If you ever wondered what’s the difference between a morph and a morpheme, this blogpost contains an easy answer: Your stereotypical “morpheme” is actually a morph! No need to worry about all the problems with morphemes anymore: We can simply say “morph”, and continue to live happily. A morph is a minimal form, and if anyone has further questions about this definition, I’ve answered them in a forthcoming paper (“The morph as a minimal linguistic form”)
We could of course also continue to use the better-known Bloomfieldian term “morpheme” (in the sense of morph), but every linguist who knows a little more than what we find in Youtube videos (e.g. here or here) knows that there are huge problems with this notion. Anyone who has taken an advanced course or is well-read in the literature knows that “morphemes” don’t work the way Bloomfield (1933) may have imagined it. Languages simply don’t only consist of strings of meaningful pieces (= morphs) strung together – things are a lot more complicated, a lot of the time. Anderson (2015) concludes his comprehensive survey paper in a sobering tone (which basically says “down with the morpheme”, but more politely):
“We commonly assume that both the intension and the extension of the term [morpheme] are virtually self-evident, but it turns out on closer examination to hold the keys to some of the deep questions we can ask about the nature of language. One of these, indeed, is whether or not there is any such thing as a ‘morpheme’.”
These deep questions are unanswered (and will not be answered definitively anytime soon), so should we base our introductory lectures in linguistics on one or another very questionable theory of the nature of language?
Probably not, so we want to have the clearly defined term morph in our textbooks (and Youtube videos), because anyone can understand what a minimal form is (= a string of segments that has a meaning of some sort and cannot be broken down into smaller forms).
So how is this different from the Bloomfieldian definition of a “morpheme” as “the smallest meaningful unit”? One simple difference is that there may be unpronounced meaningful units, often called “zero signs” (at least since Jakobson 1939). For example, many people say that there is a zero plural article in English, based on the paradigm
“the tree – the tree-s – a tree – Ø trees”
(e.g. Koeneman & Zeijlstra 2017: 21) This zero article would be a “zero morpheme”, but it not a “zero minimal form”, because forms cannot be zero.
The crucial difference here is between a “unit” and a “form”. What exactly is a “unit”? This is hard to say, but we know what a form is: a pairing of a segmental string and a meaning of some sort. A form is something that we can hear (or see, in the case of a sign language), whereas a “unit” can be an abstract element of a language system. Linguists have often made use of zeroes as elements of the system, for very good reasons. Without abstract notions such as zeroes, derivations, and movements, it’s rather difficult to describe a language concisely. But we cannot hear a zero, or a derivation, or a movement. These may well be “units” of a system (depending on our analytical preferences), but they aren’t forms that we can hear. As a result, there is very little agreement among linguists about the kinds of “units” that we should use to describe/analyze languages – and our students are confused.
But if we define a morph as something that a composite expression may be broken down into, then we can continue to use this term uniformly, and to assign tasks to students that are actually manageable – because all linguists agree on how to do it in principle (and when they don’t agree on particular cases, they disagree for well-understood reasons, along clear dimensions such as “lumper vs. splitter”). This is of course the same as the stereotypical “morpheme”, so for the purposes of quick explanation, it may be sufficient (and it’s probably everything that an average student remembers anyway).
But when we consider less stereotypical cases, things get difficult. To see to what extent linguists agree on “morpheme analysis” of English, I did a little experiment a while ago: On Twitter, I asked my followers (most of them linguists) how they would count morphemes in an English vowel-changing Past-Tense form such as sang (Paste Tense of sing). The result was about evenly split, with about two fifths saying that such a form contains two morphemes, and two fifths saying that it contains one morpheme. The remaining fifth says that the question makes no sense (perhaps because these respondents agree with Anderson and many other morphological theorists that the “morpheme” concept is on the wrong track).
In the English past-tense form „sang“, is there one morpheme (expressing the event and the tense), or are there two morphemes (the root morpheme and the tense morpheme)?
