Why Computational Linguistics?
I was thinking about what to say to a student who is contemplating a career in computational linguistics. How can I convey my enthusiasm? How can I explain my fascination with language? Here are some of the things that came to mind:
I believe that most of my knowledge of the world comes from reading and talking (that is, from language), rather than from direct experience (from doing and sensing). I know that this belief is not very scientific (we need to figure out how to measure the amount of knowledge from linguistic and non-linguistic sources), but I am inclined to believe it until science says otherwise. So I am curious about language because I believe it is my main source of knowledge of the world.
Like most people, I have a constant internal flow of words in my mind; I’m always talking to myself in my head. There is a meditation exercise in which one attempts to halt this internal discussion. I’ve tried this exercise occasionally, and I find it extremely difficult. On the rare occasions that I have been able to halt the internal flow of words, I find that I am immersed in the present; I lose all sense of the past and the future. I hypothesize that language is what gives us our sense of the past and the future. I think that language is what distinguishes human consciousness from the consciousness of an animal, such as a dog. Again, this belief is not very scientific, but it is part of what attracts me to the study of language.
Language is a digital encoding of (what appears to introspection as) analog experience. This encoding is powerful for data compression and pattern matching, but any analog-to-digital conversion introduces aliasing noise. I’m very interested in how using language as a tool for thought distorts our perception of the world.
The Sapir-Whorf hypothesis is that language can channel our thoughts in certain ways. New words give us new ways of thinking. Consider how the evolution of mathematical notation has facilitated mathematical thinking. For example, compare Roman numerals to Arabic numerals; compare Newton’s notation for calculus with Leibniz’s notation; compare modern algebraic notation with earlier notation. It is relatively easy to see the importance of notation for math, but, although it is less visible, I believe similar evolutionary advances can be found in non-mathematical language.
Language seems to give us a glimpse into how another person views the world. This is most striking when we read a novel by a great author, but it happens to a lesser degree in every conversation. Language is as close as we can get to other minds.
Language is a kind of living, evolving creature; a communal creation that (to some extent) transcends the individual speakers of the language. Historical linguistics has much in common with cladistics. I think it’s exciting to think of language as a new kind of life form.
Historical linguistics gives us an intriguing window into the past. Although there is no written record of Proto-Indo-European, linguists have been able to reconstruct much of the language. By reconstruction of the language of the Proto-Indo-Europeans, we have learned much about their daily lives.
This may explain my interest in language, but I haven’t yet mentioned computers. I believe that computers are a tool for keeping linguistics honest:
Science is a way of trying not to fool yourself. The first principle is that you must not fool yourself — and you are the easiest person to fool. So you have to be very careful about that. — Richard Feynman
This point also applies to cognitive psychology. Both linguistics and cognitive psychology are especially sensitive to self-fooling, due to the dangerous temptations of introspection. We think that introspection gives us an advantage in these sciences, but I believe it is often a disadvantage. If a linguistic or psychological theory can be implemented as a computer algorithm, and its performance can be evaluated on some realistic test, then we have evidence that we are not fooling ourselves.
Filed under: Computational Linguistics, Computer Science, Philosophy of Mind, Philosophy of Science | Tagged: language, Sapir-Whorf, meditation
I like the idea that language is a kind of analog-to-digital encoding. There is even more: language is a powerful compression “algorithm” in that you can express complex matters with just a few words. And language (broadly conceived) is what enables us to go beyond sensory experience, i.e. abstraction isn’t really possible without language.
But on the other hand I think that there isn’t a binary divide between the conscious experiences of a human and that of all animals there are; some animals may be capable of crude representations of their perceptions and actions and also have rudimentary capacities for memory, which may count as a predecessor for consciousness.
And related to that point is that I believe there are aspects of conscious experience which really are not quite accessible in language, like “what it is like” to move your fingers or what it feels like to hear some piece of music, read a poem etc. One can use words to have somebody else try to experience the same, nevertheless it may be impossible to get everything into words that makes up the phenomenal experience. So you could “compress” into several words how somebody might move their fingers and you may even try to describe how you yourself feel doing this, but they will have to try out the movement themselves before they really know what you are talking about.
There’s no doubt about the mechanics of language being fascinating. Not to mention the knowledge that it can convey.
But I’d argue that knowledge which is NOT mediated by language is even more interesting. Pure awareness - pure knowing rather than knowledge of propositions - is even more interesting, I think, and much harder to study too, let alone communicate about.
But if you’re a Wittgensteinian about language and what can be spoken of, you should be sympathetic to that idea, no?
I think of computational linguistics as tools extending my reach.
I can look at a 1, 2 or even a 100 entries that are supposed to represent the same thing. But with CL tools, I can look at one thousand of them and find 20 that do not match my expectations. Then, those twenty are the ones I need to understand and possibly account for.
Also, the basic tools used in CL are cropping up in other fields with all those web2.0 companies. I am talking collaborative bookmarking, interest profiling, movie recommendations, etc.