In less than three months, Christina Benson had to complete an enormous electronic discovery project for a 2005 antitrust case. To make matters worse, out of the millions of pages of text under review, one-third were in Japanese.
Normally for a task like this, Benson, an associate at Hunton & Williams in Washington, D.C., would contract at least a dozen translators to review the foreign language documents. But with the clock ticking, she had to find a quicker solution.
“Human translation is a bottleneck in time and cost because people can only translate so fast,” Benson says. “You also are paying them by the hour for that time.”
Benson decided to try translation software, commonly known as machine translation. Although the program she used didn’t provide a 100 percent accurate translation, it did the job well enough to allow contract attorneys to determine which Japanese documents were relevant to the case. Documents that the contract attorneys flagged as relevant went to human translators.
“We ended up using only about six human translators and saving many hours of translation time,” Benson says.
Machine translation isn’t perfect, and it may never replace human translation. But with increased pressure to manage the skyrocketing costs of e-discovery and shorter discovery deadlines, some companies are finding a new use for it.
“Machine translation can be spotty in terms of quality,” says Jorge Montanez, executive vice president of First Advantage, a litigation consultancy that offers translation software for legal departments. “But what it can allow you to do is quickly and cheaply make decisions about documents that are grossly irrelevant during document review.”
Deus Ex Machina
One of the chief benefits of translation software is that it is extremely easy for anyone to install and use. Most translation software works within Microsoft Office and Internet Explorer. A user opens a document in one of those programs and then identifies the source language–the language the text is in–and the target language–the language the text will be translated into. With the click of a button, the software automatically conducts a word-by-word translation of the document.
Lawyers, however, have been hesitant to embrace this technology. The word-by-word translations aren’t ideal for sensitive legal documents where accuracy is crucial. For example, when SYSTRAN, a popular translation product that contains 13 languages, translates “Ciudad de Mexico,” or “Mexico City” in English, the output is a literal word-for-word translation–”City of Mexico.”
“A document passed through translation software can’t be used as evidence,” says Nina Ivanichvili, CEO of All Language Alliance Inc., which provides human translation services for lawyers. “Not only would that document not stand the test at trial, but it might embarrass the party presenting the document.”
Over the past year, in-house counsel, however, have been realizing the potential of machine translation thanks, in part, to improvements in the software.
Most software packages have added built-in dictionaries. These are sets of preprogrammed, industry-specific terminology that can provide more accurate translations for certain fields, including law.
“Words have different meanings in different contexts,” says Reba Rosenbluth, a spokesperson for SYSTRAN. “By selecting a technology-specific dictionary, the software will know the word ‘chip’ in ‘computer chip’ is referring to computers and not a chip like in a chocolate chip cookie.”
But the major driver in the increased use of machine technology is e-discovery.
Under the new Federal Rules of Civil Procedure, which are set to go into effect Dec. 1, 2006, parties must meet soon after a filing to develop a discovery plan. Part of this plan is to hash out how long the producing party will take to turn over documents to the requesting party. Prior to the new rules, parties could wait several years before even beginning discovery. Because the parties must determine a time frame before discovery even begins, it’s crucial that nothing during document review slows down the process.
For example, a U.S. company gets slapped with a lawsuit dealing with defective car seats that a Korean subsidiary manufactures. The two parties meet, and the U.S. company determines it will take six months to collect, review and produce the requested documents. Now, the countdown begins.
The company must suddenly sort through millions of pages of text in both English and Korean. Without machine translation, the company could spend weeks waiting for translators to decipher the Korean documents, possibly missing its deadline.
But with machine translation, a contract attorney who is hired to review documents can specify English as the target language and Korean as the source language, click a button and generate a rough translation in a matter of seconds. From the translation, the attorney can scan the document for the term “car seat.” If the term shows up, the attorney can assume it’s relevant and pass it off to human translators for a more accurate translation, shaving off countless hours from the review process.
“Machine translation puts the focus on only the most relevant documents, allowing counsel to do what they need to do in the time frame they have,” Montanez says. For now, translation software’s use in the legal world is limited to document review. However, improvements are on the horizon that will allow more uses.
For the past decade, programmers and linguists have been working on what’s called “statistical machine translation.” This form of translation uses statistics to determine reliable patterns for word usage. This will increase the likelihood of a correct translation for words with multiple meanings that vary depending on context.
“We are moving in the right direction with the statistical model,” says Philip Resnik, associate professor at the University of Maryland’s department of linguistics. “I definitely think we’ll see increases in quality in the future.”
This will allow legal departments to use machine translation for more than just document review.
“When those word meanings and word group meanings get translated better, people will be able to search foreign language documents for specified terms more accurately and faster,” -Montanez says.
However, even as the technology improves, it will never completely replace human translators.
“For certain problems, I don’t think we will be able to remove the human from the loop,” Resnik says. “You need to ensure quality. If it would be catastrophic to translate a word incorrectly, then you will need a human being.”