塞翁失马: The Challenges Faced by Translation Software

Faced with today’s rapid globalization, it is no wonder that facilitating cross-cultural dialogue is at the forefront of many tech companies’ minds. Every day, it seems more and more start-ups are being funded with the promise that their patent-pending technology is going to change the way people from different societies interact. It only makes sense that these types of gadgets get the attention that they do: though the number is ever-shrinking, there are currently somewhere around 7,000 living languages spoken in the world, and no one speaks all of them. Even learning the 23 most common languages, which would put you in communication with more than half of the world’s population, seems an impossible feat. How, then, is one supposed to connect with others in our ever-diversifying society? The answer, some believe, may be more attainable than we think.

The Japanese company Logbar is one proponent of “wearable tech” aimed primarily at assisting travelers in their intercultural exchanges. Recently, this video has been making the rounds on Facebook and several other social media platforms, illustrating the use of the company’s latest pursuit, a device called Ili. This real-time translating device boasts the ability to instantly translate between spoken English and Japanese: simply press the record button and talk into the device, and it spits out the translated sentence(s) via a built-in speaker. According to reports, the operating system is completely self-contained, with almost 50,000 travel-related words and phrases stored directly on the OS, eliminating the need for a wifi or data connection.

“The future is here!” claims the video’s caption, and at first glance, perhaps that appears to be the case. To be honest, the Ili is not such a futuristic concept: it combines voice recognition technology that is now common in most smartphones and computers with the lexicon of software like Google Translate. The most obvious difference, of course, being that the data is spoken, rather than orthographic. It is certainly convenient (especially since the device’s dictionary is stored offline), but how practical is it? As a linguistics student and a speaker of other languages, I am inclined to think realistically about how the development of devices such as Ili may affect the field to which I have dedicated years of studies.

Ili and other artificial translators rely heavily on the fact that human language is systematic – that is, that there are rules and formats which govern the way that you and I express our ideas. That is the reason that we can understand sentences that have never before been uttered; for example, Your kitten station is full of horseradish. It is also the reason we can assign grammatical categories to meaningless words, like Lewis Carroll does in Jabberwocky. Even if we all had the same vocabulary, if everybody applied their own structure to that vocabulary, we would not be able to understand one another. This need for structure is what leads many programmers and linguists alike to believe that human language can, indeed, be qualified using technology. To an extent, they are right. With the right set of skills, one could indeed direct a computer to convert individual words and phrases in one language to their counterparts in another language. This is especially true for concrete nouns and verbs, such as tree, dog, run, or eat. Even more contextually-dependent pronouns could potentially be translated artificially, without necessitating a huge margin of error. Google Translate alone is proof of the power that technology asserts over the spontaneity of human language.

The real trouble, of course, lies not within the systemic aspect of language, but within the potentially infinite variability of these systems between languages and the way that each one represents the world around us. We have all certainly come across “untranslatable” words or phrases in another language – meaning, of course, that an idea which is qualified by a few syllables in one language requires many more syllables to convey in another (one need look no further than the Wikipedia article on the Spanish word duende for a prime example). You may have even stumbled across much more rigidly-defined vocabulary in another language  (one that comes to mind is the hotly-debated claim that a particular group of Eskimos have over 100 words for snow) that is more general in your own. These and many other examples are often cited as the downfalls of translation driven by artificial intelligence. However, even supposedly untranslatable words, with the right set of skills and ample time, could be systematically defined and, eventually, communicated in any given language (assuming that, allotted an infinite number of words/phrases, every language possesses the capacity to describe what may be encoded with only one syllable in another language). This is clear when looking at the list of Eskimo terms for snow: even if /tliyel/ is an idea that is not communicated using one equivalent word in English, it can be sufficiently described using the phrase “snow that has been marked by wolves.”  So, while language-specific words and phrases may not be the easiest task for translation app developers, they are certainly not insurmountable, given the sophistication of current technological advancement as well as the far-reaching abilities of coding.

The most difficult challenge faced by programmers and companies such as Logbar, in my opinion, is how to approach idioms and other contextually-dependent manifestations of language. Words, phrases, and even entire sentences that, in context, mean something entirely different than their literal interpretation are often the downfall of both second language learners and translation software, for the same reason: often, one must be intimately familiar with the culture encapsulated within the language in order to correctly interpret its idioms. For example, there is a Chinese proverb that roughly translates to “old man Sai lost [his] horse” (塞翁失馬), which refers to an old folktale in which an old man named Sai’s horse runs away. All of his neighbors offer their condolences, but Old Sai tells them that it is still unclear whether this loss is a good or a bad thing. Later, the horse returns with another equally fine steed following close behind, illustrating that losing his horse the first time was actually a blessing in disguise. How might one code this idiom to be displayed in English? Should it be replaced with a literally different but idealistically similar English idiom (e.g. “when one door closes, another one opens”)? Or perhaps the entire story behind the proverb would most clearly illustrate its intended meaning. Google Translate displays this idiom as “blessing in disguise” – but that may cause confusion for language learners. Unlike the individually untranslatable words mentioned above, there is no clear “right” way to approach the translation of idioms, especially those whose meaning is founded primarily in cultural understanding. Fundamental differences between the society in which one language is spoken and the society in which another language is spoken can create a disconnect between idiomatic expressions, even if they are perfectly translated. This is where our technology falters, as artificial intelligence still does not possess the capacity to deeply analyze the context in which a sentence is spoken.

So, AI-driven translations lack the ability to interpret non-literal language- what’s the big deal? It’s not like our day-to-day interactions make heavy use of idioms or proverbs. More often than not, we take for granted just how many non-literal expressions exist in our daily speech – I have used six in the last three sentences alone. We haven’t even talked about the mental processing power required for other types of non-literal language – it is much more demanding than one might think to identify the one or two traits that distinguish an object or idea, then apply those traits abstractly to another, as we do with metaphors. Oftentimes, such non-literal language is not indicated by anything within the linguistic structure (use of comparison words/phrases for simile being an exception in English), and instead is only identifiable by a disconnect between the content of previous statements and one’s own knowledge of non-literal phrases in their language. How is a computer or other operating system supposed to retain and sift through the relevant information to make these judgments?

All of this is to say that while devices such as Ili are indeed a technological feat from the consumer’s perspective, they may not be quite so groundbreaking from the linguist’s end. Because while language is, at its core, a systematic representation of human thought, it is also open-source. Anybody that has tried to learn a foreign language could tell you that grammar is rife with exceptions (I’m looking at you, Spanish conjugation) and that things most certainly aren’t always what they seem. They could also tell you that the best language learners immerse themselves not just in the grammar and the vocabulary, but in the culture(s) of the places where their target language is spoken, for precisely the reasons mentioned above: language and society reflect one another (which is partially why abbreviations like “LOL” and “OMG” weren’t popularized 100 years ago).

As someone who grew up frequently translating from Chinese to English for my parents, I can understand the appeal of translation devices; it is mentally taxing to switch between languages, much like performing complex math equations in your head. But my own experience is also the reason that I understand the limitations of such technology: while machines can indeed streamline many of the time-consuming processes required for basic linguistic analyses, they cannot answer questions about the nature of language, nor can they interpret the meaning behind human utterances with even 90% accuracy, whether intended for translation or other purposes. The good news is, this probably means I will not be replaced by a robot in any job I pursue with my degree. The bad news is, this means we are still several groundbreaking discoveries away from any concrete proof for many linguistic theories.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s