Saab Mansour

Recent advances in SRI'S IraqComm™ Iraqi Arabic-English speech-to-speech translation system

We summarize recent progress on SRIs IraqComm™ Iraqi Arabic-English two-way speech-to-speech translation system. In the past year we made substantial developments in our speech recognition and machine translation technology, leading to significant improvements in both accuracy and speed of the IraqComm system. On the 2008 NIST-evaluation dataset our twoway speech-to-text (S2T) system achieved 6% to 8% absolute improvement in BLEU in both directions, compared to our previous year system [1].

Improved Sentence-Level Arabic Dialect Classification

The paper presents work on improved sentence-level dialect classification of Egyptian Arabic (ARZ) vs. Modern Standard Arabic (MSA). Our approach is based on binary feature functions that can be implemented with a minimal amount of task-specific knowledge. We train a featurerich linear classifier based on a linear support-vector machine (linear SVM) approach. Our best system achieves an accuracy of 89.1 % on the Arabic Online Commentary (AOC) dataset (Zaidan and Callison-Burch, 2011) using 10-fold stratified cross validation: a 1.3 % absolute accuracy improvement over the results published by (Zaidan and Callison-Burch, 2014). We also evaluate the classifier on dialect data from an additional data source. Here, we find that features which measure the informalness of a sentence actually decrease classification accuracy significantly.

Unsupervised Adaptation for Statistical Machine Translation

In this work, we tackle the problem of language and translation models domainadaptation without explicit bilingual indomain training data. In such a scenario, the only information about the domain can be induced from the source-language test corpus. We explore unsupervised adaptation, where the source-language test corpus is combined with the corresponding hypotheses generated by the translation system to perform adaptation. We compare unsupervised adaptation to supervised and pseudo supervised adaptation. Our results show that the choice of the adaptation (target) set is crucial for successful application of adaptation methods. Evaluation is conducted over the German-to-English WMT newswire translation task. The experiments show that the unsupervised adaptation method generates the best translation quality as well as generalizes well to unseen test sets.

A comparison of segmentation methods and extended lexicon models for Arabic statistical machine translation

In this article, we investigate different methodologies of Arabic segmentation for statistical machine translation by comparing a rule-based segmenter to different statistically-based segmenters. We also present a method for segmentation that serves the needs of a real-time translation system without impairing the translation accuracy. Second, we report on extended lexicon models based on triplets that incorporate sentence-level context during the decoding process. Results are presented on different translation tasks that show improvements in both BLEU and TER scores.