The use of machines in the construction of a grammar and computer program for structural analysis

The present paper describes progress made on the building of a descriptive grammar of Russian with the complementary efforts of man and machine. Linguistic research at the Rand Corporation begins with the collection on punched cards of a large quantity of raw text from Russian physics journals. As described elsewhere in detail, a total of 250,000 running words of text is being processed, in corpora of about 30,000 words each. Post editors supply codes to indicate (a) the structure of the Russian sentence and (b) its translation into English. In this way the relative position of each word in the structure of the whole sentence is recognized and codified. Dependency codes are then punched back into the text cards. The entire corpus is then machine-sorted and listed according to the structural and morphological type of each item in the text, and according to lexical entries. Syntactic analyses of these listings lead to the identification of word classes according to function (the extension and modification of traditional grammatical classifications) and to identification of the relations between syntactic units of the sentence. The word classes and functional relationships thus determined are imbedded in a computer program for sentence-structure determination that is now being tested. The program establishes a relationship between two words in a specific sentence when: (a) the words belong to classes that, in general, can be related, and (b) all intervening words in the sentence have previously been related to one or the other of the words in question. The sum of the word classes and functional relationships that can exist among them is a grammar for Russian physics texts, while the computer program for translation is a working statement of the grammar. The empirical questions now under test are: (a) What word classes and functional relationships are to be recognized for Russian ? (b) Do the computer-determined sentence structures match those given for the same sentences by linguists ? Harper • The use of machines in the construction of a grammar and computer program for structural analysis 189