Parsing with Coordinate Structure Analysis based on Sequence Alignment

对于出现and或or“or”等并列连词的给定句子，检测并列连词所连接的单词序列的范围，以及区分其语法范畴，被认为是自然语言处理的基础。然而，即使是最先进的解析器也只能识别大约50%的这种协调范围及其语法类别。为了解决这个问题，我们开发了一种方法，该方法使用了基于我们为协调定制的语法的CKY算法，以及我们通过序列比对和平均感知器定义的可调相似性度量，这很好地提高了性能。

For a given sentence in which coordination conjunctions such as and or "or" occur, detecting the scope of word sequences that the coordination conjunctions join together, as well as distinguishing the grammatical category of it, are deemed fundamental in natural language processing. However, even the state-of-the-art parsers can identify only around fifty percent of such coordination scopes and their grammatical category. To cope with the problem, we developed a methodology that employs a CKY algorithm based on the grammar we customize for coordination as well as a tunable similarity measures we define via sequence alignment and averaged perceptron, which improves performance very well.