What is Knowledge-based Translation in Artificial Intelligence?

Artificial Intelligence is a term which we all are familiar with and have often come across in some form or the other. However, what we see today is the culmination of decades of research and development made upon the base idea of creating intelligent machines which was first brought to reality in the 1950s. At the time of its introduction, the idea of creating machines that could “think” was thought of as something impossible, but the development being made in different directions such as computer memory, computation speed, etc has opened up the tremendous potential of the field. It is also believed that till now we have only scraped the surface of true artificial intelligence.

Artificial Intelligence in text-translation

Text Translation is one of the most common applications of artificial intelligence and it is often that one does not even know that AI is being used in it. Each one of us at some point in our lives has gone to google to translate a particular phrase or sentence to another language and may have also seen that sometimes while translating a sentence as a whole, it returns word to word translations which often makes an error grammatically or in the meaning. This is mainly because a machine cannot understand the meaning of the sentence and the emotion it contains. For a machine to effectively and accurately auto-translate sentences, it would require some level of understanding of the language and should have the ability to ‘read’. For example, when we read something, we take the help of our common sense and our knowledge about various objects, people, events, etc. in the real world and connect it with them, gaining an understanding of what we have just read.

In the initial days of machine translation, Bar Hillel stated that “the problem of competent automatic translation of a text is equivalent to the full understand of the text”. At the time, true machine translation was considered non-feasible because of the huge amount of understanding required to accurately translate a text and also the time complexity of applying this information. However, over the years, with considerable practical and theoretical advances such as the speed of computation as well as improvement in memory management and other factors, scientists believe that some progress towards knowledge-based machine translation can be achieved. Another important development that has opened up new possibilities is the contribution to the ‘database of common sense. Numerous models depicting world knowledge have been made and attempts to imbue intelligence has been done which have acted as stepping stones for scientists who have been able to create considerably advanced analyzers and translators of natural language.

What are the problems faced?

One of the most important problems which arise in automatic language translation is the lack of consistency between different languages. It can be often seen that a particular word in one language may not have a representation in another language and also that sentences and phrases may often change in meaning depending on the context they are used in. Therefore, it becomes even more important for a machine to be able to understand the text to accurately perform the translation and detailed information knowledge is required consisting of various settings and situations. After conducting several experiments and researches, it was widely understood that to accurately perform auto-translation, a level of understanding of the text was required and that too in a deep sense. It was also noticed that translation cannot be done just by performing syntactic manipulations, even if the notion of context is removed.

To overcome this, a method known as script-based machine translation was used. A script is a method to express peoples’ everyday events in language-free meaning representations. This means that scripts contain information about day-to-day events of people such as going to the museum etc. and include sequential information called causal chains. Each causal chain contains information about a particular action that can take place which further includes the sequence variables. So basically, when a text is given as an input to a translator, it first tries to understand the text and converts it into a language-free representation of it using the help of scripts and then finally tries to convert it into the target language. The in-depth details about the actual process of translating a text will not be covered in this article and maybe covered in the future. However, we will focus more on the basic understanding of the process, its requirements, and its problems.

How much understanding is required?

A question that always arises when talking about knowledge-based machine translations is that how much understanding of a language is required to perform accurate translations. It is believed that a complete understanding would be quite inefficient as the task in itself is very complex and a detailed representation of every word and its usage would be required. Therefore, the general approach to fully automated high-quality translation (FAHQT) has been found to be inadequate and huge debates and arguments have been going on regarding the same. It is also considered that minimal understanding of the text is enough to perform an acceptable accurate translation between a pair of languages as it would become much more complex and efficient.

The goal to create an ideal auto-translation system is still a long way from where we are and would require continuous efforts from our scientists to one day achieve this goal. As Artificial Intelligence itself was considered impossible a few decades back, we might not know what the future holds for us and what all development will be made to make this goal a reality. But one thing which we can say right now is that Artificial Intelligence will play an integral role in shaping our futures.