— Martin Haspelmath (@haspelmath) April 5, 2020
What explains this discrepancy? Presumably, it’s not that different linguists on Twitter have radically different views about the nature of language – they probably understand the word “morpheme” in two different ways: The problem is terminological!
Fortunately, terminological problems (unlike deep questions about the nature of things) are eminently solvable. We are dealing with three different concepts here:
(1) minimal linguistic form (= morph): if “morpheme” is used in this way, then sang consists of a single morph(eme), and the first answer is correct.
(2) minimal abstract “unit of analysis”: if “morpheme” is used in this way, then sang consists of two morphemes, because clearly, this form is used to denote singing and is at the same time a Past Tense form
(3) minimal meaningful all-purpose piece for analysis: if “morpheme” is understood like this, then the question makes no sense, because there can be no “all-purpose pieces” for analysis: Linguistic analysis is complicated and necessarily includes a range of different kinds of “elements” or “units” (minimally forms, classes and constructions, but quite possibly also zero, derivations, movement, interfaces, depending on one’s predilections).
Bloomfield doesn’t seem to have been aware of the differences between these three minimal entities, but by the time of Hockett (1947), some of the issues had come to light, and more came to light later (Matthews 1974; Anderson 1992; and so on). Hockett proposed the term “morph” for a minimal linguistic form, but unfortunately, Bloomfield’s older term “morpheme” continued to be used by most people, often in confusing ways (see Mugdan 1986 for a scathing review of the confusions of the first hundred years of the term “morpheme”; and Mugdan 2015 for a more comprehensive update).
In the morphology textbooks, the term “morph” is rarely mentioned, but it is often observed that a morpheme may be “realized” in different ways, and the term allomorph is prominent in all textbooks (and in the literature more generally) for such different realizations. Hockett, for example, said that the English plural morph -s in cow-s and the plural morph -en in ox-en “are grouped into a single morpheme” if the occur in complementary contexts (1947: 322), and such sets of suppletive forms are commonly discussed unde the heading of “allomorphy”. But on the other hand, phonological variants (e.g. cow-s vs. bush-es) are also often called “allomorphs” – so what is an “allomorph”?
Is each “allomorph” a morph, as the term would suggest? If so, are there also morphs that are not “allomorphs”? What is the difference between an allomorph and a morph. Most linguists would find it difficult to answer these questions, which shows that there is no common understanding of this term in the discipline either. But these terms occur in the textbooks as well, confusing our students in the same way as the term “morpheme”.
I have therefore proposed that sets of suppletive morph sets should be called supplemorphs, and phonological variants should simply be called morph variants (see this earlier blogpost). Supplemorphs are different forms (the plural form -s is not the same form as -en), but morph variants are variants of the same form (-s and -es are variants of the same morph).
If you read this as a non-expert, then you may be shaking your head: Is it really this bad? Do linguists use their most basic (and most famous) terms in such confusing ways? Yes, I’m ashamed to say (and I didn’t fully realize it when I wrote a morphology textbook quite some time ago, Haspelmath 2002) … If you read this as an expert and think that I’m wrong, then please make an alternative proposal, or at least point out errors in what I say here (and in the paper).
So to end on an upbeat note: There’s a good chance that linguists could converge in the future on the term morph (with its good Hockettian pedigree) to denote a minimal linguistic form (or of they want to continue to use the better-known term “morpheme”, then they will be aware that they actually mean “morph”). We may not agree on our visions for understanding the nature of language, but we can agree on some basic terminology for our textbooks.
Anderson, Stephen R. 1992. A-morphous morphology. Cambridge: Cambridge University Press.
Anderson, Stephen R. 2015. The morpheme: Its nature and use. In Matthew Baerman (ed.), The Oxford handbook of inflection, 11–33. Oxford: Oxford University Press.
Bloomfield, Leonard. 1933. Language. New York: H. Holt and Company.
