At Captionmax, we believe that good, quality translation is an art. Translation is not science or math; you can’t simply replace one word in a language for the other in the destination tongue. Machine translation can be a helpful tool in the process, but human input is still invaluable in creating quality subtitles. Here, we’ll outline a few reasons as to why the human touch is still necessary in the localization process.
Machines may be getting better at connecting phrases and identifying meaning, but they aren’t particularly good at knowing when to apply the correct phrase. Humans can translate the cultural meaning of a sentence from one language to the other. When you take into consideration that words can have several different connotations or meanings, it is important to have a human ensuring that the best style and word choice are chosen for the translation to be effective. According to Westman (2021), “Even languages like Spanish and Chinese that were usually accurate could have Google Translate errors that could confuse [medical] patients. An instruction for a patient taking the blood-thinning medication Coumadin read ‘Your Coumadin level was too high today. Do not take any more Coumadin until your doctor reviews the results.’ It was translated into Chinese as ‘Your soybean level was too high today. Do not take any more soybean until your doctor reviews the results.’”
According to Popel, et. al, 2020, machine translation tends to work best in unambiguous and matter-of-fact writing scenarios, such as political news and business articles. Nature also found that translation engines tend to perform poorly on documents with more nuanced lexicons, like creative writing and sports news.
Humans are very good at understanding context. When you break down human speech to the base maxims, people can make assumptions based on logical steps and situational intuition. Machine translation struggles the more specific that context becomes. This happens because machine translation is much more meticulous in its approach to contextualizing: it starts at the word level, then broadening its scope to the segment the word is contained within. After that, it expands to the full sentence, onwards to the paragraph and entire document to figure out the context.
For example, let’s take a look at the Spanish word “pendejo.” In Mexico, the word “pendejo” means “idiot” or “stupid”– clearly an insult. However, in Chile, the word “pendejo” can be used to refer to young kids, or someone who is a bit immature. The word also holds the same meaning in Venezuela. Cultural context matters when translating, and humans are built to understand and shift meanings between contexts.
Translating slang, figures of speech, and colloquialisms present a slew of issues that machine translation engines aren’t equipped to handle alone. For starters, how do you convert English slang to a new language in a culturally sensitive manner? Translating isn’t just about converting words and grammar– you’re also translating culture.
Machine translations have a hard time recognizing colloquialisms. For example, English phrases like hit the hay or break a leg lose their meaning when translated literally. According to Captionmax translator Eladio Canibano, in Spanish, most of these expressions have counterparts, but they might not be used in the same way. Therefore, it is important to understand the meaning first before finding an appropriate translation.
Grammar and Punctuation
Grammar and punctuation in translation benefits greatly from having humans involved. Grammatical structures vary from one language to another, and machines have a hard time figuring that out. According to Johnson (2021), “But they [machine translators] struggle with sentences that are difficult to analyse, precisely because they are ungrammatical…”
These nuances are easily spotted and accounted for by trained subtitlers and captioners.
For instance, machines don’t always know if they “you” is used in a singular context or a plural context, or when the Spanish adjective must be masculine or feminine. In areas such as grammar and punctuation, is it important to remember that we are working not to translate words, but to translate messages.
Presentation is another area in which humans add value to the localization process. With subtitles, you are limited by physical space, as you can only fit so many words on-screen at once. Most machine translation systems are not trained to work within such limits. Even the specialist systems that try to accommodate space considerations very often fail to capture the full meaning/intent of the dialogue in the space available. Human translators are skilled at creatively crafting subtitles that can fit within these space restrictions, as they know that there are only two lines of text with a character limit they can work with. Translators have the ability to carefully paraphrase and condense the translation in order to fit the subtitles on-screen. Machine translations routinely fall short of human translators when it comes to keeping a consistent look and feel for the subtitle delivery.
All of this isn’t to say that machine translation is useless– on the contrary! Translation engines are an incredibly valuable tool when managed by a trained human translator. Their ability to quickly parse text in one language and simultaneously translate it into multiple different languages is something that the human brain is incapable of doing. When it comes to crafting quality subtitles for media, however, human translation will always manage to beat the machines.
For more information, check out our webinar on The Art of Media Localization!