Introduction
Inter-language Vector Space is the advanced, neural network-based technology which lays foundation for the AI strategy of XTM Cloud, along with Neural MT. This is a unique mathematical algorithmic approach to the advancement of language technology based on massive neural networks. In short, it indicates the approximate closeness between distinct source and target words within a segment.
The technology is being used to enhance translators’, reviewers’ and post-editors’ productivity during their work in XTM Workbench.
Inter-Language Vector Space supports such XTM Cloud-related operations as:
auto-insertion of inline tags,
auto-bilingual term extraction,
auto-subsegment matching,
auto-fuzzy correction.
How does this technology work?
Inter-Language Vector Space enables direct alignment at the word and phrase level for the translation of a particular segment in XTM Workbench. It is built on extensive resources by Google and Facebook, with XTM contributing key elements from the L10N, multilingual point of view.
This AI-based technology draws on massive Big Data resources, including the resources of all of the Internet and XTM’s massive bilingual dictionaries, to calculate the probability of a particular target language word being the correct translation of a source word for over 250 language pairs. The purpose of this technology is to aid linguists in simple tasks by offering algorithm-driven automation, to improve their productivity and user experience.
Manual transfer of inline elements to target segments or bilingual term extraction take an inordinate amount of time and effort, negatively impacting linguists’ productivity and creativity. Therefore, auto-inline elements and auto-bilingual term extraction, supported by AI-based Inter-Language Vector Space, help eliminate the mundane work from translation processes. This, in turn, brings quicker turnaround times, significant cost reduction and a marked increase in quality.
Finding equivalents of source language words and phrases if no dictionaries or existing translations are available is a laborious and time-consuming activity. Since Inter-language Vector Space is based on extensive neural network analysis of the whole of the Internet, encompassing 150+ languages, according to this new approach, language is represented as a set of relationships between words and vector space points from one word to another, resulting in simple answers to operations, such as king is to ‘man’ as ‘woman’ is to queen. As a result, Inter-Language Vector Space is able to automatically work out relationships between words and how they relate to one another.
It knows, for instance, that king relates to ‘man’ and therefore can also provide the equivalent queen for ‘woman’. The unique point of this technology is that this can be done across different languages.
Take a look at the following example:
王 (king in Japanese) is to ‘man’, as ‘woman’ is to 女王 (queen in Japanese).