UCD : Diachronic Text Classification with Character, Word, and Syntactic N-grams

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 879–883, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics UCD : Diachronic Text Classification with Character, Word, and Syntactic N-grams Terrence Szymanski Insight Centre for Data Analytics School of Computer Science and Informatics University College Dublin, Ireland terrence.szymanski@ucd.ie Gerard Lynch Centre for Applied Data Analytics Research University College Dublin, Ireland gerard.lynch.@ucd.ie Abstract We present our submission to SemEval-2015 Task 7: Diachronic Text Evaluation, in which we approach the task of assigning a date to a text as a multi-class classification problem. We extract n-gram features from the text at the letter, word, and syntactic level, and use these to train a classifier on date-labeled train- ing data. We also incorporate date probabili- ties of syntactic features as estimated from a very large external corpus of books. Our sys- tem achieved the highest performance of all systems on subtask 2: identifying texts by spe- cific time language use. 1 Introduction This paper describes our submission to the SemEval-2015 Task 7, “Diachronic Text Evalua- tion” (Popescu and Strapparava, 2015). The aim of this shared task is to evaluate approaches to- ward diachronic text analysis of a corpus of English- language news articles from The Spectator1 archive, originally published between 1700 and 2014. We solely address subtask 2: “texts with specific time language usage.” The goal of this subtask is to infer the composition date of a text based on im- plicit clues in language of the text, as opposed to overt mentions of datable named entities or events. This task has inherent utility, for example, for his- torians dating texts in an archive with no external datable properties. However, it is equally interest- ing as an investigation into methods for quantifying 1http://www.spectator.co.uk/. changes in language and writing style over a period of centuries. We approach this task in a similar manner as pre- vious work on stylistic text classification (Argamon- Engelson et al., 1998) in that we aim to model stylis- tic, rather than topical, features of the text. From each text we extract a variety of character, lexical, and syntactic features, as described in section 3. We also use a set of syntactic features whose frequencies over time have been estimated from a very large cor- pus of books (Goldberg and Orwant, 2013). While many of these features have previously been used for stylistic analysis, our approach is not to model style per se. Many types of variation may be captured indirectly by our features: the spelling, typogra- phy, lexicon, and grammar of English have changed markedly over the past centuries, as has the genre of news writing. We consider any time-correlated vari- ation to be useful for dating. 2 Data We used the two training sets of texts provided by the challenge organizers for subtask 2. After remov- ing errors (repeated items, items containing no text, items with invalid dates), our training set consisted of 4130 items. Each item contains the text of a snip- pet …

Content truncated to 500 words