Can characters reveal your native language? A language-independent approach to native language identification

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1363–1373, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Can characters reveal your native language? A language-independent approach to native language identification Radu Tudor Ionescu⋄, Marius Popescu⋄, Aoife Cahill† ⋄University of Bucharest Department of Computer Science 14 Academiei, Bucharest, Romania raducu.ionescu@gmail.com popescunmarius@gmail.com †Educational Testing Service 660 Rosedale Rd Princeton, NJ 08541, USA acahill@ets.org Abstract A common approach in text mining tasks such as text categorization, authorship identification or plagiarism detection is to rely on features like words, part-of-speech tags, stems, or some other high-level lin- guistic features. In this work, an approach that uses character n-grams as features is proposed for the task of native language identification. Instead of doing standard feature selection, the proposed approach combines several string kernels using mul- tiple kernel learning. Kernel Ridge Re- gression and Kernel Discriminant Analy- sis are independently used in the learning stage. The empirical results obtained in all the experiments conducted in this work in- dicate that the proposed approach achieves state of the art performance in native lan- guage identification, reaching an accuracy that is 1.7% above the top scoring system of the 2013 NLI Shared Task. Further- more, the proposed approach has an im- portant advantage in that it is language in- dependent and linguistic theory neutral. In the cross-corpus experiment, the proposed approach shows that it can also be topic independent, improving the state of the art system by 32.3%. 1 Introduction Using words as basic units is natural in textual analysis tasks such as text categorization, author- ship identification or plagiarism detection. Per- haps surprisingly, recent results indicate that meth- ods handling the text at the character level can also be very effective (Lodhi et al., 2002; Sander- son and Guenter, 2006; Popescu and Dinu, 2007; Grozea et al., 2009; Popescu, 2011; Popescu and Grozea, 2012). By disregarding features of natu- ral language such as words, phrases, or meaning, an approach that works at the character level has an important advantage in that it is language inde- pendent and linguistic theory neutral. This paper presents a state of the art machine learning system for native language identification that works at the character level. The proposed system is inspired by the system of Popescu and Ionescu (2013), but includes some variations and improvements. major improvement is that several string kernelsA are combined via multiple kernel learning (Shawe- Taylor and Cristianini, 2004). Despite the fact that the (histogram) intersection kernel is very popular in computer vision (Maji et al., 2008; Vedaldi and Zisserman, 2010), it has never been used before in text mining. In this work, the intersection kernel is used for the first time in a text categorization task, alone and in combination with other kernels. The intersection kernel lies somewhere in the middle between the kernel that takes into account only the presence of n-grams and the kernel based on the frequency of n-grams (p-spectrum string kernel). Two kernel classifiers are …

Content truncated to 500 words