COMMUNICATING ACROSS LANGUAGE BARRIERS: LANGUAGE TRANSLATION TECHNOLOGIES

(ProQuest Information and Learning: ... denotes formulae omitted.)

Consider the following situations:

Scenario 1: A distraught Spanish-speaking offender visits a probation/parole office during a lunch hour and tells the duty officer, "Finalmente encontre el valor de decirle a mi P.O. que

Scenario 2: While doing his rounds at a county jail, a correctional officer is summoned by a Saudi resident who says:

"...

("Some of the inmates are planning a disturbance tonight. I'm afraid I'll get hurt."). Knowing an Arabicspeaking officer is scheduled to work the next morning, he places a note in the officer's mail bin advising him of the need to speak with this resident. The note is found several hours after officials regained control of the jail.

Scenario 3: After months of investigating a vicious leader of a Laotian drug cartel, narcotics detectives receive a break. They are told the brother of the target wishes to give a statement that will seal the case. However, the inability to quickly locate a Laotian translator stalls the investigation. The following evening, the target murders the brother after he falls under suspicion by the cartel.

Although language barriers are nothing new, as the story of the ancient city of Babel attests, events like these are now happening in the nation's criminal justice system with increasing regularity. Thousands of people cross U.S. borders each month hoping to find a better life (Rossmo, 2005). Unfortunately, not all the new residents are committed to a law-abiding lifestyle. Processing non-English-speaking defendants through the criminal justice system is a problem that has mushroomed during the past several years.

Typically, agencies have managed to get by with a few dedicated bilingual employees, who do their best to bridge the language gap. However, with an increasingly diverse population of immigrants entering into even smaller to mid-sized communities, criminal justice agencies are finding they cannot effectively communicate with their clientele. Agencies do not have sufficient bilingual staff available, and the current demand to converse in many languages makes staff's reliance on "office linguists" impractical.

Fortunately, there is an emerging technology that promises help. Although language translation systems have been available for many years, they have been used only marginally by criminal justice agencies - until recently. The early systems were awkward to use and not always effective, while the newest technology makes language translation amazingly simple and relatively affordable.

It is now possible to converse with someone who speaks a different language in real time, using bidirectional speech translation equipment that consists of a microphone-equipped computer with language translation software installed. The Probation and Parole Division of New Mexico recently tested such a system in Albuquerque. The results of the experiment are discussed later in this article.

A Nation of Immigrants

Immigration now may be one of the nation's most divisive topics. The extent of the issue was recently revealed when millions of immigrants and their supporters marched in the streets of the country's largest cities to protest restrictive immigration laws that were being contemplated in Congress. Images of the masses that were broadcast worldwide by every major news outlet clearly show that the ethnic and cultural composition of the U.S. population is rapidly evolving.

According to the 2000 U.S. Census, 21.3 million people living in the United States have difficulty conversing in the English language. This represents a startling 52.5 percent increase over the 1990 census data. Of those responding to the census, 3.3 million spoke no English at all. Considering that the census data is now six years old and that those who speak little or no English are very likely to be underreported in a census, it would be a fair assumption that the language barrier in America is even greater.

It is not surprising that many non-English-speaking U.S. residents find themselves in the criminal justice system. Non-English speakers have a harder time finding gainful employment, advancing their education and locating a stable residence - all factors that have been found to increase the risk of criminality. With this segment of the population growing so quickly, the criminal justice system has been severely impacted, with some agencies scrambling to find solutions to their language barrier crisis.

Early Translation Systems

Computer software for recognizing speech, translating text and producing voice output has been under development for several decades. The key goal has been and continues to be combining these language functions into a system that really works - an instant translation of anything that is said or heard at any time, at any location, for any language. As an initial step toward achieving this goal, early translation systems simply focused on translating text. Unfortunately, these initial systems often neglected to use context in generating the translation. For example, the word "right" can mean a statement of correctness (e.g., "You are right."), the authority to do something (e.g., "the right to vote") or a statement of direction (e.g., "Turn right at the corner."). It was common for these early systems to simply insert a word-for-word substitution using the most commonly used meaning (in this case, translating "right" as a direction), thus often constructing completely misleading translations. It was a start but certainly did not provide the accuracy needed for actual use.

