SuMT: A Framework of Summarization and MT

International Joint Conference on Natural Language Processing, pages 270–278, Nagoya, Japan, 14-18 October 2013. SUMT: A Framework of Summarization and MT Houda Bouamor Behrang Mohit Carnegie Mellon University Doha, Qatar hbouamor@qatar.cmu.edu, behrang@cmu.edu, ko@cs.cmu.edu Kemal Oflazer Abstract We present a novel system combination of machine translation and text summariza- tion which provides high quality summary translations superior to the baseline trans- lation of the entire document. We first use supervised learning and build a classifier that predicts if the translation of a sentence has high or low translation quality. This is a reference-free estimation of MT qual- ity which helps us to distinguish the subset of sentences which have better translation quality. We pair this classifier with a state- of-the-art summarization system to build an MT-aware summarization system. To evaluate summarization quality, we build a test set by summarizing a bilingual corpus. We evaluate the performance of our sys- tem with respect to both MT and summa- rization quality and, demonstrate that we can balance between improving MT qual- ity and maintaining a decent summariza- tion quality. 1 Introduction Machine Translation (MT) has been championed as an effective technology for knowledge trans- fer from English to languages with less digital content. An example of such efforts is the au- tomatic translation of English Wikipedia to lan- guages with smaller collections. However, MT quality is still far from ideal for many of the lan- guages and text genres. While translating a docu- ment, there are many poorly translated sentences which can provide incorrect context and confuse the reader. Moreover, some of these sentences are not as informative and could be summarized to make a more cohesive document. Thus, for tasks in which complete translation is not mandatory, MT can be effective if the system can provide a more informative subset of the content with higher translation quality. In this work, we demonstrate a framework of MT and text summarization which replaces the baseline translation with a proper summary that has higher translation quality than the full transla- tion. For this, we combine a state of the art English summarization system and a novel framework for prediction of MT quality without references. Our research contributions are: (a) We extend a classification framework for reference-free prediction of translation quality at the sentence-level. (b) We incorporate MT knowledge into a summa- rization system which results in high quality translation summaries. (c) For evaluation purposes, we conduct a bilin- gual manual summarization of a parallel cor- pus.1 Our English-Arabic system reads in an En- glish document along with its baseline Arabic translation and outputs, as a summary, a subset of the Arabic sentences based on their informa- tiveness and also their translation quality. We demonstrate the utility of our system by evaluat- ing it with respect to both its MT and the sum- marization quality. For summarization, we con- duct both reference-based and reference-free eval- uations and observe a performance in the range of the state of the art system. Moreover, the trans- lation quality of …

Content truncated to 500 words