Haspelmath, Martin. 2002. Understanding morphology. London: Arnold.
Haspelmath, Martin. 2021. The morph as a minimal linguistic form. Morphology (to appear) https://ling.auf.net/lingbuzz/004926.
Hockett, Charles F. 1947. Problems of morphemic analysis. Language 23(4). 321–343. https://doi.org10.2307/410295.
Jakobson, Roman. 1939. Signe zéro. In Mélanges linguistiques offerts à Charles Bally, 143–152. Genève: Georg & Cie.
Matthews, Peter H. 1974. Morphology. Cambridge: Cambridge University Press.
Mugdan, Joachim. 1986. Was ist eigentlich ein Morphem? Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 39(1–4). 29–43. https://doi.org10.1524/stuf.19184.108.40.206.
Mugdan, Joachim. 2015. Units of word-formation. In Peter O. Müller, Ingeborg Ohnheiser, Susan Olsen & Franz Rainer (eds.), Word-formation: An international handbook of the languages of Europe, vol. Vol. 1, 235–301. Berlin: De Gruyter Mouton. https://doi.org10.1515/9783110246254-017.
I know this is a pretty old post, but I thought it couldn’t hurt to share my thoughts anyways. 🙂
A lot of people seem concerned with the defining a minimal form as “a pairing of a segmental string and a meaning of some sort.”
You clarify after providing this definition that “A form is something that we can hear (or see, in the case of a sign language)”. But it is not actually possible to hear an abstract, toneless string of segments, because every time a string of segments is pronounced, it is pronounced with a particular tone.
Perhaps the definition could be tweaked to “a pairing of a segmental string (including tone) and a meaning of some sort”?
Under this definition, we can clearly say that in tonal languages, tá and tà are distinct forms, and may even be distinct morphs. In non tonal languages, we would say that tá and tà, are still technically different forms, even though they do not contrast, and say that they are “morph variants” in exactly the same way that we currently say that the [z] and [s] of [dogz] and [kats] are morph variants.
[By the way, the concept of “morph variants” seems to imply that “morphs” are actually equivalence sets of actual forms. So it seems like much of the discussion here boils down to what sort of equivalence relation we should use to define a “morph”.]
To make this a good retro-definition which will work even for previous work that does not provide tone-marking, we could say that “if a source does not provide tone-marking, you should assume that tone is not contrastive in the language, unless the author of the source states otherwise”.
I hope this is helpful!
A note: Unfortunately, it seems to me that my proposed definition will still not be able to handle “floating” tone morphs/morphemes. However, I think it is an improvement over the original definition, because it is capable of handling non-tonal languages just as well as your original definition, while also being applicable to tonal languages of the Vietnamese/Chinese sort, where each morph is associated with a tonal syllable.
I agree with all of this. I would definitely say that “tá and tà are distinct forms”, but I wouldn’t say that the tones are forms, just as I wouldn’t say that segmental features are forms (e.g. the fronting in German Bücher, vs. Buch, is not a form). I thought that my definition entails this, but maybe it could be still clearer.
Hi Martin, nice that people are debating these issues here, hopefully you’ll get lots of feedback as well on your Morphology article!
I have some reactions. First, you write ” the only way to define “morph” is as a type of “form” “. Agreed. But I have a more liberal conception of “forms”. It’s true that canonical linguistic forms consist solely of segments. But segments are but one type of primitive which can combine to form new forms. Such as strings /æ k t/ ‘act’, /t æ k/ ‘tack’, /k æ t/ ‘cat’, etc.
Equally, why can forms not also consist of other kinds of phonological primitives, i.e. tonemes (on another, suprasegmental tier), accent marks, floating features without segments (e.g. floating [+Advanced Tongue Root] in some vowel harmony systems), or floating segments without features (e.g. coda “Q” archiphoneme in Japanese which assimilates to following consonant).
