Matthias Eck

Language model adaptation for statistical machine translation with structured query models

We explore unsupervised language model adaptation techniques for Statistical Machine Translation. The hypotheses from the machine translation output are converted into queries at different levels of representation power and used to extract similar sentences from very large monolingual text collection. Specific language models are then build from the retrieved data and interpolated with a general background model. Experiments show significant improvements when translating with these adapted language models.

Improving statistical machine translation in the medical domain using the unified medical language system

Texts from the medical domain are an important task for natural language processing. This paper investigates the usefulness of a large medical database (the Unified Medical Language System) for the translation of dialogues between doctors and patients using a statistical machine translation system. We are able to show that the extraction of a large dictionary and the usage of semantic type information to generalize the training data significantly improves the translation performance.

Translation Model Pruning via Usage Statistics for Statistical Machine Translation

We describe a new pruning approach to remove phrase pairs from translation models of statistical machine translation systems. The approach applies the original translation system to a large amount of text and calculates usage statistics for the phrase pairs. Using these statistics the relevance of each phrase pair can be estimated. The approach is tested against a strong baseline based on previous work and shows significant improvements.

Performance of an online translation tool when applied to patient educational material

BACKGROUND Language barriers may prevent clinicians from tailoring patient educational material to the needs of individuals with limited English proficiency. Online translation tools could fill this gap, but their accuracy is unknown. We evaluated the accuracy of an online translation tool for patient educational material. METHODS We selected 45 sentences from a pamphlet available in both English and Spanish, and translated it into Spanish using GoogleTranslate™ (GT). Three bilingual Spanish speakers then performed a blinded evaluation on these 45 sentences, comparing GT-translated sentences to those translated professionally, along four domains: fluency (grammatical correctness), adequacy (information preservation), meaning (connotation maintenance), and severity (perceived dangerousness of an error if present). In addition, evaluators indicated whether they had a preference for either the GT-translated or professionally translated sentences. RESULTS The GT-translated sentences had significantly lower fluency scores compared to the professional translation (3.4 vs. 4.7, P < 0.001), but similar adequacy (4.2 vs. 4.5, P = 0.19) and meaning (4.5 vs. 4.8, P = 0.29) scores. The GT-translated sentences were more likely to have any error (39% vs. 22%, P = 0.05), but not statistically more likely to have a severe error (4% vs. 2%, P = 0.61). Evaluators preferred the professional translation for complex sentences, but not for simple ones. DISCUSSION When applied to patient educational material, GT performed comparably to professional human translation in terms of preserving information and meaning, though it was slightly worse in preserving grammar. In situations where professional human translations are unavailable or impractical, online translation may someday fill an important niche.

Incremental Adaptation of Speech-to-Speech Translation

In building practical two-way speech-to-speech translation systems the end user will always wish to use the system in an environment different from the original training data. As with all speech systems, it is important to allow the system to adapt to the actual usage situations. This paper investigates how a speech-to-speech translation system can adapt day-to-day from collected data on day one to improve performance on day two. The platform is the CMU Iraqi-English portable two-way speech-to-speech system as developed under the DARPA TransTac program. We show how machine translation, speech recognition and overall system performance can be improved on day 2 after adapting from day 1 in both a supervised and unsupervised way.

Using web text to improve keyword spotting in speech

For low resource languages, collecting sufficient training data to build acoustic and language models is time consuming and often expensive. But large amounts of text data, such as online newspapers, web forums or online encyclopedias, usually exist for languages that have a large population of native speakers. This text data can be easily collected from the web and then used to both expand the recognizers vocabulary and improve the language model. One challenge, however, is normalizing and filtering the web data for a specific task. In this paper, we investigate the use of online text resources to improve the performance of speech recognition specifically for the task of keyword spotting. For the five languages provided in the base period of the IARPA BABEL project, we automatically collected text data from the web using only Limited LP resources. We then compared two methods for filtering the web data, one based on perplexity ranking and the other based on out-of-vocabulary (OOV) word detection. By integrating the web text into our systems, we observed significant improvements in keyword spotting accuracy for four out of the five languages. The best approach obtained an improvement in actual term weighted value (ATWV) of 0.0424 compared to a baseline system trained only on LimitedLP resources. On average, ATWV was improved by 0.0243 across five languages.

Jibbigo: Speech-to-speech translation on mobile devices

Jibbigo is a speech-to-speech translation application for iPhone, iPod touch, and iPad devices. Jibbigo allows the user to simply speak a sentence, and it speaks the sentence aloud in the other language, much like a personal human interpreter would. The speech-to-speech translation is bi-directional for a two way dialog between participants.