Man v. Machine: Social and Legal Implications of Machine Translation

In a predominantly English-speaking country such as the United States, it can be easy to take for granted the essential relationship between the arts of law and translation. Yet, as David Bellos notes in Is That a Fish in Your Ear?: Translation and the Meaning of Everything, legal texts are translated in much greater quantities than typical literature, and interestingly, in more complex and various directions. Not only is the translation of law from one language to another “a prerequisite for the construction and maintenance of a global society,” as Bellos puts it, legal writing is already an act of translation in and of itself. Law in English, for instance, isn’t really English at all anymore; legal language often takes on different connotations from common parlance and is frequently an amalgamation of clearly non-English terms. The self-referential and essentially enclosed language of the law is what leads Bellos to conclude that it is inherently untranslatable, though by necessity the object of much translation.

As the future of legal translation undoubtedly lies–at least in part–in machine translation (MT), there is potential for valuable increased efficiency as much as costly detours: instances of “getting lost in translation.” In order to understand what increased adoption of MT means for the legal world, particularly in the United States, it is important to look back on the history of US Legal Translation, to look across the Atlantic to the complex yet streamlined translation infrastructure practiced by the European Union, and look to the future as cases regarding MT use at our southern border are beginning to cause courts to consider the implications of MT on consent.

US Legal Translation

Louisiana

Even before the enhanced globalization of the 20th and 21st centuries, the United States, perhaps surprisingly, already had a rich history of legal translation, particularly in Louisiana. In fact, Louisiana’s first constitution was written in French before it was translated to English, although it held that both versions had equal authority. This system of enacting English and French laws of equal authority remained in place for 35 years until the state’s second constitution, in 1845, held only English law to be official in the future as Lousianna became a state. For this reason, the Civil Code of 1870 appeared in only English, though the issue of conflicting translations remained. Conflicts in translation can arise for numerous reasons: from outright mistakes, ambiguities arising from the structure of a language, or from the fact that there often is no such thing as a perfect match from one word to another between languages (Bermann 91-93). For instance, Shelp v. National Surety Corporation, a case regarding whether a lessor was obligated to repair doors on his leased property, dealt with a discrepancy between the Article 2716 of the Civil Code of 1870, which appeared only in English, and its parallel in the French-language provision of the 1825 code:

“The repairs which must be made at the expense of the tenant are those which, during the

lease, it becomes necessary to make … [notably] to windows, shutters, partitions, shop

windows, locks and hinges, and everything of that kind, according to the custom of the

The 1825 French version, however, had included portes (“portes, croisees, planches de cloison et de fermeture de boutique, gonds, tarjettes, serrures et autres, suivant l’usage des lieux”), which did not make its way into the English version. While doors could have been reasonably construed as belonging to the category of things mentioned in the English version, the court came down on the side of the lessee on account of the greater authority of the French original text. According to the court, while English-language law was to be generally dispositive, the French version of the 1825 code was authoritative because the Civil Codes of 1808 and 1825 were enacted in French before being translated into English. The decision to honor the French original was especially validated since, as the state Supreme Court found, “the English translation of the French text of the Civil Code of 1825 and, for that matter, the Code of Practice of 1825, was spectacularly bad.” The Court also quoted from Edward Dubuisson in the decision:

“Even where the translations do not contain misleading errors, the vigor, the spirit, the

clarity and finish of the originals are lost in the translation.”

If a team of human translators, even while producing an error-free translation, miss “the vigor, the spirit, the clarity and finish of the originals” how can we expect MT to capture these expressly humanistic qualities of legal writing?

Consent

Moving beyond the not inconsequential aesthetics of “vigor” and “finish,” research has found that MT use in high risk settings such as court and legal proceedings, despite its intention to enhance accessibility, can exacerbate existing inequalities. For instance, two recently reported cases revolved around US transport police officers using Google Translate to gain the consent of Spanish-speaking individuals to search their vehicles. Both cases resulted in the officers charging the individuals with a crime after finding illegal substances in the vehicles; however, the use of Google Translate to gain consent was challenged in court as an insufficient method of overcoming the language barrier, to differing degrees of success. In one case the motion was dismissed, yet in the other, the evidence was suppressed. These cases highlight both the lack of uniformity with which courts address issues of MT and consent as well as the pressing relevance of MT.