Second, what I see this boiling down to is interpreting non-segment phonological primitives as too abstract to stand on their own, but seeing segments as not abstract. This cannot be right. Segments are not speech sounds, so it is not the case that “a “form” is necessarily segmental [because] (its something that we can hear)”. We do not hear segments, we hear the phonetic correlates of segments as they are dictated by the phonological/phonetic grammar(s), in conjunction with linguistic and non-linguistic factors. The idea that we can hear just the segments /k æ t/ in ‘cat’ neglects the fact that vocalic segments are accompanied by suprasegmental intonational structure (even, perhaps especially, in isolation). This holds for both tone languages with “grammatical tone” (add H tone for the past perfective) as well as non-tone languages with intonation (add H*!H for an imperative meaning). A phonological representation /k æ t/ in English (or the four-co-exponent realization of [HABITUAL] TAM in Makonde [H + nda- + LHL + -a] with four floating tones – Manus 2014) is not something which can ever be heard.
Third, Sebastian writes “ablaut processes like in sing/sang should not be called morphs or morphemes”. I agree if they are non-productive (a separate issue of how one defines this). But if it is productive (like parts of Icelandic inflectional paradigms if my memory serves me), then I would say the exponent of such meaning is [+Front] (or some such equivalent). If segments can make up forms, and features make up segments, it doesn’t make sense to me to say that a single feature cannot be a form. Especially if the feature categorically co-varies with a particular semantic meaning (i.e. the duality of patterning).
In short, I believe that morphs-as-forms is the correct way forward. But what the forms consist of depends on the phonological system and how it is organized. If phonology is symbolic, then forms consist of whatever symbols we have evidence for (whether segments, features, tones, accents, hand configurations, etc.).
Many thanks, Nik, for continuing this discussion! You are making many good points, and this discussion shows that some of our basic concepts are not generally agreed. On the one hand, you agree that “a morph is a form” (which not everyone seems to find intuitive), but on the other, you say that “segments are not speech sounds”, which I do not find intuitive.
The reason why I don’t want to say that a form can consist of a floating feature or other abstract elements is that we don’t talk like this. I want to find definitions of terms that correspond to the way we actually do talk. And I think Bloomfield (1933) was right to say that a form is a sequence of segments – this is how linguists use the term “form” (or “expression”). (Or maybe Bloomfield was actually instrumental in creating this notion of form.) It does not HAVE TO be like that, of course, and you CAN use the term “form” in an extended (“more liberal”) sense – just like you can extend the sense of any term (often without people noticing). The meanings of our terms are arbitrary, just like the meanings of all ordinary forms. But what I’m arguing is that they should be FIXED, i.e. just because a floating feature is in some ways like a segmental form does not mean that we should call it a form. We’d have to find some other term for what morphs and floating features share.
An objection that I would accept would be if you showed me that the term “form” is actually used for floating features in the literature.
(I am fully aware that the term “morph” is often used for zero expression or other non-segmental ways of rendering meanings; but you already seem convinced that “zero morphs” are not such a good idea. But why not? I think the only reason is that we want a good definition of “morph”, and conversely, a good term for a minimal form.)
And there is also one other important point to keep in mind: I am distinguishing sharply between general concepts and language-particular concepts, and many other linguists (especially generative linguists) do not want to make this distinction – they want to use the same concepts for description and for general linguistics. I think that Larry Hyman has understood the difference (e.g. here: https://dlc.hypotheses.org/263), but others insist that there is no difference (e.g. Kiparsky: https://dlc.hypotheses.org/1817).
Language-particular description “depends on the phonological system and how it is organized” (as you put it in your last paragraph). But this does not mean that general terms like “form” or “segment” depend on systems – on the contrary, I think that general terms must be defined in terms of phonic or conceptual substance, not in terms of phonological or semantic systems. The difference between these two is the key insight of structuralism (https://dlc.hypotheses.org/2356), and I think we need to keep this in mind. But I know that the generative view (that all languages are made up of the same innate building blocks and that we are reasonably close to finding them) is currently more popular. I have long had this kind of view myself, but I realized that it is incompatible with rigorous general/comparative terminology, so I switched to a more traditional structuralist view.
