UHEL will conduct experiments on creating translation memory data with GF grammars. Grammars of a given domain are used to generate bilingual aligned data, which can be converted into a translation memory. Then, when translating new material, the translation memory can provide fuzzy matches for cases where the constructions are similar, but words are different. This is one way to compensate for the lack of lexicon in a situation where adding new vocabulary is hard.
Jonson (2006) describes an experiment on synthesizing a corpus with GF for training speech recognition models. The idea is similar: use a grammar to generate reliable data for a data-driven approach.
By using GF translation suggestions in a translation memory it is also possible to use standard translation tools like Trados to generate pre-translation reports of exact and fuzzy matches. These percentages of different matches are easier to demonstrate to translation industry stakeholders, as the scientific metrics used in MT evaluation (BLEU, NIST, Rouge and so on) are not generally used or well understood within the industry.
Services in the GF cloud will be linked to each other by UGOT. Syntax editor (http://cloud.grammaticalframework.org/syntax-editor/editor.html) will be used within the Simple Translation Tool (http://cloud.grammaticalframework.org/translator/). As described in the DoW, there will be a mode for editing source text, where structural changes to the document can be made by manipulating abstract syntax trees. This functionality will be added to the Simple Translation Tool.
Simple Translation Tool will be extended to a bilingual, controlled language document authoring tool, with useful ways to enter and edit the source text too. Additions include a text input guided by word completion and syntax tree editing, by invoking the syntax editor (see section 2.2) on a source segment.
plan to submit a system demo paper to the MT Summit, http://www.mtsummit2013.info/ deadline 22 April. About a unique MT platform,
Grammar editing in the translator's tools is still an open question. One of the main shortcomings of a MOLTO type translation is the limited coverage, and that's why it is important that a translator can easily extend and modify the lexicon. Just importing raw lexicon data, with or without TermFactory, is described in Listenmaa (2012). This is a question of adding more content, but another question is modifying existing grammars, usually in a case of an error.
There are some steps taken to further this issue. D11.2 presents a multilingual semantic wiki, where it is possible for every user to modify the grammar behind the wiki. This is still an expert work, as the editing is done with raw GF code, but there are methods for guided grammar editing, such as the cloud-based IDE (see documentation). This environment offers an easy way of multilingual grammar writing and editing.