Phonological theory that denies phonemes?

[Muikkunen]

Some time ago I found a website proposing phonology without phonemes, because illiterate people don't have awareness of them, arguing that phonemes are a byproduct of writing system development. On that website, there were links to papers in psycholinguistics about phonemes having no mental representation in illiterate people and also article about history of writing systems.
Unfortunately, I forgot what that website and theory are called, but really need to find it now.
Can you help me?

[Daniel]

Hm, that specific website doesn't sound familiar to me, but the topic is interesting, so I'd want to take a look if you find it.

The general idea that phonemes are not a real part of our mental grammars is not a new question/suggestion, and phonemes are actually somewhat controversial, but for more subtle reasons than the summary of the website you described suggests. That is, there are problems with uniquely identifying phonemes (e.g. https://en.wikipedia.org/?title=Archiphoneme&redirect=no), although in general I think there is scientific consensus that we generally have categorical perception in some sense for phonology.

And this paper came to mind, which might be relevant for you (if you don't already have it):
Maddieson, Ian. 2018. Is phonological typology possible without (universal) categories? In Larry M. Hyman & Frans Plank (eds.), Phonological Typology, 107–125. Berlin: De Gruyter. https://doi.org/10.1515/9783110451931-004

[Forbes]

but the topic is interesting

It is!

It is undeniable that speech is continuous, but we feel it is segmented. The question is though whether we feel that way because we write alphabetically. Is the concept of the phoneme motivated by the fact that we write alphabetically, or do we go back and insist that the alphabet assumes a phonemic analysis? A chicken and egg question.

In another forum there was a discussion about whether affricatives are one or two phonemes. I think it has to be a bit flexible and to an extent language dependent, It comes down to whether you think you can make a phonemic analysis of language without reference to morphology and semantics. In the end any division is speech is arbitrary and the question should be what units are helpful in any context.

Emphasising that speech is continuous, spectrographic analysis has shown that phonemes "react" with each other. For example, in /kæt/ the k is a "k before æ"; the æ is a "æ after k and a æ before t"; the t is a "t after æ". So phonemes are an abstraction.

[Daniel]

It would be relevant to investigate this with speakers of languages that have different kinds of writing systems. In my Morphology class, I've discussed how although orthography does not directly indicate structure, there do seem to be some ways in which certain orthographies fit the organization of some languages better than others. The consonant-only writing of Ancient Egyptian and Semitic languages almost certainly has something to do with their templatic morphology and focus on consonantal roots, while the Greeks then accidentally created vowels as a misinterpretation of the Phoenician alphabet (really abjad). Chinese is another example of a different kind. Or consider Japanese where the written language is organized into content words (kanji = Chinese characters), grammatical particles/suffixes (hiragana syllabary) and borrowed content words or onomatopoeia (katakana syllabary). (I wonder if Japanese speakers would have different phonemic perception of content words vs. grammatical morphemes, although they can also write everything in hiragana, and children are taught that way, so it may no longer be so cognitively salient. Similarly for Chinese they know how to use Pinyin or similar systems of transliteration, so phonemes may also be familiar. But also consider the organization of Japanese into moras instead of syllables, which is somewhat reflected in the orthography.)

So phonemes are an abstraction.

Yes, an abstraction, but one that seems cognitively real at least in some ways. We have categorical perception of phonemes. And also more generally, take a moment to realize how remarkable it is that we can in fact write spoken language using discrete symbols. So far, signed languages have evaded any kind of conventional writing system because they are not (entirely) discrete in the same way, so it is not clear what units to write down. There have been many proposals, but remarkably for a visual medium, it has not been successfully represented in (visual!) writing, at least not in terms of any writing system being widely adopted. And even in principle, it may not be possible to write signed languages completely in discrete symbols, because some signs simply aren't discrete where the signing space is used as a continuous 3D space for representing spatial relationships (although we could approximate that in writing). There's a similar problem with intonation for oral languages, actually, and linguists sometimes use impressionistic systems of prosody but it's not clear whether this fully captures the information in speech: for example, https://en.wikipedia.org/wiki/ToBI

[panini]

There are various things that might be denied. One is that humans have a symbolic system for categorizing speech. I take that to not be reasonably deniable, however I believe that certain phoneticians adhere(d) to a non-categorial analysis of speech recognition.
Another that can be denied is the specific package-deal "phoneme" as contrasted with allophone with all of that biuniqueness stuff. I do deny that there is such a thing as a "phoneme", where /t/ is a phoneme of English and [tʰ t̚ ɾ] are allophones. It is true that [tʰ] can be derived by rule from /t/, but that doesn't justify creating a special theoretical category. There is a difficult question as to which structures exist in the phonological component and which only exist in the phonetic component, and for example there is good evidence that "nasal vowels" are not a phonological category of English, they are the result of phonetic implementation: but the flap is clearly a phonological entity.

A further question is what kind of thing a "phoneme" is. Is it part of phonological grammar, phonetic grammar, or something on the periphery of language, at the auditory interface with phonetics? The "speaker awareness" test is, indeed, not reliable exactly because speakers need specific training in their writing system, and they do not learn a general skill about "phonemes" from one language that can be translated to other languages. E.g. a speaker of Logoori may know how to write English and Swahili, and even Logoori but the writing system is defective in not representing all of the phonemes. Unless one engages a speaker in training in linguistics, the standard reaction to a vowel-quality, vowel-length or tone minimal pair is that "they are spelled the same". You can experimentally get them to identify words, but they don't have any awareness of what's a "phoneme" in their language.