I’m with the others in thinking of the ambitions as noble, but the exclusion of suprasegmentals as arbitrary. You said:
A “form” is necessarily segmental (its something that we can hear), so tonal units cannot be forms (and hence cannot be morphs).
While technically true, the way I see it, an exponent creates a difference in a word to bring in its meaning (presence of s in cats = PL, absence of s = absence of PL; presence of i in geese = PL, absence of i/presence of u in goose = absence of PL).
So while a purely suprasegmental exponent is unpronouncable, the change it brings to a word is hearable, (just) like with a segmental exponent. Outside of linguistic discussion you’ll never hear a segmental exponent detached from a word either (minor exception in next paragraph). Thus I don’t see this as a reasonable basis for excluding suprasegmental exponents.
Stronger evidence for an exclusively segmental morph is suspended affixation imo (German: An- und Verkauf, ess- und trinkbar). That’s most likely a phenomenon a suprasegmental exponent wouldn’t be able to undergo. Maybe it would find a default host or materialize. I think suspended affixation is rare enough for the conditions to likely not be met in the languages we know of sadly. If there are more phenomena where segmental and suprasegmental exponents behave differently, the split in terminology will become more justified, but I still contest the need to not group as different kinds of realizations of one underlying abstract morpheme.
Your strongest argument in this distinction seems to be tradition, I just think this would be a great opportunity to not only level the terminology but enhance it.
I guess at the end it’ll always at least partially boil down to personal opinions whether a change in form and the concatenation of form should be regarded equal kinds of exponence or not.
I didn’t know the field was so ill termed in this regard. I know linguistics is generally bad with terminology, but I’ve been taught four different types of morphs in my first semester of undergrad linguistics (surface phonological allomorph s/z/iz, phonologically conditioned true allomorph i/ka, lexical allomorph s/en/0 and another one I can’t even remember).
Yes, the exclusion of “suprasegmentals” (= “suprasegmental entities”?) is indeed arbitrary – my paper is about a terminological convention, and conventions are (by definition) arbitrary. I think what linguists need to understand is that their terminology must be separate from their theories, because our theories have not been converging much for decades – but we do have a chance of converging on better common terminology.
Yes, we can say that “an exponent creates a difference in a word to bring in its meaning”, but what is “an exponent”? Apparently it is defined as “a way of creating a difference in a word to bring in its meaning”. We know that there are many ways of expressing meanings, and many ways of talking about such “meaning-expressing methods”. We don’t agree on how to talk about them, but is this necessary?
I don’t think so – we can agree to use the term “form” when there is a segment sequence that is association with a meaning, and “construction” when there is more. My next paper will be on the definition of the term “construction”, I hope.
I like your attempt to make the terminology clear; when I was studying Morphology, I also found the term “morpheme” very confusing.
But I think your definition of “form”, and thus “morph” can make the analysis of languages with lexical tones very problematic. So my native language, Vietnamese, has lexical tones, and words that have the same segmental shape but different tones are usually unrelated in terms of meaning (so I think it is not possible to regard tone differences as “process” in this case). For example, the six following words contain the same string of segment /ta/ but different tones (the one written without diacritics means mid-level tone).
ta: ‘us, we, one’, kind of similar to “on” in French
tá: ‘a dozen’
tà: ‘wicked, evil’
So the segmental string alone, without any tone, means nothing. A weird scenario appears, where there is no pairing of content and segmental shape exists in such language, thus results in a language with no form (contrasting to the claim of Bloomfield that you cite in the article).
Please correct me if I get your idea wrong.
