The authors describe the first phase of a parser-based information retrieval system. After a brief review of probabilistic retrieval techniques, they propose a “deeper” analysis using the Moulton-Robinson dependency theory of syntax and Gruenewald’s syntax crystal parser SCRYP [1]. Instead of conventional phrase grammar parsing, although not incompatible with it, the authors seek to adapt the Moulton-Robinson syntactic grammar, which is structured on relationships of “scope” and “dependency.” Since the Moulton-Robinson processing model encodes complex relationships, rather than rules, it offers a high degree of modularity. A module can be viewed as a four-sided figure with a potential linkage to another module as detemined by relationships of parity (scope) or hierarchy (dependency). According to surface string order, when codes of two modules match horizontally or vertically, the two entities can be linked. Successive passes add these simple structures together in increasingly complex combinations. Although the parser is basically bottom up, it combines high level features of top down parsing as well. A combinatorial problem of exponentially increased structures after each pass can be solved, according to the authors, by local restrictions or heuristic rules: for example, the separate processing of noun and prepositional phrases. Long range advantages will include rapid or realtime processing of full texts, formal simplicity of output which enhances heuristic estimates of relatedness between query and document, and the ability to process incomplete or ungrammatical strings.

The authors discuss briefly implementation issues in which scope/dependency relationships among nouns do or do not match query and string. They conclude with strategies focusing on simple sentences in which the parser retrieves a relationship between two or more nouns and a verb (e.g., Poland attacks Germany, and not the reverse) and problems related to cross-sentence relations. In their next report I would look forward to more details about conventions of document and query input.