that understanding. This computers still can't do. The second technique is what would bem called in computer terminology "hard wiring". You can feed a list of metaphors and their meanings to a computer - but that is not saying much.
As far as I know, what you are suggesting is not yet possible. If it would be, I think it would constitute real artificial intelligence.
| | |  A thread worth reading | | |
| | |
hillel roman
[Guest] |
| | | | | | posted 10/21/2001 12:59 | | | | | | | | | | I think the real problem is still understanding "understanding". Let us (again) consider Witgenstein's concept of meaning as use, which Jean seems to have ignored - For example the sentence "the rain in spain..." for most humans would say nothing of the weather in Spain but rather suggest a word game, or a song (even assuming they are not familiar with "my fair Lady"), and the expected reaction would not be "really? I thought the rain in spain falls mainly in the mountains", rather we would expect the person to join in and sing with us. Jean assumes a layered model of language where some sort of litteral or basic meaning lies at the bottom, a model which has many times been shown to be not only false but also innefective in constituting what is reffered to as Strong AI. Language is never isolated from the total living experience, so is the knowledge of folk tails, literary pieces, current events, and world knowledge. Without use, i.e. a connection between utterances and actions (not only speach ones) language does not only loose its meaning, we also loose our ability to appreciate lingual performance.
| | |
| | |
yaki |
| | | | | | posted 10/22/2001 10:17 | | | | | | | | | | Wow, Hillel. I couldn't have said it better myself. Let me just add a word about metaphors: A metaphor is nothing more than a new way to use words. Once it's not new anymore, it's no longer considered a metaphor. (If it's origins can still be traced, it is sometimes called a "dead metaphor"). An example: A table typically has 4 LEGS. A leg of a table is not considered a metaphor, but there was a time when it was. ("Leg" was originally used strictly as a body organ. Later the term was "borrowed" and used for inanimate objects such as furniture).
| | |
| | |
Spydre |
| | | | | | posted 3/12/2007 07:55 | | | | | | | | | | This is a topic that has bumbled about in my head for a while. The thought of what it is that lies at the heart of language and meaning.
Language is, at its root a way for an individual to broadcast a status of circumstance to another individual. Whether it be trees in a forest communicating to other plant life surrounding them of a deadly infestation by chemical means alerting all to increase their biological defenses in advance; insects communicating the location, quality, and quantity of a newly located food source; or the advanced quality of humans to communicate their individual thoughts (no matter the complexity) to other humans. The key is not why this is done, but how.
In our Star Trek example what is lacking between the Tamarians and the Enterprise crew is any form of context in which to frame what is being expressed. It is like viewing an ancient pictograph writing system that conveys an event that once occurred, which to a member of the group that drew the pictographs, whose meaning is formed from the pictures' context to current social culture and environment. It is little more than a mnemonic device, and this is the basic development that all written languages transitioned from if their development was not influenced by the exposure to a more advanced writing system. The problem is, once the context for the pictures is lost, through either significant change, or loss, of either culture or environment, the meaning of the pictures is lost as well.
The major revolution in writing systems from pictures and pictographs seems to have evolved to deal primarily with the complexities involved in keeping track of property ownership. Simple images representing individuals and their property is fine for basic tabulations, but once complex transactions of property and wealth begin to take place, this method is not up to the task. When two parties enter into an agreement to charge interest on a loan with consequences for lack of repayment of the loan on time, more than word of mouth and the veracity of eyewitnesses is going to be required to maintain order, and mere pictures aren't going to represent that Bob loaned Bill $10 last Tuesday under the agreement that Bill repays the $10 by next Wednesday at a cost of $2 and an additional 50 cents for each day the loan went unpaid. Such initial difficulties in cultures without a system of written language may even have led to cultural and religious taboos against lending money to people within the same society as a means to promote harmony within the group. But large systems of government eventually need a means to collect fair and equitable taxation, and this promoted a break through in writing.
