Digital Grammars builds on the Grammatical Framework (GF), which is a software system for multilingual translation and other applications of grammars. GF works in the same way as compilers, which translate program code into machine code. The translation is based on an abstract semantic representation, which guarantees that the source and the target have the same meaning. This model for processing is perfect for the main mission of Digital Grammars - to create language technology that one can rely on.
The key to reliability is the development of technologies that can be understood and where problems can be traced back to their origins. The proliferation of statistical methods in language technology led to a situation where the outcome of a process is hard to predict and almost impossible to control. In exchange, statistical methods give us robust systems with wide coverage but at the cost that certain error rate must be tolerated. In Digital Grammars we bet on carefully designed and verified language resources which give us high precision. Since we focus on producer oriented translation, we sometimes sacrifice coverage for precision. We still welcome statistical methods for rapid bootstrapping of new resources or in situations where we have to deal with incomplete information.
GF has been applied to over 30 languages and has large resources for processing them. The GF community has over 100 active members around the world. In addition to this, Digital Grammars has expert knowledge about the rich language resources of Språkbanken, the Swedish Language Bank, which is a leading producer of resources for Swedish. The cooperation between GF and Språkbanken has long traditions within the Centre for Language Technology at the University of Gothenburg.
Explainable Machine Translation (XMT)
Digital Grammars uses a translation technique that one can rely on. The translation is not given by a black box, but by a structured pipeline in which every step can be verified. This technique, Explainable Machine Translation, is an example of Explainable Artificial Intelligence (XAI). The importance of XAI is increasing as more and more critical tasks are performed by machines.
Our translation solutions are tailored for customers, to follow your terminology and style guidelines. You can safely use them to publish your content, with no or just a little manual corrections (post-editing). Our translation tools assess the confidence of translations, so that post-editors can focus on those parts that might need attention.
Digital Grammars translation solutions are highly multilingual, with a potential for over 30 languages. The more languages you need, the more cost-efficient it is to use our solutions. Our translation is based on meaning: we analyse the content of the source text and render the same meaning in the target language. Learn more.
Data-Driven Documentation (DDD)
The core of meaning-based translations is the technique of Data-Driven Documentation. In many cases, this core technique can eliminate text-to-text translation entirely: the translations are generated automatically, directly from data. Data-Driven Documentation is faster, cheaper, and more reliable than traditional text-based translation, since it eliminates the uncertainty involved in interpreting a text.
Data-Driven Documentation is applicable if your content already exists as machine-readable data: for instance, databases, product catalogues, financial reports and sports results.
The technology also supports text robots, which produce text in real time from streamed data - of course in many parallel languages if you want.
Data-Driven Documentation can be smoothly combined with translation, if your content is a mixture of free text and data. Learn more.
Natural language reasoning and question answering
The meaning analysis used in Data-Driven Documentation also makes it possible to process queries. The queries can be very accurate and complex, similar to SQL queries in a database. This is much more powerful than usual string-based search.
Digital Grammars search and question answering is of course multilingual: your customers can make the queries in their own languages, and they will get answers in the same language. But the answers can be found in a set of sources written in other languages.
- Explainable Machine Translation by Aarne Ranta. Abstract of an invited talk given at LaML-2017, Gothenburg. [abstract]
- Data-Driven Documentation: A Technique for Reliable Multilingual Information Access by Aarne Ranta. Abstract of keynote talk given at PIC-2015, Nanjing. [pdf]
- Grammar Engineering for a Customer: a Case Study with Five Languages by A Ranta, C Unger, DV Hussey - ACL/GEAF 2015, 2015. [pdf]
- Embedded Controlled Languages by A Ranta - Controlled Natural Language, LNCS/LNAI 2014. [pdf]
- Speech-enabled hybrid multilingual translation for mobile devices by K Angelov, B Bringert, A Ranta - EACL 2014, 2014. [pdf]