英语论文网

EXTRACTING USEFUL INFORMATION FROM TM DATABASES

The first widely available translation memory(TM) tools stored data by dividing 英国留学生信息科技毕业论文 source and target texts into sentences and storing each sentence pair as a unit. TRADOS, for example, was designed using this sentence-based approach.

Recently, a character-string/bidet (CSB)-based approach has emerged that focuses not on sentences but on character strings, keeping the bidets intact so that any string can be examined in context. Multitask, an example of this approach, is not advertised as a TM, perhaps to differentiate it from tools using the sentence-based approach, but still fits the broad definition of TM as an archive of previous translations structured for reuse.

Earlier attempts at comparative evaluation of TM tools focused only on the sentence-based approach, partly because the CSB-based approach is so new. Each type of tool generates different types of matches when a new source text is compared with the database. Most TM tools involve automatic search and retrieval functions. I sought to compare the performance of these functions in each of the two categories. Assuming that the databases have been filled with identical texts, and given the same new text for translation, which approach provides the user with the most useful information?

DESIGNING A COMPARATIVE
EVALUATION methodology

A good evaluation methodology should have the following characteristics:
Validity: end users should be able to infer from the measurement values obtained what effect the tool will have on their productivity
Reliability: an evaluation should produce similar or identical results when repeated in the same context by different evaluators
Efficient applicability: the evaluation should be performable with the least effort possible, especially by the end users.
These were prioritized in that order. TRADOS andMultiTrans were chosen as representative tools for each approach, but the methodology had to be applicable to other tools as well. Finally, usefulness had to be defined as a measurable attribute.

WHAT IS A USEFUL MATCH?
For this project, usefulness is defined as a function of validity, time saved and time spent. Usefulness and timesavings are overlapping concepts. If a match is invalid in a given context, it is useless to the user and automatically wastes time. This is serious, since TM technology is supposed to accelerate the translation process. However, the inverse is not necessarily true. A valid match is useful if and only if it saves time in terms of research, typing or revision. Even a valid match is useless if it takes more time to find and retrieve it than to translate the segment from scratch.
In designing a methodology to compare the usefulness of output from sentence-based searches vs. CSB-based searches, it is important to account for the relationship between validity and time. For any given unit of text to be translated, each tool will produce one of five results:
1. No proposal
2. An invalid proposal (that necessarily wastes time)
3. A valid proposal that wastes time
4. A valid proposal that has no effect on time
5. A valid proposal that saves time
Category 5 is the only desirable result. Category 1 cannot be measured, Category 4 can be ignored, and Categories 2 and 3 must be penalized. The tool will require a certain amount of effort on the part of the user to generate results even in Category 5, and this 本论文由英语论文网提供整理，提供论文代写，英语论文代写，代写论文，代写英语论文，代写留学生论文，代写英文论文，留学生论文代写相关核心关键词搜索。