Applying Collocation Segmentation to the ACL Anthology Reference Corpus

Proceedings of the ACL-2012 Special Workshop on Rediscovering 50 Years of Discoveries, pages 66–75, Jeju, Republic of Korea, 10 July 2012. c⃝2012 Association for Computational Linguistics Applying Collocation Segmentation to the ACL Anthology Reference Corpus Vidas Daudaraviˇcius Vytautas Magnus University / Vileikos 8, Lithuania v.daudaravicius@if.vdu.lt Abstract Collocation is a well-known linguistic phe- nomenon which has a long history of research and use. In this study I employ collocation segmentation to extract terms from the large and complex ACL Anthology Reference Cor- pus, and also briefly research and describe the history of the ACL. The results of the study show that until 1986, the most signifi- cant terms were related to formal/rule based methods. Starting in 1987, terms related to statistical methods became more important. For instance, language model, similarity mea- sure, text classification. In 1990, the terms Penn Treebank, Mutual Information , statis- tical parsing, bilingual corpus, and depen- dency tree became the most important, show- ing that newly released language resources ap- peared together with many new research areas in computational linguistics. Although Penn Treebank was a significant term only tem- porarily in the early nineties, the corpus is still used by researchers today. The most recent significant terms are Bleu score and semantic role labeling. While machine translation as a term is significant throughout the ACL ARC corpus, it is not significant for any particu- lar time period. This shows that some terms can be significant globally while remaining in- significant at a local level. 1 Introduction Collocation is a well-known linguistic phenomenon which has a long history of research and use. The importance of the collocation paradigm shift is raised in the most recent study on collocations (Sere- tan, 2011). Collocations are a key issue for tasks like natural language parsing and generation, as well as real-life applications such as machine translation, in- formation extraction and retrieval. Collocation phe- nomena are simple, but hard to employ in real tasks. In this study I introduce collocation segmentation as a language processing method, maintaining simplic- ity and clarity of use as per the n-gram approach. In the beginning, I study the usage of the terms collo- cation and segmentation in the ACL Anthology Ref- erence Corpus (ARC), as well as other related terms such as word, multi-word, and n-gram. To evaluate the ability of collocation segmentation to handle dif- ferent aspects of collocations, I extract the most sig- nificant collocation segments in the ACL ARC. In addition, based on a ranking like that of TF-IDF, I extract terms that are related to different phenom- ena of natural language analysis and processing. The distribution of these terms in ACL ARC helps to un- derstand the main breakpoints of different research areas across the years. On the other hand, there was no goal to make a thorough study of the methods used by the ACL ARC, as such a task is complex and prohibitively extensive. 2 ACL Anthology Reference Corpus This study uses the ACL ARC version 20090501. The first step …

Content truncated to 500 words