An alternative approach implemented in early translation systems used a database of preselected phrases that had previously been translated by a trained interpreter. Here, the user is limited to using only phrases that have been entered into the system previously. When one of these phrases is selected, the preprogrammed translation is simply retrieved from the database. In general, this produces reliable translations, but works only for the phrases that someone else has previously entered into the system. In this second approach, increased accuracy has been gained, but at the cost of significantly reducing the flexibility of the system. The upside is an easy-to-use, low-cost system. The downside is an extremely limited amount of phrases. Examples of current systems that use this approach are multiple products from Lingua and Ectaco, plus Marine Acoustics' Phraselator and SpeechGear's Interpreter software.

Recent Technology Advancements

The U.S. military has long recognized the need for troops to be able to communicate directly with the local populace. As a result of this need, the military has been funding the development of new systems that are coming ever closer to meeting the stated goal - translating anything that is said or heard at any time, at any location, for any language. These systems are not perfect, but significant improvements have been made. SpeechGear's Interact system, for example, supports an English vocabulary of more than 200,000 words along with a foreign language vocabulary of around 50,000 words. These words can be combined into any sentence and translated. The context of the words is used to both identify what was said as well as to generate the appropriate translation. For example, in the sentence "Please write your name on the right side of the paper," the second and seventh words are pronounced the same. Context is used to determine if the proper transcription should be "write" or "right." The same applies to words such as to, two and too; or sea and see. With an appropriate transcript, the meaning of the sentence is identified and the appropriate translation generated.

The Albuquerque Test

The Probation and Parole Division of the New Mexico Corrections Department was eager to give the new technology a try. A border state with a large population of Spanishspeaking residents, New Mexico is challenged to provide appropriate services to its ever-growing population of non-English-speaking offenders. Having relied solely on bilingual probation/parole officers to handle Spanish-speaking cases, the division has found itself in a difficult position. Many bilingual officers correctly point out that offenders who do not speak English are much more likely to have critical needs. They often have difficulty understanding the court system and the requirements of their sentences. They also are more prone to be unemployed, less educated and living in substandard housing. Unless significant time is invested and great care is taken, these offenders are more likely to be rearrested.

The bilingual officers who supervise Spanish-speaking caseloads are given a small pay differential to compensate them for their much-needed language skills. Unfortunately, with tight budgets prevailing, the extra pay is inadequate to attract and retain these valuable employees. When speech-translation systems began to gain credibility, the division took steps to begin an evaluation process of the technology.

SpeechGear Inc., agreed to assist the division with a trial. The company, primarily a defense contractor that supplies military troops with speech-translation equipment, seemed eager to expand its focus to the criminal justice field. A trial was scheduled for June 2006.

The parameters of the study were rather straightforward: A bilingual officer in Albuquerque was asked to use the equipment when communicating with his Spanish-speaking caseload. He was told to refrain from speaking Spanish unless the translation system failed to communicate the concept. The trial was limited to the offenders who spoke Mexican Spanish, because the software used was engineered for this dialect. The officer evaluated the effectiveness of the equipment in translating each portion of his interviews by rating the communication as:

* Successful - No problems with the translation were encountered, and the concept was translated without any difficulty;

* Successful after rephrasing - The concept was ultimately translated appropriately after rephrasing the concept;

* Unsuccessful communication - The officer could not successfully communicate the concept, even after rephrasing; or

* Equipment failure - The officer was unable to communicate the concept due to problems with the equipment.

Perhaps the most difficult aspect of the trial was breaking down the communication into ratable units, which are defined as a portion of the interview that is a distinct part of the entire conversation. For example, if the officer queried the offender about a change of address and then proceeded to talk about his employment status, two ratable units were discussed. A ratable unit of conversation is brief, not lasting more than one minute, using everyday, freely spoken language. The use of slang or colloquialisms was avoided. If the initial attempt at communicating failed, the officer rephrased the concept, avoiding words that may have caused the problem. The officer had a scoring sheet at his desk that was used as the conversation progressed.

English-to-Spanish and Spanish-to-English communications were rated independently. For example, if the officer asked the offender if he had changed his address since his last report to the office, and that communication was translated effectively, it would be rated as successful. If the offender responded by saying he had moved into a flophouse in the barrio, the system may have trouble properly translating the communication. The Spanish-to-English portion of the dialogue would receive a different scoring.

