A program for aligning sentences in bilingual corpora

Abstract

Researchers in both machine Iranslation (e.g., Brown et al., 1990) and bilingual lexicography (e.g., Klavans and Tzoukermann, 1990) have recently become interested in studying parallel texts, texts such as the Canadian Hansards (parliamentary proceedings) which are available in multiple languages (French and English). This paper describes a method for aligning sentences in these parallel texts, based on a simple statistical model of character lengths. The method was developed and tested on a small trilingual sample of Swiss economic reports. A much larger sample of 90 million words of Canadian Hansards has been aligned and donated to the ACL/DCI.

Evaluation of Methods for Sentence and Lexical Alignment of Brazilian Portuguese and English Parallel Texts

“…It can be noticed that only the methods that consider cognate words as an alignment criterion had success in omissions. In [7], Gale and Church had already mentioned the necessity of considering language-specific methods to deal adequately with this alignment category and this point was confirmed by the results reported in this paper.…”

“…Precision stands for the number of 6 It is important to say that CorpusPE was evaluated with 64 pairs rather than 65 because we note that one of them was not parallel at lexical level. 7 Available in http://www.d.umn.edu/ tdeperse/code.html.…”

“…The sentence alignment methods evaluated here were named: GC ( [6], [7]), GMA and GSA+ ( [8], [9]), Piperidis et al ( [10]) and TCA ( [11]). …”

“…The algorithms used for automatic extraction of multiword units were NSP (N-gram Statistic Package) 7 and another which was implemented based on the Mutual Expectation technique [19]. Through this process, three lists (for each language) were generated by each algorithm and the final English and BP multiword lists have 240 and 222 units respectively.…”

PAN@FIRE: Overview of the Cross-Language !ndian Text Re-Use Detection Competition

“…The features for the classification process are three: (i) and (ii) are the score of the most likely alignments at sentence and paragraph level between d hi and d en , respectively. These scores were computed with the length based alignment algorithm proposed by [13]. (iii) is a lexical feature: A Hindi-English dictionary was used to gloss the Hindi documents and calculate an idf-based cosine similarity between suspicious and potential source documents.…”