While there is a mixed awareness of MT technology for legal use, states with significant Spanish-speaking populations such as New Mexico have begun to consider the technology more thoroughly. As Vieira et al. find, “The state court of New Mexico is an example of an institution that has considered MT in more detail. It has a track record of appointing non-English-speaking jurors and has provided MT use guidelines in relation to these appointments. The guidelines state that unedited MT should not be used for materials expected to fulfill a formal role, for example in court proceedings or as exhibits.” As MT only promises to become more prevalent in legal proceedings, it is important that these and other solutions, as will be discussed below, are carefully considered by the courts.

Business

Another incredibly important realm of legal translation is that of business contracts, particularly in the case of US-China deals and translations, where the source and target languages are markedly different. If there is an ambiguity or a lack of a clear match in the target language, the translator may have to choose, and thus limit unbeknownst to the reader, a facet of the original meaning. According to Torbert, an expert in this field, “Even if translators can convey the ambiguity in the Chinese target-language document, a Chinese reader may interpret the ambiguity differently from the reader of the English source-language document. As a result, ambiguities can lead to serious misunderstandings.” To give an idea of the pervasiveness and seeming inescapability of such ambiguity, Torbert organized Chinese source language ambiguities into a daunting list of categories: singular or plural, verb tense, inclusion or exclusion in numbers and dates, conjunctions, abbreviations and English source language ambiguities into nouns and pronouns, prepositions, conjunctions, synonymous adjectives, adjective and adverbial phrase modification, negation, and post-modification. To provide an example of post-modification, for instance, Torbert details that

“For example, in the English translation of the Catalogue Guiding Foreign Investment in

Industry, an entry in the ‘encouraged’ category states ‘apparatus, instruments, and

machines for cultural and office use in the manufacturing industry.’ The ambiguity is whether the phrase ‘for cultural and office use’ modifies only “machines” or ‘apparatus,

instruments, and machines.’ In the original Chinese text, the phrase precedes ‘machines,’

so it clearly modifies only ‘machines’.”

These ambiguity-inducing linguistic elements are clearly essential to legal writing. The answer then lies not in avoidance, but awareness, something that must apply to MT as well, but is arguably more difficult to get a machine to appreciate algorithmically.

European Union

Beyond the scope of the US, the European Union’s parity system among its 24 official languages, in which each language version carries the authority of an original and no document is released until all translations are complete, is possibly the most impressively scaled feat of legal translation with 552 bilateral combinations. Because translation is not only cross-linguistic, but cross-cultural, additional issues arise in that the legal systems of EU member states themselves don’t align exactly, making perfect translation matches impossible. Furthermore, negotiations of course must occur in a language, creating an essential asymmetry and subtle prioritization of official and working languages. Perhaps filtering negotiations through the black box of MT, rather than the “procedural” languages of English, French, and German, could prevent this prioritization, yet in its current state, such technology would almost certainly obscure to some degree both meaning and the essential human to human understanding, even trust, that is necessary for successful negotiation.

Social Implications of Machine Translation

Law, clearly in the case of the US and more globally, has been translated and trans-linguistically negotiated more or less sufficiently over the years out of strict necessity. The laborious task of translating and negotiating terms, on a semantic as well as legal level however, is far from efficient. In order to save time and money, many firms may refuse cases brought up by a non-English speaker; or in cases where firms accept such cases, the additional time and money spent finding a translator and generating appropriate translations of documents often proves to be an unfeasible financial burden on the non-English speaker, which is especially detrimental if he or she is of an already marginalized class. In an effort to alleviate these financial burdens, speed up the process, and improve accessibility of legal services, some firms have turned to MT; however, MT is accompanied by its own set of significant advantages and disadvantages. The specialized even idiomatic vocabulary, the self-referential nature of legal texts, and the context dependent connotations of words differing from normal use are not only key components of legal writing but are also exactly the areas in which MT services such as Google Translate, which is trained on an enormous corpus of which legal texts make up only a fraction, are least accurate. For this reason, legal-specific MT software has emerged, yet its increased accuracy also comes with the risks of hyper specialization, as translation software designed for one sector of the law will lose accuracy if applied to another, and lessened accessibility, as subscription models are used to provide the revenue to train and maintain such specialized models, unlike free services such as Google Translate.

