Automatic Extraction of Bilingual Collocations from Parallel Corpora

The field of natural language processing has become extremely popular nowadays. In particular, scholars pay much attention to the task of collocation extraction due to its various applications (such as machine translation and speech recognition). A considerable amount of literature has been published on the topic of monolingual collocation extraction, including studies by Zakharov, Seretan, and Bhalla. At the same time, relatively few studies introduce methods of multilingual equivalent phrase search in the target languages; for instance, English–Portuguese–Spanish and Chinese–English corpus analyses were carried out by Garcia and Wu, respectively. Existing research into German and English languages has been limited too, whilst separately these languages were explored to a considerable extent. The purpose of the present study is to implement an algorithm of automatic German–English collocation extraction from a parallel corpus. The current study combines statistical methods, syntactic analysis, and word embeddings to extract bilingual collocations. The result is that the approach can effectively identify German–English equivalents with a precision of 87%. The proposed method can contribute to the performance of machine translation tools, facilitate second language learning and improve dictionaries and academic textbooks.