Not all communication was rated. Brief discussions at the beginning of the interviews, describing how the equipment worked and why it was being used, were typically in Spanish. After the offender agreed to participate in the study, ratable units of communication commenced. Figure 1 is a sample scoring sheet used by the officer.

Trial Results

The 30-day trial provided interesting results. The English-to-Spanish translation was extremely effective, while the Spanish-to-English communication was sometimes problematic. There were several reasons for this disparity.

The probation officer assigned to the trial spent significant time training the software to recognize the nuances of his speech. Anyone who has used commercially available computer dictation software is familiar with this process. The user typically reads a lengthy script that helps the computer learn his or her speech patterns. As a result of this training, the computer learns the user's voice. The Spanish-speaking offenders who reported to the officer did not spend time training the computer to learn their voices. As would be expected, this contributed to an increased Spanishto-English translation error rate.

The probation officer quickly learned which English phrases the software had problems translating. By avoiding these phrases and relying on speech that was successfully translated during previous conversations, the English-to-Spanish translations improved quickly. Because the Spanish-speaking offenders had no previous exposure to the system, there was no such learning curve. Similarly, the probation officer learned the most effective rate of speech and the optimal distance to place the microphone from his mouth, while the Spanish-speaking offenders did not have this advantage. Finally, the probation officer became comfortable with the equipment and learned how to speak clearly, precisely and with minimal stammering. Many of the offenders seemed intimidated by the equipment, which resulted in an increased level of stammering.

Another factor to consider is the translation engine itself. The English-to-Spanish software can recognize more than 200,000 words, while the Spanish-to-English engine recognizes only 50,000 words. Because English-speaking users train the software to learn their speech, a larger vocabulary can be afforded. The fewer words of the Spanish-to-English software actually help improve translation accuracy. Because the Spanish-speaking offenders do not train the software to learn their speech, more errors are likely. If the Spanish-to-English software had a larger vocabulary, the translation error rate would actually increase because seldom-used words that are near homonyms to the intended word may mistakenly be chosen for the translation. The trick is to find the optimal number of words that produce the fewest errors. The nature of business in which the system is used must also be considered. In this trial, many corrections-specific words were unsuccessfully translated. For example, "P.O." would probably be translated as post office. The vendor who provided the equipment for the trial indicated that trade-specific terminology will be added to the software when a criminal justice agency becomes a customer.

The translation software engines are in a constant state of refinement, with some language pairs having had more time to mature than others. It is expected that the systems will only get better with time.

Finally, another factor affecting translation accuracy was the software settings of the microphone. It was discovered that a small adjustment to the gain (the sensitivity of the audio input) resulted in a significant difference in the system's effectiveness. High gain would include unwanted background noises and low gain would allow only the loudest sound through. For example, if communication is attempted in a busy metropolitan outdoor setting with cars rushing by and police sirens wailing, the settings would need to be adjusted to minimize peripheral noise. A quiet office environment would require different settings.

With 232 ratable communications evaluated during the course of the trial, 200 (86.2 percent) were deemed successful. Of the English-to-Spanish communications, 61.2 percent required no rephrasing to be successful (see Table 1). The Spanish-to-English portions of the dialogues required rephrasing 73.3 percent of the time. There were 32 instances when the Spanish-to-English communications were unsuccessful, meaning that after three attempts, the system failed to communicate the concept to the officer. There were no mechanical failures with the system during the trial.

When evaluating successful communication, the officer was not concerned with perfect grammar or appropriate word order. The only criterion was that a concept was effectively transferred across the language barrier. If the sentence, "I have bought a red car" was translated "Car red purchased by me," the translation, although awkward, communicated the concept and was rated as successful.

For this reason, it is important to use speech translation systems in appropriate settings. Where precise language is required, these systems are not yet up to the task. Court proceedings and inmate medical evaluations are good examples where professional translators are still required. However, there are hundreds of applications where the use of language translation technologies would be appropriate. Understanding the different systems' capabilities and limitations are the keys to success.

The social and cultural landscape of the correctional workplace is changing. In the criminal justice system, the ability to respond to these changes is critical. Communication is paramount. As society evolves into a more multiethnic mix, criminal justice officials must develop strategies that will keep diverse languages from being a barrier to reaching their goals and objectives. A language translation system is one solution that will help bridge the gap and topple that formidable "Tower of Babel" that has arisen in many locales.