The aims of AnaGramma

Our aim is to invent a new paradigm and framework for syntactic and semantic analysis, which is called AnaGramma and applied in a novel paser framework.
The principles of our parsing system departs in many ways from previous approaches to sentence analysis and keeps the following aims in view:
  • The system is psycholinguistically motivated, meaning that as far as possible it follows the patterns of human language processing, as set forth in.
  • As a performance based system, it attempts to process all linguistic forms that exist in texts, not putting a particular emphasis on theoretically grammatical but naturally very rarely occurring phenomena. At the same time, our system tries to treat all texts, including ill-formed or grammatical ones, as valid linguistic input.
  • Utilizes a strictly left-to-right approach, processing the input word by word. The parser makes no assumptions about the not yet processed part of the input text and does not refer to it in any way. If a decision cannot be made based on the currently available information, the processing moves on and the decision is deferred until such information becomes available.
  • The parser has a parallel architecture. As opposed to previous approaches, where analyses were usually the result of a series of processing modules connected as a pipeline, our system uses parallel threads to process the current input word (morphological analyzer, modules responsible for different syntactic phenomena, statistical checks, anaphora resolution, focus identification, etc.). These threads communicate and co-operate with each other correcting each other’s mistakes, and the final output is a result of the interplay among all these processing units.
  • The whole text rather than just a single sentence is the basic unit of representation, making it possible to represent anaphoric relationships inside and between sentences in a uniform manner.
  • As a result of this, and the parallel nature of the whole parsing process, the final representation is not necessarily a tree, but a contiguous graph, possibly with different types of arcs.