MONO91 NIST Monograph 91: Automatic Indexing: A State-of-the-Art Report Other Potentially Related Research chapter Mary Elizabeth Stevens National Bureau of Standards but some never, used in SYNTOL itself, he reports that both substantives and textual expressions indicative 0£ certain specific SYNTOL relations can be unambiguously identi- fied. Contextual clues are used: for example, if the word "homme'1 occurs it is trans- lated as "sexe masculinit if "femme" also occurs, as 11etre humain11 if 1'animal11 is also mentioned, and as "suj et experimental" otherwise. Melton and her associates at the Center for Documentation and Communication Research, Western Reserve, have also been investigating machine processing of input text with a view to the automatic selection and manipulation of clue words and relation - ships between them for information retrieval purposes. Their material consists of abstracts from the metallurgical section of Chemical Abstracts. From sample abstracts, a lexicon is developed which involves classification of words into those that are signifi- cant from a metallurgical point of view;those that name materials, compounds, environ- ments; those denoting processes; those denoting characteristics of materials; preposi- tions; those which will not operate `Ln the analysis of the text, and the like. On the basis of analysis of a number of sentences from the sample text, rules for combination and selection of specified words in specified relationships can be set up. These rules are designed to identify sentence types which: (1) Describe performance of a process on a material. (2) Discuss a material in terms of properties, components, form, or environment. (3) Describe a process without reference to specific materials. (4) Discuss metallurgical properties without reference to specific materials. (5) Discuss two or more materials, properties or processes. (6) Describe a causal relationship between two properties. (7) Give a comparison of materials. (8) Contain no words of interest in the system. Computer programs to explore the possibilities for automatic analyses of the kind developed manually for the sample abstracts will be written with the objective of finding an effective compromise between mere word identification and total linguistic analysis. Melton says: "If one considers this method of analysis from the point of view of the linguist, he can immediately describe many grammatical constructions, which will prevent the meaningful reduction of these sentences. It is not known at this time how often such sentences will appear in the corpus of this investigation Nor is it known how adversely such failure would affect the retrieval of the information in these sentences. The answers to these questions will be available only after a large sample has been analyzed and put to an extensive retrieval test. At its most successful the project will achieve an automatic processing of metallurgical text which will permit retrieval of the type of information which can be stated in its own terms with a tolerable amount of inappropriate selections. Should this goal be unattainable, the project will have generated a file of abstracts automatically searchable on the word level 141