A Corpus for Multilingual Document Classification in Eight Languages

A Corpus for Multilingual Document Classification in Eight Languages Holger Schwenk, Xian Li Facebook AI Research, Facebook AML {schwenk,xianl}@fb.com Abstract Cross-lingual document classification aims at training a document classifier on resources in one language and transferring it to a different language without any additional resources. Several approaches have been proposed in the literature and the current best practice is to evaluate them on a subset of the Reuters Corpus Volume 2. However, this subset covers only few languages (English, German, French and Spanish) and almost all published works focus on the the transfer between English and German. In addition, we have observed that the class prior distributions differ significantly between the languages. We argue that this complicates the evaluation of the multilinguality. In this paper, we propose a new subset of the Reuters corpus with balanced class priors for eight languages. By adding Italian, Russian, Japanese and Chinese, we cover languages which are very different with respect to syntax, morphology, etc. We provide strong baselines for all language transfer directions using multilingual word and sentence embeddings respectively. Our goal is to offer a freely available framework to evaluate cross-lingual document classification, and we hope to foster by these means, research in this important area. Keywords: cross-lingual document classification, multilinguality, Reuters Corpus Volume 2 (RCV2), evaluation framework 1. Introduction There are many tasks in natural language processing which require the classification of sentences or longer paragraphs into a set of predefined categories. Typical applications are for instance topic identification (e.g. sports, news, . . .) or product reviews (positive or negative). There is a large body of research on approaches for document classification. An important aspect to compare these different approaches is the availability of high quality corpora to train and eval- uate them. Unfortunately, most of these evaluation tasks focus on the English language only, while there is an ever increasing need to perform document classification in many other languages. One could of course collect and label training data for other languages, but this would be costly and time consuming. An interesting alternative is “cross- lingual document classification”. The underlying idea is to use a representation of the words or whole documents which is independent of the language. By these means, a classifier trained on one language can be transferred to a different one, without the need of resources in that trans- fer language. Ideally, the performance obtained by cross- lingual transfer should be as close as possible to training the entire system on language specific resources. Such a task was first proposed by (Klementiev et al., 2012) using the Reuters Corpus Volume 2. The aim was to first train a classifier on English and then to transfer it to German, and vice versa. An extension to the transfer between En- glish and French and Spanish respectively was proposed by (Mogadala and Rettinger, 2016). However, only few com- parative results are available for these transfer directions. The contributions of this work are as follows. We extend previous works and use …

Content truncated to 500 words