The notion of ambiguity was explored in the previous chapter in depth. Several categories of ambiguities have therefore been identified. In this chapter an experiment will be conducted using Personal Translator, a commercial machine translation system based on rules, and Google translate, a free online machine translation system that implements the statistical method. A brief description of these tools are given below in section one. Since a critical part of this chapter is devoted to evaluation, section two shall dwell on MT evaluation with a focus on Costa MT evaluation tool which assisted annotators during the evaluation task. Section three shall be devoted to the experiment which is carried out in two steps. The first step …show more content…
It has been shown that the best way to understand the meaning of a word is to place it in context, since without context, it would be almost impossible even for a human being to determine the meaning of certain words. Based on this observation, PT has developed a model that performs a contextual analysis in order to identify the semantic network that the candidate ambiguous word pertains to. This contextual analysis is called “neural transfer”. Vera Aleksić & Gregor Thurmair, (2011) note that contextual analysis is not limited only to the sentence, as is often the case in some translation systems, but the analysis is performed beyond sentence level, allowing greater reliability of the contextual analyzer. PT reports improvement of the translation quality by about 40% for texts containing the affected concepts. This has been exemplified with “Gericht”, a German word which means a court or a dish depending on the context in which it is used. In the following examples, “Gericht” appears in the same sentence. Hints for the context are provided by subsequent …show more content…
Indeed, evaluation of MT can be classified according to its aims. MT evaluation is particularly important to at least two categories of people: designers and users of MT devices. Evaluation allows designers of MT to identify weaknesses in their system and find adequate solutions. MT evaluation also allows potential buyers to get an idea of what should be expected from the purchased system. MT evaluation also allows LSP to assess the need to invest in a particular machine translation system. The question of how much post-edition is needed to attain a translation of publishable quality is also decisive in the purchase process. In the following description borrowed from Van Slype (1979), MT evaluation is categorized in terms of the aim of the evaluation and the doer of the evaluation. When MT evaluation is described in terms of who carries out the evaluation, then a distinction can be made between upstream evaluation and downstream evaluation. When MT evaluation is described in terms of the aim of the evaluation, then the macro evaluation is opposed to the