The break through, of course, is in making the written language undergo the same revolution that spoken human language had already found necessary--syntax. A way of insuring that, no matter who read what was written, the process and order of the symbols recreated the exact process and order of the thoughts and ideas of the original author; whether they are read the day transcribed or years later. There are several different ways this could be, and was, achieved in written languages--and spoken languages were no different. At some point simply commenting about a leopard and having everybody in the social group panic because they did not know if you meant that you saw a leopard now, or were commenting about a leopard you saw earlier, or, perhaps were just thinking the thought of a leopard, became not as useful as a linguistic tool which would enable you to link your spoken words in time and space relation the same way that your thoughts move and track linearly through time and space.
Thus, every modern human language has some means to indicate syntax (the specific ordering of speech in a consistent pattern to convey the temporal and spatial interactivity of spoken thoughts and ideas). Some languages achieve it solely through word order, others through sophisticated usage of prefixes and suffixes and conjugations--but they all have it. And while there is some discussion as to whether this ability to express our thoughts in an arbitrary but orderly fashion within societal groups for advanced communication of thoughts is a feature entirely hard coded into our brains, or is instead a skill that must be learned, the anecdotal evidence would seem to indicate the latter. Circumstantial findings are that if humans at some point do not learn syntax before late adolescence, it cannot be learned at all. It also does not seem that apes or other animals (dolphins, etc.) that have been taught symbol or sign languages (no matter at what age) ever learn the skill of syntactical language. Instead of "Shaka, when the walls fell," you get "Shaka walls fell" or "walls fell Shaka" "fell Shaka walls" and so forth. Did the walls fall on Shaka, did Shaka make the walls fall, or just what did or is happening? Without even the simple syntax of the wording "when the walls fell," indicating that Shaka was there when it happened, it is hard to glean anything meaningful from the words in the phrase (which would seem to indicate our Tamarians once had a syntactical language that has long since passed into disuse because of...? a preprogrammed collective ancestral memory at birth? or who knows. It makes an interesting query though as to how many anecdotes they would have to have stored away in order say everything that they wanted to say in a meaningful manner. Imagine trying to say in Tamarian, "I went to work today and forgot my lunch; so I went to the cafeteria to buy a ham sandwich, but all they had was turkey.")
So, if writing evolved to communicate the way we use spoken language, and if the way we use spoken language evolved to communicate the way we think, what does this say about language? That language as we know it may be arbitrary to the human brains that use it. Seeing as we have yet to fathom how conscious thought is actually produced in our brains, this would easily explain why there are still some features of grammar hotly debated to this very day, and why most discussions of linguistics quickly veer off into the realm of philosophy. Much like a program written to operate using the Mac OS is not going to run on Windows, human language is a function only useful to minds that think in the same manner as ours (or at least the same as those parts of our brains that process language); in which case we'd be only so lucky as to find aliens whose language were only as difficult to understand as the Tamarians.
And here is the point where most people who follow my supposition on human language throw up their hands and say that if I'm right then there is no hope of producing a computer program that can use natural language. However, I would beg to differ though, and offer that there are, as I see it, at least two distinct possibilities for making it happen. One would be the approach taken by A-i Research wherein a program is imprinted with an algorithmically recorded human mind--a complete, preconstructed, human personality, either real or imaginary; the recording process could be something as simple as the HAL training process, or something vastly more high tech, like a direct brain feed. The other method is more difficult, and would require the construction of an operating system that functions or mimics that part (or parts) of the human brain that perform thought and language. And while the latter requires a great deal of in depth biological and neurological study, innovative programming, and perhaps development of currently non-existing hardware and software; the former only needs sufficient memory and processor power, and a clever way to quickly allow linguistic input to generate the most appropriate output of a given personality.
| | | Last edited by Spydre @ 3/12/2007 6:19:00 PM | | |
| | |
yaki |
| | | | | | posted 3/12/2007 09:47 | | | | | | | | | | I couldn't have said it better myself (and why should I?)
| | |
|
|
|
|
|
