coding t realisations English: how to do it objectively? any tips?

Question

For my MA thesis, among other things, I am coding the realisations of /t/ in English, by L2 speakers, as: 1) "normal" t, or 2) flapped, d-like t, 3) missing t, and 4) glottalised t.

While I have tried to code it by ear, I would prefer a more secure method. I have been looking at the wave forms of the sounds in praat (http://www.fon.hum.uva.nl/praat/), but still find it difficult. There are times when a /t/ sounds like a [d] to me, but looks like a [t] in praat. I think the reason that it sounds like a [d] may be due to a reduced VOT (voice onset time). Also, there are times when the t looks like it is invisible or glottalised, but on closer examination, there does seem to be a tiny t-wave form. I also find it difficult to tell objectively whether a t is simply missing or whether it is glottalised.

Do any of you have any tips or resources for this?

Answer 1

Describing a /t/ as "glottalized" without including an alternative "glottal stop" may be unwise, because it invites confusion between between a /t/ pronounced like an ordinary [t] but with simmultaneous closure of the glottis (which I would write [tʼ]) and a glottal stop [ʔ], because both are possible. (Also, "glottalized t" in other languages can refer to an ejective t.)

It's hard to give good examples for /t/, so please consider the difference between three kinds of /p/ in my speech: (1) plain [p] in "spore", (2) glottalized [pʼ] in "stopped", (3) glottal stop [ʔ] in "pop bottle". For (3), there is no lip closure for the second "p": [pʰɑʔbɑdl̩]. There are correspondingly three varieties of /t/ also, but for "pot bottle" the difference between glottalized t and glottal stop is just a matter of timing the alveolar closure. (You may not have these exact pronunciatiions in your own speech.)

Answer 2

The problem with this is that what you are looking at can be described in terms of a gradient (from the full form to the missing t), but in reality there is no single phonetic measure along which the various categories you are looking at are placed. So I guess the best option is actually coding them impressionistically, as you are doing already.

What you could do is to have two or three other independent people transcribing the data and then check the inter-transcriber agreement (http://www.statisticshowto.com/inter-rater-reliability/). If the agreement is high, you can be more confident that the categorisation you made was possibly correct.