Can computers truly understand human language? This research explores the use of transition network grammars to analyze natural language sentences. The paper delves into structure-building actions associated with the arcs of the grammar network. These actions enable the reordering, restructuring, and copying of constituents, facilitating the production of deep-structure representations akin to those obtained from transformational analysis. Conditions are set on the arcs to allow for selectivity which rules out analyses that have no meaning. The study highlights the advantages of this model in natural language processing, offering detailed examples to illustrate its capabilities. A powerful selectivity based on semantic information is included to guide parsing. It briefly describes the implementation of an experimental parsing system for transition network grammars. The discussion is presented in detail and exemplified. This research is vital for computer scientists and linguists, providing insights into more effective and meaningful natural language analysis. It may be very helpful in future computer natural language analyses.
This article's exploration of augmented transition network grammars for natural language processing fits within the Communications of the ACM's scope of computer science and software engineering. The journal emphasizes practical applications and theoretical advancements, and this paper contributes to both by proposing a detailed approach to language analysis with an experimental parsing system.