Legal Implications of Machine Translation

Between 1980 and 2000, while the US population increased by 25 percent, the number of Americans speaking a language other than English at home nearly doubled. Thus, the malpractice issues posed by lawyers dealing with non-English documents are more relevant, but also with the growth of MT technology, potentially more addressable than ever. While there is a precedent for regulating interpreters in court, no such thing exists for translators or translation systems as of yet. Investing in machine translation software has recently proven to be more efficient, both in terms of saving time and money, than hiring human translators, yet the issue arises that lawyers are unable to verify accuracy of output themselves and are likely unaware of or even unlikely to understand the exact complexities of machine translation software behind the user interface. In addition to the dangers of a blind reliance, use of some MT systems such as Google Translate can also breach client confidentiality, sharing client information or data with the MT provider. Furthermore, consequences of poor translation can be detrimental to individuals as well as court systems more broadly: For instance, in the mid 1900s a mistranslation of some German phrases in a European Court of Justice judgment ultimately resulted in over 200,000 suits being inappropriately filed in German courts.

In her discussion of potential malpractice issues arising from MT use for preparing legal documents, Wahler cites the 1932 T.J. Hooper case which held the defendant responsible for damages resulting from tugboats lost in a storm for failure to equip them with proper radios as precedent that under the standard of care sufficiently available modern technology can be reasonably expected to be employed. Thus, as non-English speaking individuals have a right to gain representation and understand their case proceedings as enabled by translation, lawyers have a responsibility to make use of modern technology as safely, effectively, and informedly as possible.

Solutions

While the impacts of MT in legal translation, and human-conducted legal translation more broadly, are undoubtedly far reaching, from challenging consent to enhancing accessibility to expediting business proceedings, clear and necessary solutions, as Whaler advocates, include increasing regulation of translation providers, as is the case for interpreters and is substantiated by President Carter’s 1978 Court Interpreters Act, promoting a hybrid model of initial MT then review by a human translator, and certifying law-specific machine translation services for enhanced accuracy and confidentiality.

In the specific English-Chinese business translation realm, Torbert suggests back translations, or memos highlighting preexisting ambiguities, and comparing the lengths of paragraphs to check for obvious additions or omissions (Chinese texts are typically three-quarters the length of English texts). Offering a front-end, pre-translation solution, Torbet also suggests that “English speakers can help their translators by writing precise and unambiguous English. These extra steps can be time-consuming, but they can save foreign businesspeople from spending even more time and money dealing with the serious consequences of unintended misunderstandings” (53). Although certainly useful in a business context, these solutions don’t scale across all translation issues in the US, indicating a need for categorized regulation based on the type of litigation.

Other solutions include adopting ABA’s recommendations regarding translation, specifically Standard 7 drafted by ABA’s Standing Committee on Legal Aid and Indigent Defendants:

“To ensure quality in translated documents, courts should establish a translation protocol that includes: review of the document prior to translation for uniformity and plain English usage; selection of translation technology, document formats, and glossaries; and utilization of both a primary translator and reviewing translator.”

Notably, this recommendation recognizes the reality of machine translation, the need for human review, and the benefits, when applicable, of drafting the English original in a more easily translatable form, or at least identifying potential problem areas, which Torbet argues for as well. Additional steps to ensuring a more equitable use of MT in legal circumstances include requiring consent forms in the speaker’s first language before using Google Translate to protect client confidentiality, requiring some degree of education for lawyers regarding translation and machine translation services, developing an agency to regulate MT use in the law, and requiring proof of human translator review. As the world becomes more globalized, multilingual, and technologically advanced, the issue of machine translation and the law will only continue to grow in relevance.

Robert Frost remarked that “poetry is what gets lost in translation;” we have a responsibility to ensure that legal meaning, that individual rights are not lost in translation as well.

Share this: