NONPROFIT ORG. U.S. POSTAGE PAID May–June 2010 CHICAGO, IL Volume 44, Numbers 1–2 PERMIT #7706 50 East Washington Street Suite 500 Clearinghouse REVIEW Chicago, Illinois 60602 Taking action to end poverty Volume 44, Volume Numbers 1–2 ASSET BUILDING May–June 2010 by People with Disabilities Local Prosecution of Real-Estate Fraud Rights of Limited-English-Proficient Individuals 1–92 Machine Translation of Legal Information Toward Veterans Courts Language Access in State Courts The September–October 2010 Clearinghouse Review Mapping to Show Discrimination Unemployment Benefits Appeals Will be a Special Issue on Postracialism or Targeted Universalism? Climate Change and a Green Economy New Advocacy Opportunities How Effective Is Machine Translation of Legal Information? By Michael Mulé and Claudia Johnson Michael Mulé nline services offer instantaneous translation of text or website content. Staff Attorney Before using translation software or “machine-translation” tools to assist Empire Justice Center Olimited-English-proficient (LEP) clients obtain meaningful access to the 1 W. Main St. Suite 200 information, advocates should consider several factors.1 Rochester, NY 14614 585.295.5724 [email protected] Machine translation is the automatic translation of text from a source language into a target language without human intervention. Machine translation generally takes a dic- Claudia Johnson tionary approach to language and does not factor in culture or custom. Here we describe Local Initiatives Coordinator the role of a human translator, discuss when and how to use machine translation in con- junction with human translation, and review available machine-translation services. Pro Bono Net 2414 Lariat Lane Richland, WA 99352 Human Translation 509.396.7934 [email protected] A translator converts written material from one or more source languages into a tar- get language. Translators must have excellent writing and analytical ability, knowl- edge of the vocabulary, context, and meaning of words in both languages, and ensure that the translated version of the text conveys with precision the idea and form of the original. A description of interpreting and translating as an occupation notes that “[t]ranslat- ing involves more than replacing a word with its equivalent in another language: sen- tences and ideas must be manipulated to flow with the same coherence as the source document so that the translation reads as though it originated in the target language.”2 This characteristic distinguishes human from machine translation, and it is why hu- man translation is usually the only option in the following circumstances: 1Meaningful access is attained when a translation accurately communicates to a limited-English-proficient individual the programs and services of an organization or agency (see U.S. Department of Justice, Guidance to Federal Financial Assistance Recipients Regarding Title VI Prohibition Against National Origin Discrimination Affecting Limited English Proficient Persons, 67 Fed. Reg. 41455, 41461 (June 18, 2002), http://bit.ly/bH3Jpe). 2U.S. Bureau of Labor Statistics, Interpreters and Translators, in Occ U P ATIONAL OUTLOOK HAND B OOK , 2010–11 EDITION (last modified Dec. 17, 2009), http://bit.ly/9BWpez. 32 Clearinghouse REVIEW Journal of Poverty Law and Policy n May–June 2010 How Effective Is Machine Translation of Legal Information? ■ Nuanced vocabulary—documents where be unacceptably inaccurate.”4 Before us- the source text or vocabulary is ambigu- ing machine translation for any text or a ous or unclear. Documents that have document, use the following suggestions text with nuanced meanings require the to improve accuracy and efficiency: skills of a translator who understands ■ the context and message behind the Most important, limit sentence length. words. Legal terminology is nuanced Sentences of more than twenty-five vocabulary since concepts do not have words often become ambiguous and too the same meaning in all legal systems. complex for machine-translation tools to translate correctly. Keeping sentenc- ■ Individualized documents—personal- es to no more than twenty words or so ized letters and documents with con- will improve the quality of the output. cepts and specialized terms. A human However, machine translation is not translator knows the vocabulary and very good at translating the names of can convey the ideas and intent of the legal documents because they include source text. legal jargon that may not be part of an- other legal culture. Short phrases such ■ Sufficient time—if the written text does as “motion for replevin,” “petition for not have to be translated immediately dissolution,” or “request for restitu- and no detriment to the client will re- tion” still require human translation to