LinguList
commented
6 years ago I will drop this for the time being. It should be easy to reduce the variation by, i.e., searching for certain patterns in the data and changing the features. But this needs to be strictly planned, and not just done ad-hoc. For example, such a system would also have to answer whether voiceless + breathy exists (which it does according to phoible and other datasets). So this is a too big task for now and will be ignored for the time being.
What we mean by the "broad" IPA so far is also that we define feature combinations based on what we're given in the literature, and we already had some discussions whether this is useful, for example, allowing for a "breathy voiceless bilabial stop consonant" or an "aspirated voiced bilabial stop consonant", although it is clear that in both cases, we are most likely dealing with a "breathy voiced bilabial stop consonant". This problem can be addressed by adding another TS to our sample in which we are much more strict regarding the features. We'd thus no longer tolerate these distinctions but define them as alias of the same sound. The BIPA would still serve as our reference system, I suppose, as it is the system that is closest to the big transcription data bases, but the SIPA/STIPA/STRIPA would serve as a system that reduces the feature space by labelling many sounds as alias which would not be labelled as alias in the BIPA system. Needless to say that translation from BIPA to SIPA won't be always possible, while translation from SIPA to BIPA will be, but this is the general feature of transcription systems, that they may cover a different feature space.