Across the United States today English to Spanish and Spanish to English are the dominant language pairs in business translation. Experienced translators are kept busy working on Employee Manuals, Benefit Guides, Settlement Documents, product labels and Workplace Safety Manuals.
But lately, Artificial Intelligence (AI) assisted machine translation is being promoted as being the next best alternative in getting the job done.
Machine translation, which has been around since just after World War II, is getting better and better with each iteration, and in the end may prove to be the breakthrough that we all, in this industry, fear.
But, fear not.
According to Cornel University’s Dzmitry Bahdanau, Kyunghyun Cho and Yosha Bengio, “neural” machine translation is the most recently proposed approach to machine translation.
According to the authors, “Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation.”
They go on to state that “we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task” of English-to-Spanish translation. “Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.”
This is hardly an easy couple of statements to understand and unpack. In essence, it is an attempt to give the computer a type of mind allowing it to make decisions based on contextual intuition. Depending on how frequently it finds a specific word in all relevant writing between the words preceding it and the words following it. This is only possible to contemplate with a significant amount of processing power.
Of course what this is describing is a basic kind of AI.
Will it work and if so, what is the cost?
When will we be able to ask the machine a question and get an answer that does not reveal itself to be a machine? Can a computer have a mind?
This was the question posed by Allen Turing in 1950. Are we getting ever closer to that scenario?
According to http://www.human-memory.net/brain_neurons.html “The average human brain has about 100 billion neurons (or nerve cells) and many more neuroglia (or glial cells) which serve to support and protect the neurons. Each neuron connects to up to 10,000 other neurons, passing signals to each other via as many as 1,000 trillion synaptic connections. By some estimates, this is equivalent to a computer with a 1 trillion bit per second processor. Estimates of the human brain’s memory capacity vary wildly from 1 to 1,000 terabytes (for comparison, the 19 million volumes in the US Library of Congress represents about 10 terabytes of data).”
Meaning that one human brain could have many more neural connections than all the computer processors in the US put together. A human that uses a small percentage of their brain, but taking into account all the nuances of word choice, can still outperform an AI translation. Not only in accuracy of meaning, but also accuracy of intent, and at cheaper cost.
There does not appear to be any danger that anytime soon the English to Spanish flyer or employee manual will be translated by AI. Still, there is little doubt that AI usage aims to tackle multi- thousand page documents such as technical manuals in a more efficient and cost-effective manner than the teams of human translators needed for very large projects. However, human translation is often needed for final proofing.