The state of the art in machine translation (MT) is governed by neural approaches, which typically provide superior translation accuracy over statistical approaches. However, on the closely related task of word alignment, traditional statistical word alignment models often remain the go-to solution. In this paper, we present an approach to train a Transformer model to produce both accurate translations and alignments. We extract discrete alignments from the attention probabilities learnt during regular neural machine translation model training and leverage them in a multi-task framework to optimize towards translation and alignment objectives. We demonstrate that our approach produces competitive results compared to GIZA++ trained IBM alignment models without sacrificing translation accuracy and outperforms previous attempts on Transformer model based word alignment. Finally, by incorporating IBM model alignments into our multi-task training, we report significantly better alignment accuracies compared to GIZA++ on three publicly available data sets. Our implementation has been open-sourced 1 .

No abstract

Active learning (AL) for machine translation (MT) has been well-studied for the phrasebased MT paradigm. Several AL algorithms for data sampling have been proposed over the years. However, given the rapid advancement in neural methods, these algorithms have not been thoroughly investigated in the context of neural MT (NMT). In this work, we address this missing aspect by conducting a systematic comparison of different AL methods in a simulated AL framework. Our experimental setup to compare different AL methods uses: i) State-of-the-art NMT architecture to achieve realistic results; and ii) the same dataset (WMT'13 English-Spanish) to have fair comparison across different methods. We then demonstrate how recent advancements in unsupervised pre-training and paraphrastic embedding can be used to improve existing AL methods. Finally, we propose a neural extension for an AL sampling method used in the context of phrase-based MT-Round Trip Translation Likelihood (RTTL). RTTL uses a bidirectional translation model to estimate the loss of information during translation and outperforms previous methods.

No abstract

Speech recognition has often been suggested as a key to universal information access, as the speech modality is a "natural" way to interact, does not require literacy, and relies on existing telephony infrastructure. However, success stories of speech interfaces in developing regions are few and far between. The challenges of literacy, dialectal variation, and the prohibitive expense of creating the necessary linguistic resources are intractable using traditional techniques. We present our findings evaluating a low-cost, scalable speech-driven application designed and deployed in a community center in rural Tamil Nadu, India, to disseminate agricultural information to village farmers. (c) 2007 by The Massachusetts Institute of Technology.

Transliteration of named entities in user queries is a vital step in any Cross-Language Information Retrieval (CLIR) system. Several methods for transliteration have been proposed till date based on the nature of the languages considered. In this paper, we discuss about our transliteration algorithm for mapping English named entities to their proper Tamil equivalents. Our algorithm employs a grapheme-based model, in which transliteration equivalents are identified by mapping the source language names to their equivalents in a target language database, instead of generating them. The basic principle is to compress the source word into its minimal form and align it across an indexed list of target language words to arrive at the top n-equivalents based on the edit distance. We compare the performance of our approach with a statistical generation approach using Microsoft Research India (MSRI) transliteration corpus. Our approach has proved very effective in terms of accuracy and time.

We introduce a novel active learning algorithm for speech recognition in the context of accent adaptation. We adapt a source recognizer on the target accent by selecting a matched subset of utterances from a large, untranscribed and multiaccented corpus for human transcription. Traditionally, active learning in speech recognition has relied on uncertainty based sampling to choose the most informative samples for manual labeling. Such an approach doesn't include explicit relevance criterion during data selection, which is crucial for choosing utterances to match the target accent, from datasets with wide-ranging speakers of different accents. We formulate a cross-entropy based relevance measure to complement uncertainty based sampling for active learning to aid accent adaptation. We evaluate the algorithm on two different setups for Arabic and English accents and show that our approach performs favorably to conventional data selection. We analyze the results to show the effectiveness of our approach in finding the most relevant subset of utterances for improving the speech recognizer on the target accent.

The conventional paradigm in speech translation starts with a speech recognition step to generate transcripts, followed by a translation step with the automatic transcripts as input. To address various shortcomings of this paradigm, recent work explores end-to-end trainable direct models that translate without transcribing. However, transcripts can be an indispensable output in practical applications, which often display transcripts alongside the translations to users. We make this common requirement explicit and explore the task of jointly transcribing and translating speech. Although high accuracy of transcript and translation are crucial, even highly accurate systems can suffer from inconsistencies between both outputs that degrade the user experience. We introduce a methodology to evaluate consistency and compare several modeling approaches, including the traditional cascaded approach and end-to-end models. We find that direct models are poorly suited to the joint transcription/translation task, but that end-to-end models that feature a coupled inference procedure are able to achieve strong consistency. We further introduce simple techniques for directly optimizing for consistency, and analyze the resulting trade-offs between consistency, transcription accuracy, and translation accuracy. 1