In this paper, we report an analysis of the strengths and weaknesses of several Machine Translation (MT) engines implementing the three most widely used paradigms. The analysis is based on a manually built test suite that comprises a large range of linguistic phenomena. Two main observations are on the one hand the striking improvement of an commercial online system when turning from a phrase-based to a neural engine and on the other hand that the successful translations of neural MT systems sometimes bear resemblance with the translations of a rule-based MT system.

The problem of a total absence of parallel data is present for a large number of language pairs and can severely detriment the quality of machine translation. We describe a language-independent method to enable machine translation between a low-resource language (LRL) and a third language, e.g. English. We deal with cases of LRLs for which there is no readily available parallel data between the low-resource language and any other language, but there is ample training data between a closelyrelated high-resource language (HRL) and the third language. We take advantage of the similarities between the HRL and the LRL in order to transform the HRL data into data similar to the LRL using transliteration. The transliteration models are trained on transliteration pairs extracted from Wikipedia article titles. Then, we automatically back-translate monolingual LRL data with the models trained on the transliterated HRL data and use the resulting parallel corpus to train our final models. Our method achieves significant improvements in translation quality, close to the results that can be achieved by a general purpose neural machine translation system trained on a significant amount of parallel data. Moreover, the method does not rely on the existence of any parallel data for training, but attempts to bootstrap already existing resources in a related language.

Grapheme-to-phoneme conversion (g2p) is necessary for text-to-speech and automatic speech recognition systems. Most g2p systems are monolingual: they require language-specific data or handcrafting of rules. Such systems are difficult to extend to low resource languages, for which data and handcrafted rules are not available. As an alternative, we present a neural sequence-to-sequence approach to g2p which is trained on spelling-pronunciation pairs in hundreds of languages. The system shares a single encoder and decoder across all languages, allowing it to utilize the intrinsic similarities between different writing systems. We show an 11% improvement in phoneme error rate over an approach based on adapting high-resource monolingual g2p models to low-resource languages. Our model is also much more compact relative to previous approaches.

In this paper we present a novel approach to minimally supervised synonym extraction. The approach is based on the word embeddings and aims at presenting a method for synonym extraction that is extensible to various languages.We report experiments with word vectors trained by using both the continuous bag-of-words model (CBoW) and the skip-gram model (SG) investigating the effects of different settings with respect to the contextual window size, the number of dimensions and the type of word vectors. We analyze the word categories that are (cosine) similar in the vector space, showing that cosine similarity on its own is a bad indicator to determine if two words are synonymous. In this context, we propose a new measure, relative cosine similarity, for calculating similarity relative to other cosine-similar words in the corpus. We show that calculating similarity relative to other words boosts the precision of the extraction. We also experiment with combining similarity scores from differently-trained vectors and explore the advantages of using a part-of-speech tagger as a way of introducing some light supervision, thus aiding extraction.We perform both intrinsic and extrinsic evaluation on our final system: intrinsic evaluation is carried out manually by two human evaluators and we use the output of our system in a machine translation task for extrinsic evaluation, showing that the extracted synonyms improve the evaluation metric.