sult. Even if time is short, however, be incorporated accurately into a docu- use machine translation only for indi- ment, website, or form. vidual words or small phrases, not as a replacement for professional transla- ■ Avoid metaphors, jokes, slang, puns, tion. idiomatic expressions, and regional or national expressions. Since these are ■ Official translations—to ensure accu- often translated literally, they tend to racy when a certified or official trans- lose their meaning. The literal transla- lation of a document is necessary in a tion of “break a leg,” for example, will legal proceeding, government agen- not make sense to the target reader. cy, or application process.3 Machine translation cannot ensure the preci- ■ Avoid abbreviations, acronyms, and sion required for translation of vital contractions, which might not have documents that affect a client’s rights, equivalents in different languages. benefits, or services. Instead spell out the entire word. Machine-translation services do not Machine Translation always recognize abbreviations and may omit them from the translation. Unlike human translation, machine Instead of “Sr.,” “Jr.,” “DHS,” or “SSI,” translation merely replaces a word with use “senior,” “junior,” “department of its target language equivalent, without human services,” “supplemental secu- considering context or meaning. Trans- rity income.” lated words without context do not main- tain the idea or message of the original ■ Use simple, direct sentences with basic text. grammatical construction. Ensure that the sentence structure is grammatical- Agencies that rely on machine transla- ly correct and do not omit words (e.g., tion alone to ensure that LEP individu- “Make sure that you use grammatically als have meaningful access to document correct sentence structure” rather than and website content should know that “Make sure you use grammatically cor- machine translation has “been found to rect sentence structure”). 3See FED . R. CIV. P. 44(a)(2), http://bit.ly/9c4xpf. 4Federal Interagency Working Group on Limited English Proficiency, Top Tips from Responses to the Survey of Language Access Strategies Used by Federal Government Agencies (2008), http://bit.ly/chlup5. Clearinghouse REVIEW Journal of Poverty Law and Policy n May–June 2010 33 How Effective Is Machine Translation of Legal Information? ■ Avoid ambiguity. To produce a clear material for meaning to that target translation, minimize use of words and community. A larger group triggers sentences that have multiple meanings obligations under Title VI of the Civil (e.g., the word “right,” which can mean Rights Act of 1964 to ensure meaning- “correct” or “opposite of left”). ful access by translating vital docu- ments and thus requires a systemic ■ Avoid compound verbs, which are of- approach to implementing quality ten mistranslated. control in translations.7 ■ Use the international standard date ■ Importance of agency reputation or format (YYYY-MM-DD) when writing trust—because its results are not reli- dates, the format for which varies from ably accurate, using machine transla- country to country. Using the interna- tion can signal to the target group that tional standard format will help elimi- its needs are unimportant. nate translation problems. ■ Volume of material and degree of rep- ■ Because some languages do not use etition—a large quantity of material, the present participle verb form, use particularly if vocabulary and concepts the infinitive form instead (e.g., “click are repeated, is ideal for machine here to select the icons and to view the translation because programs usually images” rather than “click here for se- contain dictionaries that can be cus- lecting the icons and viewing the im- tomized and updated as needed. ages”). ■ Feasibility of human translation—if the ■ Give the translator a list of words that volume of material is extremely large, should remain in the source language time is short, and frequent updates will (e.g., proper names and titles and 5 be necessary, human translation may names of benefits and agencies). be unrealistic. When to Use Machine Translation ■ Possibility of “gisting”—the practice of using machine translation to get a While machine translation cannot re- rough idea of the source text content is place human translation when quality called “gisting” (from “to get the gist and accuracy are crucial, it may be suit- of”); it can be effective and appropri- able for several forms of written content ate when an “official” translation is not or to supply an initial translation that a needed and to determine if a human human translator then reviews for accu- translation is necessary. racy and context.6 Take the following fac- tors into account before relying solely on ■ Extent of dissemination