Computerized haiku - Margaret Masterman (1971)

Cybernetic Serendipity展 (1971年) で発表された俳句生成システム

'. . . playing with any new technique is the first stage of handling it seriously and really exploiting it.' (Masterman)

A Japanese haiku is a three-line poem of 17 syllables with the following line-pattern:

Line 1: 5 syllables

Line 2: 7 syllables

Line 3: 5 syllables

It is not limited with regard to subject-matter, but strictly speaking, should contain some reference, however distant, to the season of the year.

Through lack of sufficient knowledge of these facts, our first effort to produce a machine-aided haiku was irregular; since, though the number of syllables was right, the line- pattern of the frame of the haiku, which was stored in the computer, was:

Line 1: 7 syllables

Line 2: 5 syllables

Line 3: 5 syllables

The computer's simulation of the action of the poet's mind: the thesaurus and the frame.

In order to use a machine to handle natural language for any purpose whatever, you have to make a hypothesis about the nature, the potentialities and the structure of the sample of natural language, which you are putting into your machine. As a radio critic recently said, you have got to 'crack language'. This fact has either not been realized, or has been evaded up to now, particularly by those working on information retrieval and in mechanical translation, two fields in which the capacity to 'crack language' is quite evidently required.

To put 1,000,000 words of Russian-English alphabetized dictionary on disc, make the computer match it with the alphabetized set of words in a Russian text (with the same word matched against the same translation each time it comes), record the matched entries in the serial text order and then print out the English: to do all this without first gaining any insight or understanding of what translation is, this is not to translate at all but to use the computer simply to transliterate. Any true translation there may be will be performed by the unfortunate man who tries to make out what the computer output with which he is presented can possibly have meant; and such a man will be translating (if he can) from computer output into English. What the computer itself produces could only be called 'idiot translation'.

The same goes for poetry. To put a set of words on disc in the machine, program the machine to make a random choice between them, constrained only by rhyming requirements, and to do nothing else, this is to write idiot poetry. Judged by this test (the test being how much insight was used in making them), the poems produced by computers are, on the whole, at present very inferior to computer-produced graphics where sophisticated mathematical formulae are mechanically converted to produce abstract topological designs, some of these being of deep metaphysical beauty. In poetry, we have not as yet got the generating formulae; though who would doubt that a poem, any poem, has in fact an interior logic of its own? The analytic attack made upon the Japanese haiku, therefore, in order to computerize it, represents a first attempt to get the glimmer of a glimmer of what the interior logic of a simple poem-form could be like.

This fact that the difficult problem of producing a poem's generating formula was for the first time made explicit by a rudimentary generating formula being actually displayed, sufficiently explains the haiku's unexpected success at the Cybernetic Serendipity exhibition. People who were neither Japanese nor poets could, and did, work the algorithm and produce a poem; which then, in some cases. they carefully treasured and placed in their breast pockets. (The same phenomenon was observed at an earlier stage at the Cambridge Language Research Unit. An engineer and an information-scientist, each with an international reputation, both wrote poems from the algorithm at a research meeting and both were observed slinking secretly back into the meeting-room afterwards to copy their own poems off the display-board, put them in their pockets and take them home. When a poet, Alan Trist, came to the exhibition, of course, he adopted the idea, not the haiku: and invented his own frame and wrote his own poems, from his own algorithm - which was just what had been hoped that someone would do. But long before this, the exhibit giving the poem-game had succeeded, beyond all hope, in its primary object which was that of enabling non-poets to be able to make poems - by play.)

Not only has it been shown by Huizinga, in his book Homo Ludens, that play is a fundamental religious activity; but also playing with any new technique is the first stage of handling the said technique seriously, of really exploiting it. Here, we have got genuine art creating new techniques? which will emerge as soon as more business executives who have on-line consoles in their offices, find it more fun to write poems on them than to explore the current state of the market, or to model their own firm's production flow. The computer can process more words, faster, than can any human being; it can multi-classify to an extent far beyond the ultimate limit of classification which man can contemplate: it can file-handle, mechanically generating any new file out of any old one with no constraint (e.g. it can turn all the men mentioned in a newspaper report into women, or turn an apologetic business letter into an abusive one, etc.). A moment's reflection should suffice to convince of the immensity of this power. Likewise, it can produce degrees and categories of randomness: and the philosophic re-examination of the whole notion of randomness has been one of the intellectual advances which have occurred partly as side-effects from the intensive study of mechanized calculation. But of all these techniques, the one which is most immediately relevant to compute poetry making, is that of man-machine interaction by means of a reactive online console connected to large rapid-access computer memory. For this enables the computer to enhance and extend the live poet's creativity; not to replace it as would batch-programming a poem on a computer. Larger vocabularies and unusual connexions between the words in them, together with intricate devices hitherto unexplored forms of word-combination, all these can be inserted into the machine, and still leave the live poet, operating the console, free to choose when, how and whether they should be employed: for the machine is being used here as an extension of a typewriter, not as an extension of a desk-calculator.

