Microsoft and Skype joining forces to provide a solution to the AI-driven translations dilemma.
This dissertation provides a comprehensive overview of the Skype Translator machine translation system. The main goal of this analysis is to give a full computational analysis of its capabilities, in order to compare it to other automatic translation tools available for customers and to do an in-depth analysis of its linguistic development. As an automatic tool trained with a large number of corpora extracted for proper and spontaneous conversations, our focus has been pointed to its speech recognition and translation approach, given the fact that most systems like Skype Translator have a quite interesting difficulty in handling linguistic issues such as phonetic, morphologic, syntactic and semantic disambiguation.
The first chapter gives an overview of Skype’s evolution, followed by the Microsoft acquisition process. The role of the Microsoft Research department has been crucial for the creation of this machine translation system. In fact, its groundbreaking discoveries enabled the creation of a very powerful speech recognition system, one of the first in the field trained with authentic spontaneous conversations corpora.
The second chapter shows the architecture of the machine translation system technologies used for Skype Translator, followed by an analysis of the current status of the Machine Translation field of research. Most of the documents have been received directly from the Skype Translator’s team, which provided all that they could in conformity with their confidentiality obligations.
In the third chapter all the conversation logs, recognition and translation analysis are provided in order to understand the actual Skype machine translation system’s pros and cons, with a specific focus on the linguistic research of the phonetic, morphologic, syntactic and semantic approaches. In the last section of this chapter is also possible to understand how Skype Translator interfaces with dialogues belonging to the specific field as economic, humanitarian, medical and juridical ones.
The last chapter gives the conclusion on Skype Translator speech recognition and translation approach and aims to give a proposal of possible future developments of the Machine Translation research and suggestions on how to use Skype Translation properly in order to avoid most of the recurrent recognition and translation errors.
Paolo La Montagna is a linguist, passionate about the new frontiers of Natural Language Processing. He works as a freelance linguist for different clients that are specialized in NLP. Moreover, he is the localization manager of heartvoiced.com and the founder of thinkborderless.co.