Note that I’m not talking about language-particular analysis in my paper – a key point is that general linguistics must be based on comparative concepts, and these are separate from language-particular analyses. So my paper has no consequence for your analysis of Vietnamese. But of course, it is useful to say that in Vietnamese, “ta” is a “morph” (adapting a general concept for language-particular purposes), and of course, we want to say that the various forms that you mention re different morphs. There’s nothing in my paper that goes against this traditional view. – But if you were to say that a certain tine sandhi process is “a kind of tonal affix”, that would not be compatible with my proposal anymore, because an affix is a morph, and a morph is segmental (while tone sandhi is a process).
Hi Martin. Long time listener, first time caller. I have a small comment. First, I applaud the efforts to (re)instate morph as the most useful and basic morphological primitive, rather than morpheme. In fact – as you point out in your paper but I don’t think stress enough – you have a perhaps unexpected parallel with the Distributed Morphology literature, where distinguishing ‘morphemes’ (essentially your ‘syntacticosemantic content’) from ‘vocabulary items’ has been standard for decades.
Now, my concern: you define ‘morph’ as ‘a minimal linguistic form’, and define this as “a minimal pairing of syntacticosemantic content and a string of phonological segments”. Reasonable. However, you also make explicit the following regarding suprasegmental shape (pitch, intensity, length)’, vowel apophony, consonant mutation, reduplication, etc.: “According to the definition proposed here, a morph (or any other form) cannot be nonsegmental”.
There is no reason to maintain this. My own angle comes from grammatical tone in African languages, where the familiar morphological categories – class, case, tense/aspect, polarity, plurality, agreement, the range of derivational categories, etc. – can be expressed either partially or exclusively via tone. Larry Hyman emphasized this in his recent LSA presidential address. While some of these may best be analyzed as tonal ‘processes’ (which is an analysis requiring argumentation, not an observation), in numerous cases it has been just as profitable to see them as autonomous tonal units of contrast on a separate tier (i.e. the Autosegmental revolution). And thus while they may be ‘nonconcatenative’ from the view of segments, they are concatenative from the view of tones.
TLDR: linguistic forms can consist of segments, phonological features, tones or accents, ‘intonemes’, or other such primitives (e.g. those in sign languages). Morphology suffers from ‘segmental bias’. By excluding all but segments out of hand is unproductive to establishing the parameters of variation needed to capture morph structure and behavior typologically, and leads to its (continued) limited role in all current morphological theories.
Thanks for the thoughts on your blog, as usual.
Many thanks, Nicholas! You’re absolutely right about the “segmental bias” in much of linguistics, probably due to the fact that in European languages, we don’t have tones. But how do we overcome such biases?
I think that in general, the best way to overcome biases is by becoming aware what our words mean, not necessarily by changing their denotation. Linguistics is biased toward spoken languages, so how do we overcome this? Should we say that ASL or DGS users “speak with their hands and faces”, condemning words like “sign”? Or should we say that linguistic shapes can be acoustic or visual, and linguistic articulation can be by speaking or by signing? In this case, our discipline opted for the latter, maintaining the meanings of the old words – and I think it’s reasonable to be similarly conservative when it comes to the term “morph”.
The reason is that the only way to define “morph” is as a type of “form” (or a type of “expression”), but it wouldn’t be in line with normal usage to say that “a tone is a form”, or “voicing is a form” (to take another nonsegmental “unit”, represented in English as in half/halve, calf/calve, house/hou[z]e). A “form” is necessarily segmental (its something that we can hear), so tonal units cannot be forms (and hence cannot be morphs).
Recall from my paper that it is nonsensical to reduce “morphology” to morphs, because there are obviously so many other things going on. It’s true that the term “morph” might suggest that it is “the basic unit of morphology”, but that’s wrong. There is no such thing as “the basic unit of morphology”, and in fact, I don’t think there is such a thing as “morphology” (I don’t know what a “unit” is, or what a “word” is). If we lived in a world where all our terms had clearly defined meanings, nobody would even think that a morph is somehow related to “morphology”, because a morph is defined as a minimal form.
