Saturday 13 May 2006

rule of metaphor

metaphor (greek, from metapherein to transfer, from meta- + pherein - to bear
- a figure of speech in which a word or phrase literally denoting one kind of object or idea is used in place of another to suggest a likeness or analogy between them

prelude

because of a project i have been involved in past few months, i have been spending quite a bit of time rolling up my sleeves and getting down and dirty into computational linguistics and machine translation, or rather as i like to call it, machine-assited translation.

whenever i bring up this topic, people usually ask some or all of the following questions: 'where are the limits of translation? and what is it taking you guys so damn long?' interestingly, answers to all these question lie in the concepts of referentiality and metaphor.

self-referentiality

as it turns out that the more abstract - and thus imore self-referential in some sense i would argue - a body of text is, the easier it is for computer to understand and to translate.

this applies in other areas of human thought as well. for instance, computers have obviously beaten humans is chess and numerical computation, but for some reason we still do not seem to make progress in machine translation

what is going on? are we just complacently waiting and hoping that everybody is switching over to english or is the task simply insurmontable to the great human mind which can discover cures for complex disease and travel in space?

clearly, chess for one is completely a self-referential system with no connection to reality despite some of my grandmaster friends' attempt to argue otherwise. but how is our language different? where do the self-referential aspects of the language that make it translatable and understood by computers end and when does that something else, the intractable begin?

metaphor over machine

perhaps surprisingly, one part of our language that will always beyond reach of computers is metaphor, or any kind of figurative language in general for that matter

to convince yourself, imagine sitting at a cafe with a friend of yours, looking out of the window, noticing a big bus passing by and quipping: 'that slowly-moving, packed bus reminds me of my fat neighbor who also has a top speed of one mile an hour'

chances are that you and your companion understand the metaphor, perhaps even laugh at it, but when you think about it, a computer never will. for it will never have the faculties right there and then to perceive the bus and form the image between the bus and your neighbor.

and even if it had vision, it would run in trouble sooner or later with another metaphor based on the smell of coffee, and so forth. the best it can do is to translate the metaphor word by word, which as you might guess can be a risky business. thus, perhaps surprisingly, it is metaphor in particular and figurative speech in general that defines the limits of machine translation.

on the other hand, we can now also see that a repetitive, abstract and mechanical text such as technical or legal literature should be very easily translated. once you review computational linguistic research, this turns out be exactly the case.

origins of metaphor

metaphor is in fact a very fundamental concept extending even beyond our language. for no creature, however primitive, can survive very long unless it can deal with issues such as: 'is this the kind of situation where I eat this, escape from it, mate with it, look after it, or ignore it?'

since most situations in life do not come with neat labels that say 'eat me!' or 'escape from me!', this implies some kind of pattern recognition, and hence some kind of comparison: 'is this new situation that is emerging just now more like an 'edible' situation, or like a 'dangerous' situation'

for simple organisms, this kind of categorization may be little more than the ability to respond to a few chemical or physical triggers, but more complex organisms can make much 'cleverer' categorisations. for instance, the part of a frog's brain that analyses vision is organised in several layers. One responds to fixed patterns of light and shade – e.g. the fixed features of the frog's pond. Another responds to small, fast-moving, patterns of light and shade – e.g. flies that the frog eats.

another responds to large, slow-moving, patterns – e.g. larger animals that eat frogs. So frogs can 'compare' their views of a situation in terms of these three kinds of analysis specially evolved to meet the frog's key needs. as you go up the evolutionary tree, the pattern-handling gets cleverer and cleverer. another kind of 'comparison' that begins to appear in more complex animals is mimicry - e.g. young animals learn by mimicking older animals

these kinds of pre-human pattern recognition, categorization, comparison, mimicry, and such like are not 'metaphor' or 'analogy' in the human sense. so it is not far-fetched to suggest that human use of metaphor and analogy has evolved from pre-human capacities such as these, much transformed by being mediated through language

there is growing evidence of brain mechanisms specifically concerned with 'mirroring' what others are doing. we do so on in the first grade when the teacher draws the first characters on the white board and most people seem to keep the habit till they die. it has also been called 'keeping up with the jones'. or as oscar wilde put it:

'most people are other people. their thoughts someone else's opinions, their lives a mimicry, their passions and quotation' - oscar wilde, de profundis

metaphor, oration and power

being able to invoke powerful metaphors has usually been considered a valuable and powerful skill because a language rich in imagery tends to affect people emotionally. this on the other hand has served a clear purpose in both politics and literature. so it is now also clear to me how we could have reagan as president or, for instance, why i like al pacino movies.

to illustrate highly metaphorical language for a moment, take, for example, a poem by the great french poet charles baudelaire below:


even as she walks

even when she walks she seems to dance!
her garments writhe and glisten like long snakes
obedient to the rhythm of the wands
by which a fakir wakens them to grace

like both the desert and the desert sky
insensible to human suffering,
and the ocean's endless labyrinth
she shows her body with indifference

precious minerals from her polished eyes,
and in her strange symbolic nature where
angel and sphinx unite, where diamond,

gold, and steel dissolve into one light,
shines forever, useless as a star,
the sterile woman's icy majesty

charles baudelaire, even as she walks, les fleurs du mal


you probably notice at least a few metaphors, or if you majored in english in college, you will probably uncover a plethora from each stanza

epilogue

as it turns out, even though the potential number of metaphors is limitless, very few of us actually have a poet in us. moreover, since we like to be in the business of imitating each other, it should be fairly easy to come up with at least a basic catalogue of core metaphors. in fact, many people have done just that and captured many of the richness of our metaphorical imagery in idiom and aforism dictionaries. thus, if one ever hopes to facilitate the translation of any kind of human language - including metaphorical - that is where we should look for final answers

"there is nothing in our experience from the world at very small - the quantum world - or from the world at very large - the cosmic distances where relativity theory applies - that would help us to build metaphors for understanding those worlds beyond purely mathematical terms"
- albert einstein, when asked why physicists do not find most results in modern physics rewarding although they are mathematically absolutely correct

No comments:

Post a Comment