Devoicing in English

[HoskhMatriarch]

I'm not saying English has any German-style final obstruent devoicing, but I was just thinking about how much weird devoicing there is in English. For example, I say second as [sɛkn̩t], and when I was researching on the Internet to see if anyone else did that, I found that some people do too, and also that people have complained about it, as in a page from a book "this guy wrote second as secont and pronounced it the same way, we must destroy this evil of uneducation" (which, while I think it's nice when people know how to spell, don't get what all the drama about it was for, nor understand why he thinks it's apparently some great sign of uneducation to pronounce second with a voiceless stop at the end instead of voiced. I'm not going to change my pronunciation for any of the people I've ran across who have complained about it anonymously on the Internet; in fact, now I only want to pronounce second with a voiceless stop on the end more). There is also the whole have to/supposed to thing, where certain words get devoiced before "to", but not all of them, because if I said "neet to" I'm pretty sure I'd just get random people going like "Are you German?". People even say it voiceless in emphatic speech with great pauses, like "it's not SUPPOST to be like that" or "I HAFF to!" so even if it was originally assimilation of some sort, it's basically lexicalized now. I also could almost swear that I've heard a small number of people say voiced stops voiceless at the end of words before pauses and even sometimes before other voiceless stops. Maybe English is in the middle of acquiring German-style final obstruent devoicing? Whether it is or not, it's still pretty interesting.

[vokzhen]

English voicing isn't straightforward in the first place, which doesn't help. The fortis /p t tʃ k/ series is aspirated, unaspirated, or glottalized (with at least four different kinds of glottalization: preglottalized, unreleased, glottal replacement, and ejective) in some combination depending on context, and the lenis /b d dʒ g/ series is partially voiced, fully voiced, or devoiced, as well as affecting vowel length (and probably tone, though I'm not 100% on that), also depending on context.

"Second" is probably just an irregular change (which, for the record, I have too). "Have," though, I'd say is going through a lexical split, the predicate possessive and perfect marker [=v, hæv] on the one hand and the modal verb [hæftə] on the other, and like other modals, "hafta" doesn't require an infinitive because it's been grammaticalized as part of the verb itself. "Supposed" is probably similar, because there's a split between "he was supposed to leave" and "he supposed he should leave."

You really need to use paragraphs. That could easily be split in two, probably three, and maybe more.

[zompist]

because if I said "neet to" I'm pretty sure I'd just get random people going like "Are you German?".

Are you sure you don't? For me "I need to go" is [aj ni: tʰə gow]. I could get a [d] in there by emphasizing and drawing out 'need', but it still might just be [ni::t] before 'to'.

[Salmoneus]

Well, I don't, I don't think. I think I'd be more likely to voice the following /t/ than to devoice the /d/. Although I don't think I do that either.

[Zaarin]

I devoice "supposed" before to, but interestingly I don't devoice "have." In the case of "need," I either voice or elide the /t/ rather than devoicing the /d/.

[vokzhen]

I was testing it out as I was posting, and in things like "need to" and "had to" I can't tell for sure if it's entirely devoiced and it's the length of the previous vowel I'm picking up, or whether it's genuinely voiced at least at first. At a guess, in connected speech "need to" is something like [nidtə]: longer than a singleton consonant, there's a slight buildup of pressure from voicing, and a voiceless but completely unaspirated release. Rather similar to phrase-final /d/ except that it's longer.

[Travis B.]

At least here, there isn't a general final devoicing but rather the following:

Fortis plosives are aspirated voiceless initially (except in the notable case of unstressed to) and before stressed syllables except when after sibilants within the same syllable, preglottalized in codas, and otherwise tenuis, except /t/ which is flapped in unstressed intervocalic positions or when after /n/ in an unstressed prevocalic position (where then it coalesces with the /n/ as [ɾ̃]), ignoring the elision of /t/.
Fortis fricatives are voiceless in all positions.
Lenis plosives are tenuis initially except in careful speech (where they may voice, and in the case of /b/, may even be prenasalized or implosive), are in free variation between tenuis and voiced between sonorants including across word boundaries (with /d/ flapping or eliding unstressed intervocalically, and, at least in my idiolect, except for /dʒ/, which I practically never voice), and in my own idiolect are otherwise tenuis except in careful speech (where they may voice), while some people may have more frequent voicing of lenis plosives in general.
Lenis fricatives are generally voiced, but in my own idiolect devoice in codas when not followed by sonorants, but many do voice coda fricatives generally.

There is generalized vowel length allophony, where all vowels followed by a lenis obstruent with no intervening fortis obstruent are long, where all vowels followed by a fortis obstruent with no intervening lenis obstruent are short, and where all other vowels are long, with this operating across word boundaries (such that the lengths of final vowels depend on what the word following it starts with). Note that this is more significant for communicating fortisness versus lenisness than phonation in many cases, especially since in many positions both fortis and lenis plosives may be tenuis.

There is also generalized voicing assimilation favoring assimilation towards voicelessness.

In my speech at least it seems as if I have final devoicing, but a key difference between what I have and what is present in, say, German, Dutch, Polish, Russian, etc. is that my final devoicing is suppressed when the following word begins with a sonorant.