So how do we express (or “capture”) cross-linguistic variation best? I think it’s by continuing to enrich the set of clear terms. I’m currently working on a paper that defines “bound form” in a clear way, and adds a new concept “welded form” (https://dlc.hypotheses.org/1779). It still leaves out suprasegmentals, so it won’t make you happy, but that’s because I have to start with the languages that I know best, and with the terms that are most widely used.
If we can say that /abada/ is a morph, we could also say that /HLL/ is a “morph”, possibly qualified in some kind of way. In a sense, it is concatenative in its own way.
Ablaut processes like in sing/sang should not be called morphs or morphemes, nor should processes acting upon tones be called such (´`´ -> ´´´ ).
In a tone language, /abada/ without suprasegmental information might be as meaningless as /`´´/ without segmental information. Neither might be a “minimal form”, so I see no reason to privilege one over the other here. But we can privilege pure *concatenation* over *processes*, which seems to be the core of the disagreements raised by sing/sang.
Yes, it is true that in a language that has /ábàdà/, neither [abada] nor [HLL] exists independently of the other – the tone pattern is an inseparable part of the form, just like the consonants and the vowels. So I agree that [abada] would not be a minimal form in such a language.
But note that we cannot say that just because there are some languages (without tone) where /abada/ is in fact a form, the element [abada] in a tone language must also be a form (so that [HLL], which has the same status as [abada] in this language, would also be a form).
I insist that a morph must be a form, and hence a tone pattern cannot be a morph, because a tone pattern is not a form.
I am not privileging [abada] over [HLL] – because neither is a form in a language that has no toneless roots. All I’m doing is privilege /ábàdà/ over /HLL/, because the former can be pronounced, while the latter cannot. I think this is in line with actual usage, and this is the only reason for my terminological choices – I’m just insisting that we should not define words in ways that are incompatible with actual usage.
Let’s give that language a /p-/ prefix. Concatenating our two morphs we’ll get [pábàdà].
You can hear [p] in [pábàdà] because the stop is released and changes the formants of the following vowel. However, [p] in isolation is not a pronouncable sequence.
[p̚] might be articulatory pronouncable but still only as audible as a tone without a tone-bearing unit.
Yes, indeed: Many (or most) segments cannot be pronounced in isolation. But I didn’t say that a form is something that can be pronounced in isolation – I said that it is something that can be pronounced. I take that as a definition of “a form”. I’m not perfectly happy with this either, but it’s the best I was able to do. We certainly don’t want to say that any feature can be a form, e.g. the “labial” feature of the vowel in “book”, or the “stressed” feature of the first syllable in “booking”, etc. In actual practice, linguists use the term “form/expression” in a vague, sloppy sense – but *very largely* in conformity with my definition. Linguists rarely think about definitions of terms – but once one starts thinking about the definition of “affix”, one arrives at these sorts of considerations…
But one problem, Martin, you don’t seem to define what you mean by “can be pronounced”. If you accept “can be pronounced in combination with other sounds” for the imaginary /p-/ prefix in the other comment, then tones are also “pronouncable in combination” as well. Both tone and plosive seem to require an accompanying continuant to be pronouncable.
I see no problem in rejecting stand-alone tones as form. But the question is that, under the same phonetic principle that you reject tone, should stand-alone “segments” be rejected as “form” as well?
Yes, it’s true: I cannot really define “can be pronounced” independently of “sequence of segments”. Maybe I wrongly conveyed that impression. So why do I say that a form must be a continuous sequence of segments? There is no deeper reason: It simply seems to conform best to the way that people actually use the word “form” (recall my principle of conservativeness). I don’t know of any other definition of “form” that is simple and easily comprehensible. If a tone pattern counts as a form, then why exclude intonational patterns? But surely we would not say that “the rising intonation pattern in English questions is a form” – this sounds weird. For segment sequences, saying that something is a form never sounds weird, as far as I can see.