Omnia Mutantur, Nihil Interit: Connecting Past with Present by Finding Corresponding Terms across Time

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 645–655, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Omnia Mutantur, Nihil Interit: Connecting Past with Present by Find- ing Corresponding Terms across Time Yating Zhang*, Adam Jatowt*, Sourav S Bhowmick+, Katsumi Tanaka* *School of Informatics, Kyoto University +School of Computer Engineering, Nanyang Technological University {zhang,adam,tanaka}@dl.kuis.kyoto-u.ac.jp assourav@ntu.edu.sg Abstract In the current fast-paced world, people tend to possess limited knowledge about things from the past. For example, some young users may not know that Walkman played similar func- tion as iPod does nowadays. In this paper, we approach the temporal correspondence prob- lem in which, given an input term (e.g., iPod) and the target time (e.g. 1980s), the task is to find the counterpart of the query that existed in the target time. We propose an approach that transforms word contexts across time based on their neural network representations. We then experimentally demonstrate the ef- fectiveness of our method on the New York Times Annotated Corpus. 1 Introduction What music device 30 years ago played similar role as iPod does nowadays? Who are today’s Beatles? Who was a counterpart of President Chi- rac in 1988? These and many other similar ques- tions may be difficult to answer by average users (especially, by young ones). This is because peo- ple tend to possess less knowledge about the past than about the contemporary time. In this work we propose an effective method to solve the problem of finding counterpart terms across time. In particular, for an input pair of a term (e.g., iPod) and the target time (e.g. 1980s), we find the corresponding term that existed in the target time (walkman). We consider temporal counterparts to be terms which are semantically similar, yet, which existed in different time. Knowledge of temporal counterparts can help to alleviate the problem of terminology gap for us- ers searching within temporal document collec- tions such as archives. For example, given a user’s query and the target time frame, a new modified query that represents the same meaning could be suggested to improve search results. Essentially, it would mean letting searchers use the knowledge they possess on the current world to perform search within unknown collections such as ones containing documents from the distant past. Fur- thermore, solving temporal correspondence prob- lem can help timeline construction, temporal sum- marization, reference forecasting and can have ap- plications in education. The problem of temporal counterpart detection is however not trivial. The key difficulty comes from the change of the entire context that results in low overlap of context across time. In other words, it is difficult to find temporal counterpart terms by directly comparing context vectors across time. This fact is nicely portrayed by the Latin proverb: “omnia mutantur, nihil interit” (in English: “everything changes, nothing perishes”) which indicates that there are no completely static things, yet, many things and concepts are still sim- ilar …

Content truncated to 500 words