Syntax as Clipping Blocks: Structures Algorithms and Rules
Jacques Vergne · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 1993
Research topic My research topic is into syntax of natural languages, with, as a guideline, the very high formal redundancy of natural languages. This redundancy allows to make a morpho-syntactic parser working without an exhaustive dictionary. Methodology The methodology comes from the vision of linguistics as a science: corpus observation, experiments upon this corpus, using a parser as an experimental device to validate the theory. Main observation: a natural language is not a formal language Most NLP systems process natural languages as if they were formal languages. Because of the important differences between natural and formal languages, theories and tools for formal languages are not appropriate to natural languages. Brought out segmentations Between sentence and word levels, 3 other partition levels are observed: the partition of the sentence in one center and some blocks, the clip-subject-predicate tripartition, and the clip-noun-verb tripartition in clips, nominal sequences and verbal sequences. These segments are defined and explained. Main algorithm The validation of the sentence pattern is described: it is a sequential process from the outside to the inside of the sentence structure by removing the block patterns. Then it is explained how to transpose relations internal to a block by simulated reclothing of the sentence pattern. Rules The inventory of rules for simplifying verbal sequences, removing blocks, and for verifying centers is given for French language, in the order of the sequential algorithm. The general structures of centers and blocks are presented. An example on a sentence of the corpus is commented. Then some statistics about patterns in the corpus are set out. Enclosure: a topological constraint on dependencies This is the main formal constraint upon the dependency tree, which drastically reduces the number of candidates when (heuristically) computing dependencies between blocks. Conclusions Syntax as clipping blocks is an explicative theory, and an operative theory. Its model applies this main principle for NLP: fidelity to structures of the processed object: a natural language.