What’s missing in User-centric MT?
18639Authors
nan
Abstract
This paper describes some of the kinds of pre- dictable errors in Machine Translation (MT). It then discusses means of alerting end-users of MT to the possible presence of such errors, including by providing training and/or by pro- viding automated MT ratings, MT color cod- ing and/or symbols, and footnotes and annotation. It also discusses the need for some kind of reliability measure and/or information to the MT consumer, and the likelihood of the MT user being open to using this kind of in- put. Some of the suggestions made for user- centric MT are also applicable to translator- centric MT. 1
Keywords
Full Content
What is Missing in User-Centric MT? Jennifer DeCamp The MITRE Corporation 7515 Colshire Drive McLean, VA 22001, USA jdecamp@mitre.org Abstract This paper describes some of the kinds of pre- dictable errors in Machine Translation (MT). It then discusses means of alerting end-users of MT to the possible presence of such errors, including by providing training and/or by pro- viding automated MT ratings, MT color cod- ing and/or symbols, and footnotes and annotation. It also discusses the need for some kind of reliability measure and/or information to the MT consumer, and the likelihood of the MT user being open to using this kind of in- put. Some of the suggestions made for user- centric MT are also applicable to translator- centric MT. 1 Introduction What is missing in MT? Some text may not be translated. Some relationships may be reversed. Some names may be wrongly translated. Some negatives may get lost. However, the text may read reasonably well, and the consumer may not realize substantive errors that may affect his/her understanding and decisions. This paper addresses some of the types of consistent meaningful errors and proposes means for communicating this varia- tion in reliability to the consumer of Machine Translation (MT) output. Some of the suggestions made for user-centric MT may also be applicable to translator-centric MT. 2 What is User-Centric MT? User-centric computing is a phenomenon that has emerged primarily in the last decade users are searching for, deciding on, and often translating the information they need. As Van der Meer ob- serves (1994), ―The one source of information pro- vided by the product manufacturer, the government, the doctor, or the hospital is now be- ing replaced by dozens—if not hundreds—of alter- native and competing information sources. Tips and tricks from other users, prescriptions from multiple healthcare organizations, and analyses of government data from private sources may be much more valuable than the ‗authoritative‘ infor- mation from the original publisher.‖ End Users Data Owners Translation These users then employ online MT to access the information in their language of choice: hence, ―User-Centric MT‖. MT End Users Microsoft was one of the notable pioneers of this approach for product literature, providing the MT developed for internal use at their company to premium users as a perk. The MT enabled the us- ers to translate and thus access larger sections of the online Microsoft website. Government organizations are also making greater use of User-Centric MT. Bemish (2008) observed that ―Using advanced tools like MT has Figure 1: Publisher-Centric Translation. Figure 2: User-Centric MT. allowed analysts and investigators to see data that would have taken years to translate and compile.‖ 3. What do MT Providers Do for Users? In the past few years, the Association for Machine Translation of the Americas (AMTA) and the MT Summit have provided an increasing focus at their conferences on providing tools for translators— primarily for post-editing—thus creating transla- tor-centric MT. However, little has been done to support the users who are not translators and who are utilizing …
Content truncated to 500 words
Labeled Text
This paper describes some of the kinds of predictable errors in Machine Translation (MT). It then discusses means of alerting end-users of MT to the possible presence of such errors, including by providing training and/or by providing automated MT ratings, MT color coding and/or symbols, and footnotes and annotation. It also discusses the need for some kind of reliability measure and/or information to the MT consumer, and the likelihood of the MT user being open to using this kind of input. Some of the suggestions made for usercentric MT are also applicable to translatorcentric MT. 1
Publication Details
- Paper ID:
- 18639
- Year:
- 2009
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan