A graph-based rule-mining framework for natural language learning and understanding

Learning and understanding natural languages are usually considered as independent tasks in natural language processing. These two tasks, however, are strongly interrelated and are presumably unsolvable as separate problems. In this paper, we present an algorithm called Frequent Rule Graph Miner (FRGM) that tackles these problems by alternately improving on the language model and the example interpretations. FRGM is based on an effective graph-mining algorithm adapted for enumerating frequent rule-graphs and is applicable to di erent layers of natural language processing such as morphology, syntax, semantics and pragmatics.