[Nortaneous]

'need to' and 'had to' have [-dtə] for me, but 'used to' and 'supposed to' always have [-stə] and 'have to' always has [-ftə] and those are lexicalized. 'second' can have either [-nd] or [-nt]~[-nʔ].

clusters of two consonants differing only in voicing aren't uncommon IMI: -t ð- sequences ('get the...') can optionally assimilate to (interdental) -t̪d̪-. (cf. -n ð- > -n̪n̪-)

[Imralu]

I notice these things in various US accents:

Second pronounced as secont.
Didn't pronounced as ditn't.
Ninety pronounced as ninedy.

[Travis B.]

I have [ˈniːt(ʰ)ːə(ː)] for need to and [ˈ(h)ɛːt(ʰ)ːə(ː)] for had to. This reflects varying levels of stress put on the to; if the to is given more stress it is aspirated, otherwise it is not.

I have (mostly*) free variation between [ˈsɜkɘ̃ʔ] or [ˈsɜkɘ̃n] or [ˈsɜkn̩ʔ] or [ˈsɜkn̩], i.e. with /t/, and [ˈsɜkɘ̃ːt] or [ˈsɜkɘ̃ːn] or [ˈsɜkn̩ːt] or [ˈsɜkn̩ː], i.e. with /d/, for second. This reflects varying fates for final /nt/ and /nd/, i.e. whether the /n/ is elided and reflected purely as vowel nasalization or whether the clusters are reduced to [n].

I have (mostly*) free variation between [tɘːnː(ʔ)], [tɘ̃ːːʔ], and [tɘ̃ːːn] for didn't. This reflects varying fates for intervocalic /d/ and /nt/, i.e. whether they are coalesced or whether the /d/ is simply elided, whether the /t/ is elided, and if not coalesced, the preservation of /n/ if /t/ is elided and the elision of /n/ if /t/ is preserved.

I have free variation between [ˈnãẽɾi(ː)] and [ˈnãːĩ] for ninety. This reflects variation between the preservation and elision of intervocalic /nt/ and an irregular fate for non-elided intervocalic /nt/, where the /n/ is elided and reflected as vowel nasalization and the /t/ is flapped; note that the underlying form has /t/ and not /d/, since the vowel beforehand is short and not long without /t/ elision and the resulting vowel is long and not overlong with /t/ elision.

* Mostly in that all forms can be found in all positions, but forms ending in [n] are somewhat favored before vowels.

[Nortaneous]

I notice these things in various US accents:

Second pronounced as secont.
Didn't pronounced as ditn't.
Ninety pronounced as ninedy.

θɚttin foɚ̯ttin ... ei̯ttin nai̯nttin

[Zaarin]

I recently realized that I actually have a split when it comes to "supposed." When "supposed to" is used as a synonym of "ought," I devoice it, but not when it is used as synonym of "alleged."

[Salmoneus]

I recently realized that I actually have a split when it comes to "supposed." When "supposed to" is used as a synonym of "ought," I devoice it, but not when it is used as synonym of "alleged."

I think that's the norm. It parallels the similar split between devoiced 'used to' (was once accustomed to) and voiced 'used to' (employed for the purpose of).

[Travis B.]

I recently realized that I actually have a split when it comes to "supposed." When "supposed to" is used as a synonym of "ought," I devoice it, but not when it is used as synonym of "alleged."

I think that's the norm. It parallels the similar split between devoiced 'used to' (was once accustomed to) and voiced 'used to' (employed for the purpose of).

To me it is a vowel length difference and a gemination shift/reduction difference, i.e. be supp[osʲtʲ~osʲːtʲ(ʰ)]-o* versus be supp[osʲtʲː(ʰ)]-o* and [jʉsʲtʲ~jʉsʲːtʲ(ʰ)]-o* versus [jʉːsʲtʲː(ʰ)]-o*. Both those examples reflect /s/ versus /z/ in the underlying form, but that cannot be the only thing going on as that does not explain the gemination shift/reduction difference. However, I do not make this distinction in very careful speech, where I use the latter pronunciations of each for both meanings.

Note that such devoicing of underlying forms is also found in h[ɛf] to versus h[ɛːf] two. Similarly this reflects underlying /f/ versus underlying /v/. Likewise this distinction is not made in very careful speech, where I use the latter pronunciation in both cases.

* -o indicates just the final vowel of to, which is [ə~ʉu~ʉ].

[Zaarin]

I recently realized that I actually have a split when it comes to "supposed." When "supposed to" is used as a synonym of "ought," I devoice it, but not when it is used as synonym of "alleged."

I think that's the norm. It parallels the similar split between devoiced 'used to' (was once accustomed to) and voiced 'used to' (employed for the purpose of).

You're right; "used to" is the same for me.

[gmalivuk]

If they were sentient, we could say typewriters used to be used to being used to write letters. The last "used" is the only one I voice all the way through. (I wasn't entirely sure about the correct spelling of "didn't use to" until I started teaching English, despite the extensive reading and writing I had done in college, because it sounds completely identical to "didn't used to".)

In my teacher voice, I've definitely caught myself a few times carefully saying things like "secont", probably in part as a hypercorrection of the flapped /t/ sound I know students sometimes find difficult to understand, and in part by analogy to the much more common "-ent"/"-ant" ending.

Devoicing of final /d/ is also something I've noticed in AAVE, especially after seeing it rendered into text with words like "gotdamn" and "Lordt" and "surfboardt". (An association with that dialect might be part of why the OP found people complaining about how "uneducated" it seemed.)