Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German

Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German Pierre-Edouard Honnet∗,1, Andrei Popescu-Belis∗,2, Claudiu Musat3, Michael Baeriswyl3 1 Telepathy Labs 2 HEIG-VD / HES-SO 3 Swisscom (Schweiz) AG Sch¨utzengasse 25 Route de Cheseaux 1, CP 521 Genfergasse 14 CH-8001 Z¨urich CH-1401 Yverdon-les-Bains CH-3011 Bern Switzerland Switzerland Switzerland pierre-edouard.honnet andrei.popescu-belis claudiu.musat@swisscom.com @telepathy.ai @heig-vd.ch michael.baeriswyl@swisscom.com Abstract The goal of this work is to design a machine translation (MT) system for a low-resource family of dialects, collectively known as Swiss German, which are widely spoken in Switzerland but seldom written. We collected a significant number of parallel written resources to start with, up to a total of about 60k words. Moreover, we identified several other promising data sources for Swiss German. Then, we designed and compared three strategies for normalizing Swiss German input in order to address the regional diversity. We found that character-based neural MT was the best solution for text normalization. In combination with phrase-based statistical MT, our solution reached 36% BLEU score when translating from the Bernese dialect. This value, however, decreases as the testing data becomes more remote from the training one, geographically and topically. These resources and normalization techniques are a first step towards full MT of Swiss German dialects. Keywords: machine translation, low-resource languages, spoken dialects, Swiss German, character-based neural MT 1. Introduction In the era of social media, more and more people make online contributions in their own language. The diversity of these languages is however a barrier to information ac- cess or aggregation across languages. Machine translation (MT) can now overcome this limitation with considerable success for well-resourced languages, i.e. language pairs which are endowed with large enough parallel corpora to enable the training of neural or statistical MT systems. This is not the case, though, for many low-resourced languages which have been traditionally considered as oral rather than written means of communication, and which often lack standardized spelling and/or exhibit significant variations across dialects. Such languages have an increasing pres- ence in written communication, especially through social media, while remaining inaccessible to non-speakers. This paper presents a written MT system for a mostly spo- ken family of dialects: Swiss German. Although spoken in a technologically developed country by around five mil- lion native speakers, Swiss German has never been signif- icantly used in writing – with the exception of folklore or children books – before the advent of social media. Rather, from primary school, speakers of Swiss German are taught to use High German in writing, more precisely a variety known to linguists as Swiss Standard German, which is one of the three official federal languages along with French and Italian. Swiss German is widely used in social media, but foreigners or even Swiss speakers of the other official lan- guages cannot understand it. In this paper, we describe the first end-to-end MT system from Swiss German to High German. In Section 2, we ∗ Work conducted while the first and second authors were at the Idiap Research Institute, Martigny, Switzerland. present …

Content truncated to 500 words