Contrastive Analysis and Native Language Identification

Contrastive Analysis and Native Language Identification Sze-Meng Jojo Wong Centre for Language Technology Macquarie University Sydney, NSW, Australia szewong@science.mq.edu.au Mark Dras Centre for Language Technology Macquarie University Sydney, NSW, Australia madras@science.mq.edu.au Abstract Attempts to profile authors based on their characteristics, including native language, have drawn attention in recent years, via sev- eral approaches using machine learning with simple features. In this paper we investigate the potential usefulness to this task of con- trastive analysis from second language acquis- tion research, which postulates that the (syn- tactic) errors in a text are influenced by an au- thor’s native language. We explore this, first, by conducting an analysis of three syntactic error types, through hypothesis testing and machine learning; and second, through adding in these errors as features to the replication of a previous machine learning approach. This preliminary study provides some support for the use of this kind of syntactic errors as a clue to identifying the native language of an author. 1 Introduction There is a range of work that attempts to infer, from some textual data, characteristics of the text’s author. This is often described by the term authorship pro- filing, and may be concerned with determining an author’s gender, age, or some other attributes. This information is often of interest to, for example, gov- ernments or marketing departments; the application that motivates the current work is profiling of phish- ing texts, texts that are designed to deceive a user into giving away confidential details (Fette et al., 2007; Zheng et al., 2003). The particular characteristic of interest in this pa- per is the native language of an author, where this is not the language that the text is written in. There has been only a relatively small amount of other re- search investigating this question, notably Koppel et al. (2005), Tsur and Rappoport (2007), Estival et al. (2007), and van Halteren (2008). In general these tackle the problem as a text classification task us- ing machine learning, with features over characters, words, parts of speech, and document structures. Koppel et al. (2005) also suggest syntactic features, although they do not use them in that work. The goal of this paper is to make a preliminary in- vestigation into the use of syntactic errors in native language identification. The research drawn on for this work comes from the field of contrastive analy- sis in second language acquisition (SLA). According to the contrastive analysis hypothesis formulated by Lado (1957), difficulties in acquiring a new (second) language are derived from the differences between the new language and the native (first) language of a language user. Amongst the frequently observed syntactic error types in non-native English which it has been argued are attributable to language trans- fer are subject-verb disagreement, noun-number dis- agreement, and misuse of determiners. Contrastive analysis was largely displaced in SLA by error anal- ysis (Corder, 1967), which argued that there are many other types of error in SLA, and that too much emphasis was placed on transfer errors. …

Content truncated to 500 words