I concurr with Marcin - GIZA has been repeatedly used for grapheme / phoneme alignment - I do not remember seing complaints about non-monotonicity. See eg.
Gerosa, M., Federico, M.: Coping with out-of-vocabulary words:open versus huge vocabulary ASR. In: ICASSP, pp. 4313-4316 (2009) Laurent, A., Deleglise, P., Meignier, S.: Grapheme to phoneme conversion using a SMT system. In Proc. of Interspeech (2009) Taraka Rama, Anil Kumar Singh, Sudheer Kolachina. Modeling Letter-to-Phoneme Conversion as a Phrase Based Statistical Machine Translation Problem with Minimum Error Rate Training. Proc NAACL-HTL 2009 F On 24/08/12 08:23, Marcin Junczys-Dowmunt wrote: > I have been using Giza (and Moses) quite successfully for letter<->phone > transcriptions and transcriptions between two different phonetic > alphabets just with the standard settings. If the data is monotone then > it rather improbable that Giza will produce crossing alignments. I'd > guess it's just worth a try. > > W dniu 24.08.2012 04:10, Chris Dyer pisze: >> It should be possible to adapt Giza's HMM implementation to produce >> monotone alignments. These are the changes that would be necessary >> (and which should be fairly easy, if you can figure out the code): >> >> 1) alignment distribution initialization. by default Giza initializes >> the HMM transition probabilities to be uniform (effectively making the >> first iteration of HMM training the same as one more iteration of >> Model 1). You would need to alter this to make "reverse" jumps have >> probability 0. >> >> 2) smoothing. by default, Giza does something to prevent probabilities >> from ending up zero (maybe add alpha?). This is fine for monotone >> jumps, but you want to make sure that "backward" jumps end up zero. >> >> I think adding this would be have tremendous value. >> >> -Chris >> >> On Thu, Aug 23, 2012 at 7:53 PM, Philipp Koehn <[email protected]> wrote: >>> Hi, >>> >>> the IBM Models of GIZA++ are too complicated to be used >>> for simple monotone alignment. I am not aware of any >>> switches that would allow this either. >>> >>> I suggest to look at finite state machine tools such as >>> OpenFST - http://www.openfst.org/ >>> >>> -phi >>> >>> On Wed, Aug 22, 2012 at 5:29 AM, Dario Ernst <[email protected]> wrote: >>>> Hello dear list, >>>> >>>> first off, i'm not quite sure this is the correct list to ask GIZA++ >>>> questions - if not, please just tell me ;). I'm sorry for the trouble in >>>> that case. >>>> >>>> Anyways, my question. I'm currently trying to use GIZA++ together with >>>> PISA (http://pisa.googlecode.com/) to create monotone (linear?) >>>> alignments of words and phoneme-strings. For PISA i believe i've already >>>> found a way (thanks to the nice help of the author!), but for GIZA i'm a >>>> bit at loss. Is there some external parameter that i can set, or would >>>> digging the source be necessary? If so (and i've already started to try >>>> to familiarize myself a bit with the GIZA internals), what would be a >>>> good starting point to look at? Unfortunately i'm not that good with >>>> SMT internals yet, so it'd be a bit hard for me ... so at this point any >>>> help, input and tips would be greatly appreciated! >>>> >>>> Best Regards from Germany (and, please excuse my bad english ;P), thanks >>>> for reading this ;) >>>> >>>> -- >>>> Man's mind, once stretched by a new idea, never regains its original >>>> dimensions. -- Oliver Wendell Holmes >>>> _______________________________________________ >>>> Moses-support mailing list >>>> [email protected] >>>> http://mailman.mit.edu/mailman/listinfo/moses-support >>> _______________________________________________ >>> Moses-support mailing list >>> [email protected] >>> http://mailman.mit.edu/mailman/listinfo/moses-support >> _______________________________________________ >> Moses-support mailing list >> [email protected] >> http://mailman.mit.edu/mailman/listinfo/moses-support > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