So, there are genuine new techniques here, for the poet to use, and also genuine areas of knowledge to be explored. One such area is that which lies between what car ordinarily be said, in normal correct English, through what, by extension of saying, might be poetically said, before the boundary is reached of complete gibberish. Another is the exploration of different poetic logics, using 'logic' here in an extended but still definite sense. And the "achievement to be conquered par excellence is the use of all this new knowledge, when we have it, and of all these new extended typewriter techniques (when we have developed them more fully) to enhance and give more power and variety to the intuitive creativity of the real live poet; not to replace him. For new techniques, once they become understood, can make possible the creation of startling new beauty. How, for instance, in music, can you have a ground bass continue, if the only musical instruments known to your culture are a conch, a drum and a flute?

After such talk of man-machine interaction-aided ideals, the computerized toy model of the Japanese haiku will seem trivial indeed. Just so, the Think-a-dot' toy computer seems trivial, when played beside, or on top of, the genuine article. But just as the 'Think-a-dot' becomes more or less insight- provoking according to the degree of insight possessed by the man who plays with it, so now. Just as elementary mathematics can be envisaged from an advanced point of view, so elementary computer poem-making can be envisaged from an advanced point of view. And the algorithm given here for generating Japanese haikus on-line endeavours to do just this.

The hypothesis employed here is that every poem has a frame, and that the activity of frame-making can be analytically distinguished from the activity of filling in the frame. In the haiku, therefore, the haiku-frame is stored in the computer; while the thesaurus represents the initial treasury of synonyms, or otherwise constrained word-classes, with which the poet, by man-machine interaction, fills up, one by one, the gaps left in the frame. The computer, meanwhile, having absorbed and inserted these fillings, prints out the final poem with the gaps all filled in; sometimes startling the poet at the computer console, who never did more than choose one word from a given class of words at any one time.

This, of course, is a gross over-simplification of what the true poet does, and in two respects. The true poet starts with inspired fragments, emerging fully formed from his subconscious; only at a relatively late stage, quite often, does he choose his frame. So there are (at least) three stages: orientating hunch, emergence of inspired fragments, choice of frame. Moreover, the true poet will never have a fixed thesaurus. Word-class-generation goes on in him even when sober; and the more so, not the less, the more frightened he is, the more drunk, the more inspired. Nevertheless, there have been periods, and there have been cultures, where presenting your own poem (like making up your own bunch of flowers to place in the bedroom of a guest in our culture now) was an ordinary social grace. If our culture (machine-aided), so changed as to require this grace, bought Christmas and birthday cards would become a thing of the past; you would make or draw your own, using at need your console, and the convention would be that the letter press of the card had to be your own specially invented poem - or your own specially invented code.

In such circumstances the process of poetic creation would be accelerated, simplified and made self-conscious - and would become something very much more like the process exemplified below in the haiku (which was, indeed, just such a socially required poem in traditional Japan). The frame would be chosen, first, to fit the social occasion; and next, from a more or less stereotyped sequence of word- classes consisting each of monosyllables or rhymes or half-rhymes, the poet would make concurrent choices to fill in the gaps in the frames. And, proceeding thus, a great many mediocre poems (see below) would be generated. But - as also in the case of the socially required poem-form of past cultures - what could be done within the social constraints of such a requirement would only emerge when a true poet, handling the medium, emerged also. A true poet might make inspired choices, even when handling the toy haiku.

...

This text is an excerpt from: Masterman, M. (1971) 'Computerized haiku', in Cybernetics, art and ideas, (Ed.) Reichardt, J. London, Studio Vista, 175-184.