This paper describes the machine translation technology of Kielikone Ltd. and gives an outline of TranSmart, a Finnish-English MT system which is a commercial application of that base technology. We argue that MT is fundamentally empirical research. Product development is a slow and strenuous process and MT systems will remain incomplete not only vis-a-vis human translations but also with respect to the system's own potential translation quality. An evaluation method is described which measures the progress in MT development. This method can also be used for system and technology evaluation. This paper ends with the claim that the real contribution of MT will be seen in a longer run in applications that do not compete with human translations but in which MT is the only choice. 1. KIELIKONE MT This section describes the base MT technology developed by Kielikone. The technology is language independent and can be used for building MT systems for various language pairs. Kielikone has built...

Abstract. In the course of four evaluations in the Advanced Research Projects Agency Machine Translation series, evaluation methods have evolved for measuring the core components of a diverse set of systems. This paper describes the methodologies in terms of the most recent evaluation of research and production MT systems, and discusses indications of ways to improve the focus and portability of the evaluation. 0. Introduction. Over the past four years, a set of evaluation methodologies have evolved within the MT initiative of the

We report the results of an experiment on automatic NE recognition from Machine Translations produced by five different MT systems. NE annotations are compared with the results obtained from two highquality human translations. The experiment shows that for recognition of a large class of NEs (Person Names, Locations, Dates, etc.) MT output is almost as useful as a human translation. For other types of NEs (Organisation Names) Precision figures are close to the results for human annotation, although Recall is seriously distorted by the degraded quality of MT. The success rate of NE recognition doesn't strongly correlate with human or automatic MT evaluation scores, which suggests that the quality criteria needed for measuring MT usability for dissemination purposes are not pertinent for assimilation tasks such as Information Extraction.

The best systems for machine translation of natural language are based on statistical models learned from data. Conventional representation of a statistical translation model requires substantial offline computation and representation in main memory. Therefore, the principal bottlenecks to the amount of data we can exploit and the complexity of models we can use are available memory and CPU time, and current state of the art already pushes these limits. With data size and model complexity continually increasing, a scalable solution to this problem is central to future improvement. Callison-Burch et al. (2005) and Zhang and Vogel (2005) proposed a solution that we call translation by pattern matching, which we bring to fruition in this dissertation. The training data itself serves as a proxy to the model; rules and parameters are computed on demand. It achieves our desiderata of minimal offline computation and compact representation, but is dependent on fast pattern matching algorithms on text. They demonstrated its application to a common model based on the translation of contiguous substrings, but leave some open problems. Among these is a question: can this approach match the performance of conventional methods despite unavoidable differences that it induces in the model? We show how to answer this question affirmatively. The main

Translation is a complex job. The presence of machine translation (MT) is to increase the productivity of translation and not to replace the job of human translator. Machine translation does not think like human translator but is well regarded as one of the alternative translation mediums that offer a rough translation for pairs of languages. This paper provides some definitions on accuracy in determining the accuracy of the machine translation. Specifically, this paper discusses a deeper understanding on the accuracy of machine translation system, as human’s trust on the system is unstable. It is practical to search beneath the surface of language in understanding translation from cognitive perspective. This paper discusses several parallel efforts that have been carried out along the challenging opportunity in offering translation. It also gives a stimulated idea in translator modeling. Thus, a deeper understanding of the accuracy of machine translation can reduce the worry and fear when using MT system. Although there is still no definite on the definition for accuracy, the role of accuracy in machine translation is important. Finally, this paper hopes to build up a deeper understanding on MT accuracy.

The importance of machine translation (MT) in the stream of text-handling processes has become readily apparent in many current production settings as well as in research programs such as the Translingual Information Detection, Extraction, and Summarization (TIDES) program. The MT Proficiency Scale project has developed a means of baselining the inherent "tolerance" that a text-handling task has for raw MT output, and thus how good the output must be in order to be of use to that task. This method allows for a prediction of how useful a particular system can be in a text-handling process stream, whether in integrated, MTembedded processes, or less integrated userintensive processes.

More often than not, MT these days is delivered as a component of a comprehensive end-toend NLP application. This paper presents two applications that integrate MT with other NLP processes. The first of the two combines MT with crosslingual information retrieval. The second environment uses MT, together with summarization and information extraction techniques, to generate monolingual (English) documents based on information extracted from documents in various languages. In particular, this application generates a time-stamped list of events connected to a particular person. One of the key factors in the document assembly process is the assignment of absolute dates to each sentence produced by the system. Both applications use a general purpose computational architecture that centers on an annotated document collection.

The rapid expansion of SAP in markets around the world has brought with it an urgent need within the company for high-quality translation that adheres to SAP-specific terminology standards. Both the trend toward outsourcing and the increased use of automatic translation tools depend critically on quick and reliable access to official company terminology.

Text prediction is a form of interactive machine translation that is well suited to skilled translators. In principle it can assist in the production of a target text with minimal disruption to a translator's normal routine. However, recent evaluations of a prototype prediction system showed that it significantly decreased the productivity of most translators who used it. In this paper, we analyze the reasons for this and propose a solution which consists in seeking predictions that maximize the expected benefit to the translator, rather than just trying to anticipate some amount of upcoming text. Using a model of a "typical translator" constructed from data collected in the evaluations of the prediction prototype, we show that this approach has the potential to turn text prediction into a help rather than a hindrance to a translator.

Abstract not found