What is automatic translation? How accurate can it be?
As the world has become more interconnected than ever before due to the rise of the internet and the digital age, there are more situations than ever before in which people must communicate with others who speak a different language than them. Whether it be in the context of a business trip or trying to order food from a popular local restaurant, the ability to be able to translate a language into another is more valuable than ever before. While the work of translation has historically been done by people who underwent specific training and education, there are software programs that will do this work automatically in a fraction of the time.
Automatic translation works by substituting a human translation for machine learning and AI capabilities. These machines typically translate words by using other words nearby as a reference point and making a substitution in one language that is as close as possible to the language being translated. However, this is often at the expense of accuracy, particularly when it comes to complex words, sentences, or statements. This is due in large part to machine’s inability to pick up context in when analyzing certain phrases and sentence structures. As language has evolved over thousands of years, some of the nuance and specificity that is conveyed when speaking to another person will be lost when a machine is translating.
What is automatic translation?
Automatic or machine translation is defined as a subfield of computational linguistics that focuses on the use of software to translate text or speech from one language to another. On the most basic level, automatic translation works by performing a mechanical substitution of one word in a given language to another word in the language being translated. For example, if the Spanish word for car is “carro”, an automatic translation program would automatically swap out these words when performing translation. However, this process alone will not provide a thorough and comprehensive translation in the vast majority of cases, as translating from one language to another is obviously more complicated than simply identifying similar words.
As such, neural and corpus statistical techniques can be used to improve upon the machine learning process, such as the translation of idioms, handling differences in linguistic typology such as subject verb agreement, and isolation of various anomalies. What’s more, some automatic translation software offerings will also allow users to customize their results based on a particular domain or profession, such as weather reports, so that overall output can be enhanced by limiting the scope of allowable substitutions. As machine learning software often struggles to pick up on context, this is particularly useful in domains where academic, technical, or formulaic language is used.
Pros and Cons of automatic translation
The primary pro of automatic translation is the time and money that is saved when compared to enlisting in the services of a human translator. As human translators will be looking to pick up all the nuances and subtleties contained within a given language in order to provide the most comprehensive translation as possible, this process will take a lot time and effort. This is compared to an automatic translation software option that will complete the process in a matter of minutes. Alternatively, the cost of hiring a human translator may not be economically feasible depending on the project or work that is being completed. Generally speaking, human translators can charge anywhere between $30 to $250 per page of translation depending on the skill, expertise, and experience of the translator in question.
Alternatively, the primary con to using automatic translation is the decreased accuracy when compared to that of a human translator. This is even more pronounced for languages that may not be commonly spoken or referenced in the English world such as Malay or Farsi, which can have automatic translation rates as low as 45%. As human translators are trained professionals with years of experience, they have a level of prowess and proficiency that will never be completely duplicated by a computer program. In instances where a translation must be 100% correct and free of errors or missed context, a human translator will prove to be the superior option.
See the example below of 100% automatic translation without any human involvement:
How accurate can automatic translation be?
The accuracy of a particular machine learning software offering will largely depend on the two languages being translated, as well as the length and complexity of the passage being translated. As an example, Google Translate works very well when translating basic sentences like “I love you” from Spanish to French, as both languages are romance languages that derive from similar linguistic sources. Furthermore, I love you is a very basic sentence with a straightforward context. Conversely, translating a paragraph about mechanical engineering from English to Mandarin would prove extremely difficult for an automatic translation program to handle, as Mandarin and English are two languages that developed completely independent of one another and mechanical engineering is a very complicated subject matter.
To use Google Translate as an example, a study that evaluated over 400 emergency discharge instructions across various languages found that Spanish translations had a 90% accuracy rate, while languages such as Tagalog, Korean, and Chinese had accuracy rates ranging from 80% to 90%. Alternatively, Farsi translations only registered a 67% accuracy rate, while Armenian translations ranked at a 55% accuracy rate. In one example, Google Translate turned “You can take over the counter ibuprofen as needed for pain” into Armenian as “You may take anti-tank missiles as much as you need for pain.” What’s more, even languages such as Spanish that had a 90% accuracy rate still struggled to understand the context of certain sentences.
In many instances, a combination of both automatic translation and manual translation will yield the greatest overall results. For example, a college professor may be looking to translate a lecture in Russian into English so that their students can gain a better understanding of a topic being discussed in the particular class. This professor could first use automatic translation to pick up the most important words and theories put forth in the lecture. Additionally, the professor could also hire a human translator to go over this automatic translation, to ensure that everything that was said in Russian was translated into English in the most effective and efficient manner as possible.
As linguistics is a vast and ever changing space that encapsulates the emotions, feelings, and sentiments of an entire culture over a long standing period of time, it is impossible to capture everything without the use of a human translator. As such, those who are looking to translate long passages or are seeking translations in a language that is very different from their own will struggle to have their needs met through automatic translation. Nevertheless, automatic translation works very well with similar and widely spoken languages like Spanish and French when using basic sentences. Additionally, automatic translation continues to improve every day with the goal of becoming as accurate a possible.