This is a brief and cursory comparison of The Bible Translator's Assistant (TBTA) to BibleTrans. TBTA is essentially the work of Steve Beale and Tod Allman.
I met Steve Beale shortly after I started developing BibleTrans. He was working on a PhD in machine translation (MT) at the time, and provided me with numerous valuable insights concerning the nature of MT ontology. I adopted his recommendation at that time that I use Louw&Nida (L&N) as a base ontology in BibleTrans. I still consider that a good choice, but Beale has modified his perspective somewhat and developed instead his own ontology. You can still see much of the influence of Beale's thinking in the "ABP Extensions to Louw&Nida" used in the current implementation of BibleTrans.
Tod Allman has been working on his MT project longer than Beale or I. Although I looked carefully at his work around the same time I met Beale, it appeared to me that his approach consisted primarily of modified word substitution coupled with word and phrase re-ordering, which the other linguists I spoke to at the same time generally held in contempt. I gave it no further consideration for that reason.
In 1999, after I made a public presentation of BibleTrans to SIL translators, Allman arranged for a meeting of himself, Beale and me at SIL in Dallas to seek common ground. We hammered out a set of agreed extensions to Louw&Nida, consisting largely of Beale's ideas with minor inputs from me and even fewer from Allman. I subsequently incorporated those concepts into BibleTrans, where they are still in use essentially unchanged. While I was working on that revision, Allman joined Beale in parting company with me and BibleTrans; Beale also seems to have developed reservations about using L&N as an ontology.
The current Allman & Beale work is called The Bible Translator's Assistant (TBTA) and a version is currently available for evaluation by download from their website. A demo version of BibleTrans is also available for download from my website.
The following comparison is based on public information on our respective websites. Please let me know if I have misrepresented their work in any way.
Beale continues to be interested in getting a good ontology, as he reports in his published papers (available for download from the TBTA website). While he understandably makes no mention of Louw&Nida nor the ABP extensions we agreed to in 1999, there is substantial continuity in his attributes for describing events and nouns. These are good concepts, and both TBTA and BibleTrans use them. Eschewing L&N for the Wierzbicka molecules gives his work a broader scope and appeal than BibleTrans; BibleTrans is useful only for Bible translation (and so far, only the New Testament). BibleTrans started out with a non-specific ontology similar to TBTA; we could at any time with minimal conversion effort, adopt Beale's ontology. However, L&N is well-defined and stable, while Beale appears to be still developing his ontology.
The Wierzbicka semantics was explicitly considered for use in BibleTrans at one time in the past, but it appeared incomplete and unstable and I declined to give it further consideration for that reason. A person more familiar with Wierzbicka semantics than I am recently told me her "semantic analysis of a word is a combination of a significant number of semantic primitives. For any given meaning of word, it takes a lot of effort to choose all of these." My original prototype had the same problem, but the L&N semantics overcomes that difficulty by supplying explicit Biblical meaning in context -- including substantial New Testament scholarship -- for each concept in the ontology. Beale's version of Wierzbicka semantics might be a better choice for general MT, but L&N has a definite advantage in Bible translation.
The TBTA translation engine is recognizably the same as Allman's previous work, with a "restructuring" phase, followed by a generation phase. Using Beale's ontology cures most of the "word substitution" criticism. It is a workable system, with demonstrable (but better documented) success comparable to BibleTrans.
The TBTA semantic data is explicitly in English sentence order, which requires "restructuring" rules to re-arrange the word order to be more suitable to the receptor language. BibleTrans uses a language-neutral non-linear database; the words still need to be generated in a language-specific order, but there is no English structure to tear down, so fewer rules are needed to achieve the receptor language requirements. I consider this to be a significant benefit for BibleTrans. Allman and Beale said they try to make TBTA rules resemble the way field linguists already describe their language grammar; BibleTrans forces the user to think in generative terms, which may or may not make BibleTrans more difficult to learn, but results in fewer total rules and exceptions to deal with. Both approaches obviously work, but it is unclear at this time which is better.
The fundamental difference between the BibleTrans and TBTA translation engines is like the difference between a person speaking in his native language, generating words and sentences from abstract ideas in his mind, as compared to most people speaking a second language, where they first think the sentence in their own language, then perform word substitutions and reordering transformations to produce something that is understandable in the output language. The foreign accent can be overcome, but it is very difficult.
As of this writing, Allman appears to be finishing up a PhD in MT; Beale already has that credential. My formal education is in computer science, so they have a definite theoretical advantage on me, as well as numerically more bodies to work on development. Their association with Wycliffe also affords them better access to field translators willing to test their product than I have been able to secure.
Comparing TBTA to BibleTrans may turn out to be somewhat like comparing VHS to Sony BetaMax, or Windows to the classic Macintosh: both do the job acceptably, but more developers and better adoption politics could overwhelm the significant technical advantage of the other product. I believe BibleTrans is the better technology, but even if my work on BibleTrans does nothing more than encourage Allman and Beale to make TBTA work and be used, I will have succeeded at my original goal, which is to see computer technology used to facilitate Bible translation for the 2000+ languages that still do not have any part of the Bible in their language.
2009 February 12