Just wanted to ask if the resulting TLG.fst transducer should always be determinizable? I can see how we can ensure determinizability of LG.fst by adding disambiguation symbols, but it is not fully clear to me how TLG.fst can be determinizable, because we don't add any disambiguation symbols on the input size of the T transducer(please correct me if I am wrong).
As an example, please consider 2 cases of the transduction using TLG.fst when input labels are phonemes.
AY -> I
AY -> eye
So, one phonetic input maps to 2 different words and there are no disambiguations involved in the process.
So, how the determinizability is ensured after obtaining TLG.fst ? Or it is OK to go without it?
Hello community,
Just wanted to ask if the resulting TLG.fst transducer should always be determinizable? I can see how we can ensure determinizability of LG.fst by adding disambiguation symbols, but it is not fully clear to me how TLG.fst can be determinizable, because we don't add any disambiguation symbols on the input size of the T transducer(please correct me if I am wrong). As an example, please consider 2 cases of the transduction using TLG.fst when input labels are phonemes.
AY -> I AY -> eye
So, one phonetic input maps to 2 different words and there are no disambiguations involved in the process.
So, how the determinizability is ensured after obtaining TLG.fst ? Or it is OK to go without it?