FLEx Parser for Translation

For further information email: Tim Stirtz, Andreas Joswig, or Beth Bryson.

Translation benefits greatly when analyzing a language with the FLEx parsers. When Paratext is integrated with FLEx, the quality of translation can dramatically increase through the rigorous application of these parsers. In addition, the number of organizations using machine translation is rapidly increasing, and parsing is foundational for rule-based machine translation.  FLExTrans, one such rule-based tool, was recently developed to reduce the time for drafting Scripture, especially in languages similar to another language that already has Scripture. A recent evaluation by SIL software developers notes that FLExTrans scored consistently higher in draft quality than statistical (AI) or neural (back translation) machine translation. As machine translation becomes the way of the future, there will likely be a high demand for those able to analyze languages with parsers for machine translation.

Link to Google page for article

Extracted text from FLEx-Parser-Andreas-Joswig.pdf:

Bangkok FLEx Parser Workshop August 2023

Andreas Joswig The first-ever FLEx Parser Workshop was held in Bangkok, Thailand with eighteen participants meeting for two weeks (07-18 August 2023). They made significant progress using one of the automatic parsers provided by the FLEx software suite. They were supported by six consultants, who are experienced FLEx users and had been successful in getting the parsers to work. Each day began with a classroom session about one of the various systems that make up a parser, but most of the time was spent on working on the databases of the participants.

Participants were very encouraged by how much progress they made modeling languages that they were analyzing. They were excited about how knowledge gained during the workshop will facilitate adaptation to related languages. Most of the participants went away with a newly built-up and mostly working database, to which they hopefully can add new forms without encountering too much trouble. Others went home with a dramatically improved database that previously had grown into something that they could no longer control. One participant in particular, who had been struggling with a huge database in which some words took 20 minutes to parse before the workshop, was very happy to see the same words parse in seconds.

A surprising number of participants decided during the workshop to switch over to the much more powerful, but also a bit more complex Hermit Crab parser, as it was found to be a better fit to the challenges provided by the morphophonology of the languages they are working on. Not only the participants, but all of the consultants gained a much better understanding of how the parser(s) operate, and how asking the right linguistic questions can get you out of a lot of trouble. It was good to have this nice group of very different consultants and participants, who all helped out each other with their experience and knowledge and different strengths. Indeed, the participants rather benefited from the fact that even the consultants found themselves on the learning end quite often. It tells them that their previous experience with parsing in FLEx is not a reflection of their poor abilities, but rather of the fact that FLEx is an incredibly complex software suit that brings together a great variety of information into a single clockwork, but which is still a little bit volatile - almost everybody will find themselves in need of help and advice once in a while. It is not only okay to ask for this help, but that is the best way to make progress. Beth Bryson, Matthew Lee, Ron Lockwood, Sara Mason and Kevin Warfel provided an extremely competent team of consultants at the workshop. Furthermore, several of the current participants showed potential to quickly develop into consultants themselves. Hopefully this workshop was not only an opportunity for making significant progress with these individual projects, but also an opportunity for strengthening the FLEx Parser community by developing new ways of helping each other and by identifying future consultants. The new FLEx Parsing List will also hopefully grow into a helpful forum for those using the parsers of FLEx. Everyone at this workshop expressed the wish for additional future workshops which could take place in various regions of the world. Regional workshops are especially helpful, as we saw that particular traction was gained in this workshop when the participants working in related languages had group sessions in which they compared notes on how they did things, and how different strategies work.

f