Human translation versus machine translation – the first demonstrating a kind of artistic precision and the other quick efficiency – one of the oldest battles in the books (or at least we think so). As of late, the two have been reconciling due to increased sophistication with machines’ ability to process language, and the need for quick turn-around time for some translation projects. Now, thanks to advancements in the field of Computational Linguistics, Machine Translation may soon reach a level of precision before only known to human translators…
Computational linguistics is a growing field at the intersection of linguistics and computer science. A focus is placed on analyzing the mathematical and structural properties of language, with many computational linguists focused on the development of cognitive models that represent the way language learning takes place in humans. The goal of this focus is to improve automatic translations and to make interactions between humans and computers more seamless. Strides towards effective machine translation have been made recently; Google’s June 2016 patent application for neural network machine learning can be applied to advanced machine translation that will “learn” through trial and error in a way that attempts to mimic human language acquisition. Giving computers the power of language will advance the quality of interaction between man and machine, which many hope will one day be seamless.
Evaluating the Utility of Vector Differences and the Application of Word Math
A paper released last year by researchers from the University of Melbourne, Department of Computing and Information Systems, and the University of Cambridge Computer Laboratory showed progress in large-scale analysis of words and their relationships to one another. In this particular case, words were analyzed for their co-occurrence with other words. For example, “Olympics might appear close to words like running, jumping, and throwing but less often next to words like electron or stegosaurus” (MIT Technology Review). The set of relationships can be represented by a multidirectional vector, – which is a quantity with both direction and magnitude that determines the position of one point in space relative to another – representing the way a given word is used in a language. The novelty of this study results from the newfound ability to treat a language like a vector space with mathematical properties that can be processed by a computer to provide a clearer picture of the language. A particularly fascinating application of this analysis is something that can be described as word math. For example:
king – man + woman = queen
This seems like a logic puzzle, but it turns out there is a mathematical basis for it. If you compute the vector values that would be assigned to each word in the equation above, the result is the vector value unique to queen. Relationships between present and past tense verbs and singular and plural nouns, among others, have also been analyzed and produced consistent vector results.
Implications for Machine Translation, Interpreters and Translators
Though the paper did not propose any concrete applications for the future, there are many. This would greatly simplify the study of unfamiliar languages that typically require linguists to immerse themselves in a linguistic community until they can report on their findings. Instead of this time-consuming and involved process, a computer could simply analyze speech and texts to produce statistically significant insights. Additionally, uncommon language translation could be more accurately performed by a machine because; with this type of machine learning the meaning of the text is not actually processed. Given the precision of this new technique it is not necessary to have somebody with advanced knowledge of both source and target languages.
At this stage it is not possible to use machines to translate between languages with different cultural expectations, leaving professional human translation the only option to accurately convey the intended meaning to the target audience. Localization services will not be impacted by this kind of machine translation since they demand an awareness of both the target language and the intended community. While machines are undeniably powerful tools making rapid progress, the best products still need a human behind the work.
Written By Kayla King, Marketing Intern
- SemanticScholar.org – “Take and Took, Gaggle and Goose, Book and Read: Evaluating the Utility of Vector Differences for Lexical Relation Learning”
- Technology Review.com – “King-Man + Woman = Queen: The Marvelous Mathematics of Computational Linguistics”
- Stanford University – “Computational Linguistics”
About Language Connections:
Language Connections is a language service provider. We specialize in technical, medical and legal translation, simultaneous and consecutive interpreting, website and software localization, and corporate language programs and interpreter training. We provide certified, professional translation in 100+ languages including Spanish, Portuguese, Arabic, Chinese, Russian, German, and French. With over 20 years of experience, we have expertise in all major industries including the life sciences, patent and immigration law, international business, global education, and advanced technology. We offer cost-efficient interpreting and conference solutions that will meet your multilingual needs for all types of international events – business meetings, conferences, lectures or presentations.
For a free quote or to learn more about our services visit www.languageconnections.com.
Language Connections Inc.
2001 Beacon Street, Suite 105,
Boston, MA 02135