We present various methods to train word-level translation models for statistical machine translation systems that use widely different knowledge sources ranging from parallel corpora and a bilingual lexicon to only monolingual corpora in two languages. Some novel methods are presented and previously published methods are reviewed. Also, a common evaluation metric enables the first quantitative comparison of these approaches.

We present a novel method for extracting parallel sub-sentential fragments from comparable, non-parallel bilingual corpora. By analyzing potentially similar sentence pairs using a signal processinginspired approach, we detect which segments of the source sentence are translated into segments in the target sentence, and which are not. This method enables us to extract useful machine translation training data even from very non-parallel corpora, which contain no parallel sentence pairs. We evaluate the quality of the extracted data by showing that it improves the performance of a state-of-the-art statistical machine translation system. 1

We describe a method that can make a scanned, handwritten mediaeval latin manuscript accessible to full text search. A generalized HMM is fitted, using transcribed latin to obtain a transition model and one example each of 22 letters to obtain an emission model. We show results for unigram, bigram and trigram models. Our method transcribes 25 pages of a manuscript of Terence with fair accuracy (75 % of letters correctly transcribed). Search results are very strong; we use examples of variant spellings to demonstrate that the search respects the ink of the document. Furthermore, our model produces fair searches on a document from which we obtained no training data. 1. Intoduction There are many large corpora of handwritten scanned documents, and their number is growing rapidly. Collections range from the complete works of Mark Twain to thousands of pages of zoological notes spanning two centuries. Large scale analyses of such corpora

Automatically acquiring synonymous collocation pairs such as <turn on, OBJ, light> and <switch on, OBJ, light> from corpora is a challenging task. For this task, we can, in general, have a large monolingual corpus and/or a very limited bilingual corpus. Methods that use monolingual corpora alone or use bilingual corpora alone are apparently inadequate because of low precision or low coverage. In this paper, we propose a method that uses both these resources to get an optimal compromise of precision and coverage. This method first gets candidates of synonymous collocation pairs based on a monolingual corpus and a word thesaurus, and then selects the appropriate pairs from the candidates using their translations in a second language. The translations of the candidates are obtained with a statistical translation model which is trained with a small bilingual corpus and a large monolingual corpus. The translation information is proved as effective to select synonymous collocation pairs. Experimental results indicate that the average precision and recall of our approach are 74 % and 64 % respectively, which outperform those methods that only use monolingual corpora and those that only use bilingual corpora. 1

I propose a novel co-training method for statistical machine translation. As co-training requires multiple learners trained on views of the data which are disjoint and sufficient for the labeling task, I use multiple source documents as views on translation. Co-training for statistical machine translation is therefore a type of multi-source translation. Unlike previous mutli-source methods, it improves the overall quality of translations produced by a model, rather than single translations. This is achieved by augmenting the parallel corpora on which the statistical translation models are trained. Experiments suggest that co-training is especially effective for languages with highly impoverished parallel corpora.

We introduce Bilingual Suffix Trees (BST), a data structure that is suitable for exploiting comparable corpora. We discuss algorithms that use BSTs in order to create parallel corpora and learn translations of unseen words from comparable corpora. Starting with a small bilingual dictionary that was derived automatically from a corpus of 5.000 parallel sentences, we have automatically extracted a corpus of 33.926 parallel phrases of size greater than 3, and learned 9 new word translations from a comparable corpus of 1.3M words (100.000 sentences).

In this work we present results from using Amazon’s Mechanical Turk (MTurk) to annotate translation lexicons between English and a large set of less commonly used languages. We generate candidate translations for 100 English words in each of 42 foreign languages using Wikipedia and a lexicon induction framework. We evaluate the MTurk annotations by using positive and negative control candidate translations. Additionally, we evaluate the annotations by adding pairs to our seed dictionaries, providing a feedback loop into the induction system. MTurk workers are more successful in annotating some languages than others and are not evenly distributed around the world or among the world’s languages. However, in general, we find that MTurk is a valuable resource for gathering cheap and simple annotations for most of the languages that we explored, and these annotations provide useful feedback in building a larger, more accurate lexicon. 1

This paper describes an algorithm to automatically generate a list of cognates in a target language by means of Support Vector Machines. While Levenshtein distance was used to align the training file, no knowledge repository other than an initial list of cognates used for training purposes was input into the algorithm. Evaluation was set up in a cognate production scenario which mimed a reallife situation where no word lists were available in the target language, delivering the ideal environment to test the feasibility of a more ambitious project that will involve language portability. An overall improvement of 50.58 % over the baseline showed promising horizons. 1

Shannon’s Noisy-Channel model, which describes how a corrupted message might be reconstructed, has been the corner stone for much work in statistical language and speech processing. The model factors into two components: a language model to characterize the original message and a channel model to describe the channel’s corruptive process. The standard approach for estimating the parameters of the channel model is unsupervised Maximum-Likelihood of the observation data, usually approximated using the Expectation-Maximization (EM) algorithm. In this paper we show that it is better to maximize the joint likelihood of the data at both ends of the noisy-channel. We derive a corresponding bi-directional EM algorithm and show that it gives better performance than standard EM on two tasks: (1) translation using a probabilistic lexicon and (2) adaptation of a part-of-speech tagger between related languages. 1.

Collocation translation is important for machine translation and many other NLP tasks. Unlike previous methods using bilingual parallel corpora, this paper presents a new method for acquiring collocation translations by making use of monolingual corpora and linguistic knowledge. First, dependency triples are extracted from Chinese and English corpora with dependency parsers. Then, a dependency triple translation model is estimated using the EM algorithm based on a dependency correspondence assumption. The generated triple translation model is used to extract collocation translations from two monolingual corpora. Experiments show that our approach outperforms the existing monolingual corpus based methods in dependency triple translation and achieves promising results in collocation translation extraction. 1