Automated Grammar Correction Using Hierarchical Phrase-Based Statistical Machine Translation

International Joint Conference on Natural Language Processing, pages 937–941, Nagoya, Japan, 14-18 October 2013. Automated Grammar Correction Using Hierarchical Phrase-Based Statistical Machine Translation Bibek Behera, Pushpak Bhattacharyya Dept. of Computer Science and Engineering IIT Bombay, Mumbai, India {bibek,pb}@cse.iitb.ac.in Abstract We introduce a novel technique that uses hierarchical phrase-based statistical ma- chine translation (SMT) for grammar cor- rection. SMT systems provide a uni- form platform for any sequence transfor- mation task. Thus grammar correction can be considered a translation problem from incorrect text to correct text. Over the years, grammar correction data in the elec- tronic form (i.e., parallel corpora of incor- rect and correct sentences) has increased manifolds in quality and quantity, mak- ing SMT systems feasible for grammar correction. Firstly, sophisticated transla- tion models like hierarchical phrase-based SMT can handle errors as complicated as reordering or insertion, which were diffi- cult to deal with previously throuh the me- diation of rule based systems. Secondly, this SMT based correction technique is similar in spirit to human correction, be- cause the system extracts grammar rules from the corpus and later uses these rules to translate incorrect sentences to correct sentences. We describe how to use Joshua, a hierarchical phrase-based SMT system for grammar correction. An accuracy of 0.77 (BLEU score) establishes the efficacy of our approach. 1 Introduction We consider grammar correction as a translation problem - translation from an incorrect sentence to a correct sentence. The correcting system is trained using a parallel corpus of incorrect and their corresponding correct sentences. The system learns SCFG (synchronous context free grammar) rules (Chiang, 2005) during translation. SCFG rules look like this:- • X → X1 of X2, X1 for X2 The above rule implies that phrases X1 and X2 in source language are translated to phrases in tar- get language, while of is replaced with for. The position of both phrases w.r.t. of remains same in the target language, which means there is no re- ordering of phrases. After generating such grammar rules, it con- verts the erroneous sentence to a tree using the rules of grammar, i.e., the left hand side of the SCFG rules. It then applies correction rules, i.e., the right hand side of the SCFG rules, to convert the tree as explained later in section 3. The yield of the tree generates the corrected sentence. Here are various types of errors that one en- counters in grammar correction: Article choice errors:- a Himalayas is the longest mountain range in the world. The correct translation is ‘the Himalayas is the longest moun- tain range in the world’. Preposition errors:- Helicopter crashed at cen- tral London. The correct translation is ‘Helicopter crashed in central London’. Word form errors:- The rain mays fall in July should be changed to ‘The rain may fall in July’. Word insertion errors:- The court deemed nec- essary that she respond to the summons should be changed to ‘The court deemed it necessary that she respond to the summons’. Reordering errors:- never we miss deadlines should …

Content truncated to 500 words