Robust Neural Machine Translation for Clean and Noisy Speech Transcripts

Robust Neural Machine Translation for Clean and Noisy Speech Transcripts Mattia Di Gangi⋆† Robert Enyedi Alessandra Brusadin Marcello Federico Amazon AI, East Palo Alto - USA ⋆Fondazione Bruno Kessler, Trento - Italy † University of Trento, Italy Abstract Neural machine translation models have shown to achieve high quality when trained and fed with well structured and punctuated input texts. Unfortunately, the latter condition is not met in spoken language translation, where the input is generated by an automatic speech recognition (ASR) sys- tem. In this paper, we study how to adapt a strong NMT system to make it robust to typical ASR errors. As in our application scenarios transcripts might be post-edited by hu- man experts, we propose adaptation strategies to train a sin- gle system that can translate either clean or noisy input with no supervision on the input type. Our experimental results on a public speech translation data set show that adapting a model on a significant amount of parallel data including ASR transcripts is beneficial with test data of the same type, but produces a small degradation when translating clean text. Adapting on both clean and noisy variants of the same data leads to the best results on both input types. 1. Introduction The recent quality improvements [1, 2] of neural machine translation (NMT) [3, 4] opened the way to commercial ap- plications that can provide high-quality translations. The as- sumption is that the sentences to translate will be similar to the training data, usually characterized by properly-formed text in the two translation languages. Poor-quality sentences are usually considered “noise” and removed from the training set to improve the final quality [5]. This practice is so com- mon that a shared task has been devoted to it [6], which got attention from major industrial players in MT. Thus, the ma- jor weakness of NMT lies in coping with noisy input, which is an important feature of a real-world application such as speech translation. The degradation of translation quality with noisy input has been widely reported in literature. Belinkov and Bisk [7] showed that the translation quality rapidly drops with both natural and synthetic noise. Karpukhin et al. [8] have observed a correlation between recognition and translation quality in the context of speech translation. In both cases the degradation is mainly due to word errors, and following works have shown that inserting synthetic noise in the train- The first author carried out the work during an internship at Amazon. Most of the time, travellers worry about their luggage. Most of the time travellers worry about their luggage. Table 1: Example of sentence in which the meaning is changed by a punctuation mark. ing data increases the robustness to the same kind of noise [9, 10]. In practice, ASR transcripts are not only noisy in the choice of words, but also come without punctuation.1 Thus, in order to feed MT systems with ASR output there are two main options: i) use a separate model that inserts punctu- ation (pre-processing) …

Content truncated to 500 words