|
Possibilities of CAT
The use of CAT in software localization provides important benefits for translators and localizers. Besides improving text consistency and terminological coherence, assisted translation tools help to save time by recycling previously translated strings (leveraging). In addition, software can be completely localized using only one application, as it can be observed in our case study.
The target of translation tools is to improve translator’s performance when completing a given project: Therefore, CAT is not a threat for professionals, since the quality of the final output will be strictly linked to the skills and competence of human translators. Learning curves for using CAT tend to be quite reasonable and multi-faced applications (such as Passolo) can be handled in a short period of time (although some more extra time may be required to master it).
Obviously, relevant differences do exist among translation tools regarding not only functionality and usability but also other important issues (such as price, license conditions, etc.). The selection of a particular tool must be done in accordance with the specific requirements and necessities of translators. However, these applications clearly offer an advantage in order to achieve a truly localized product.
Types of CAT tools and their features
More recently, some interactive MT systems have begun to shift the user’s role from analyzing source texts or editing machine output to collaborating with the machine to produce target translations. The TransType project describes a pioneering system that supports word completion and next-few-word translation predictions. Along a similar line, Koehn [2009] develops a Web-based CAT tool, Caitra, which displays one phrasal translation at a time and offers alternative phrasal translations. The main differences between their work and ours are that we make no assumptions about the user skill and experience as a translator, and we further display grammar patterns to provide the general usages of predicted translations, allowing the user to increase his/her language proficiency.
For example, Barrachina et al. [2008] investigate the applicability of MT kernels of alignment templates, phrase-based models, and stochastic finite-state transducers within the IMT framework. Nepveu et al. [2004], Ortiz-Martinez et al. [2010, 2011] further leverage user feedback to improve their IMT systems. Nepveu et al. incorporate adaptive language and translation models trained on recent user in-put histories for better user experience, while Ortiz-Martinez et al. use user-validated translation pairs and an online learning algorithm to incrementally re-train MT models. Rather than be triggered by user correction, our method is triggered by word delimiters, such as spaces, and is designed to assist in source sentence translation and target language learning.
Albrecht et al. [2009] introduce a collaborative framework in which users who do not understand the source language are asked to correct MT translation errors with the help of a visualization of multiple language resources. Mediating between automatic system and users, we set up an environment for nonnative speakers of the target language. In contrast to the previous research in CAT, we present a writing assistant that automatically suggests translation texts in syntactic patterns based on users’ translation prefixes. The assistant iteratively collaborates with users, with the goal of reducing their effort to make their own grammatical and lexical choices and of enhancing their writing performance in terms of productivity and correctness.
Recently, many online systems have been designed to identify collocations for the purposes of lexicography and language learning. Among these, Sketch Engine [Kilgarriff et al. 2004] is the most famous, which summarizes a word’s grammatical and collocation behavior. JustTheWord4 clusters co-occurring words for single-word search queries, while TANGO [Jian et al. 2004] accommodates cross-language single-word searches. An interesting approach presented by Cheng et al. [2006] describes how to retrieve mutually expected words via concgrams allowing for constituency (e.g., “AB” in constituents of “ACB” and “ACDB”) and positional variants (e.g., “AB” and “BA”). In contrast, we go one step further to syntactically summarize the regularities of the contexts of a single- or multiword query and, at the same time, the context representation is not pre-determined and not limited to certain collocation types (e.g., adjective-noun, verb-noun, and verb-noun-preposition).
Do'stlaringiz bilan baham: |
