1. a Cognitive Grammar perspective David Tuggy SIL-Mexico
Phonemes
a Cognitive Grammar perspective David Tuggy SIL-Mexico

2. Phonology is part of cognition
Phonology is the study of the sounds of language.
Language is something people do using their minds; it is a kind of cognitive activity.
Phonology is not different from the rest of language in this.
It involves moving body parts to make changes in the outside world (in this case, sounds), but so does a lot of other cognitive activity.

3. Phonemes are Categories
An extremely important aspect of phonology, as with many other areas of cognition, is categorization, the establishing of groups that can be treated as one thing.
Phonemes are a special kind of phonological categories, in which different sounds are, for certain purposes, treated by language users as one thing.

4. Categorization So let’s talk about how we do categories.
I will be using the words “category” and “class” to mean somewhat different things.
A “class” is a group of concepts (cognitive structures) of whatever sort.
A “category” is a class that has been established in a person’s mind,
Especially if it has been conventionalized for a group of language users.

5. Categorization A “class”, then, is a potential category.
Classes get turned into categories by repeated usage that
1) entrenches (establishes) the grouping in users’ minds
2) establishes the grouping’s conventionality (it’s being commonly established among all the relevant users).

6. Categorization
A category is conventional when I know it, you know it, I know you know it and you know I know it. It is shared among those in our group.
A language consists (*only!!*) of conventional categories. Its phonology is no exception.
(When you study phonetics, you study the effects of such categories on the outside world.)

7. Usage-based categorization
We learn categories through usage: repeated occurrences (active or passive) of them in our minds.
They are *always* based on similarities.
(Though as a limiting case, such as that of “cabbages … and whether pigs have wings”, that similarity may be only that of occurring simultaneously in the mind.)
Let’s look at a non-phonological example or two.

8. Usage-based categorization
Suppose a child learns (through repeatedly encountering it) the concept of a human baby.
The child then learns what a kitten and a puppy are

9. Usage-based categorization
The child compares PUPPY and KITTEN with BABY
perhaps after hearing a puppy and/or kitten called a [ˈbejbi] just like a human baby is.

10. Usage-based categorization
The comparison shows both similarities and dissimilarities.
This is traditionally represented in CG by a dashed arrow (from Standard to Target, S - -  T).
The relationship so symbolized is one of partial schematicity.

11. Usage-based categorization
Such a comparison prompts the extraction of a schema.
The schema consists in what the compared concepts have in common.

12. Usage-based categorization
The relationship is represented by a solid-line arrow from schema (Standard) to subcase(s) (Target(s)).

13. Usage-based categorization
This is a relationship of full schematicity.
There is no conflict between the specifications of the standard and those of the target.

14. Usage-based categorization
An intermediate schema may be extracted embodying what’s common to the puppy and the kitten

15. Usage-based categorization
And so the category (grouping of concepts) grows

16. Usage-based categorization
ALL these concepts are schemas

17. Usage-based categorization
Schema = pattern = generalization over specific cases

18. Usage-based categorization
All linguistic structures are patterns—they are all schematic in some degree

19. Usage-based categorization
BABY, KITTEN and PUPPY are generalizations over different specific babies, kittens and puppies.
The concept of a particular baby, kitten or puppy is in its turn a generalization over even more specific experiences of that real-world entity.

20. Usage-based categorization
As we have seen, categories can become very complex.
(The meaning of baby is of course considerably more complex than what we have represented here.)
Note that not all the members of the category are equal.

21. Usage-based categorization
Some concepts are better-established (through usage) than others, and hence more salient.
HUMAN INFANT is the most salient meaning of the word baby.
It is the prototype for the category.

22. Usage-based categorization
Similarity relations involving the prototype are naturally likely to be more salient and important than others.
Relations of full and partial schematicity are, by essence, relations of full or partial similarity.

23. Usage-based categorization
Thus the whole network is based on similarity relations.
Which concepts are compared and assimilated into a category is an interesting question
It depends on two important factors.

24. Usage-based categorization
The first is “brute” or “etic” similarity: more-similar concepts are more likely to be assimilated than less-similar ones.
But the deciding factor is the effect of usage.
The members of a category are not those that might have been included, but those that have been.

25. Usage-based categorization
A shoot from a tree root is similar to some concepts in the meaning of baby.
It might well be included in the meaning of baby, but in fact it isn’t.
(A Nahuatl speaker might call it a konēƛ ‘baby’, however.)

26. Usage-based categorization
It would be perfectly natural to compare the concepts LAMB and KID, extract a robust schema uniting them, and make that category the meaning of a lexical item.
In English we don’t do that.
In (Biblical) Hebrew they did.
This usage, often repeated, established the category as conventional for them.

27. Natural classes & Classical categories
In many Americans’ minds, the category COCKROACH has a very well-entrenched schema
Any sub-categories (sub-cases) are far from salient.
In effect, the highest schema of the category is also its prototype.
It is not necessary that the category be structured so in our minds.

28. Natural classes & Classical categories
For my wife there are two salient sub-classes
For an entomologist, the six families might be salient, and there would be a strong prototype, the American Cockroach
But for most of us a cockroach is a cockroach and there’s (hopefully) an end to it.

29. Natural classes & Classical categories
It is, then, natural (cognitively easy) to take cockroaches as a class without distinguishing among them.
Such a class may be called a natural class.
Natural classes, if they become entrenched and conventionalized, form classical categories.
In classical categories, the highest schema is also the prototype.

30. Classical categories
“Classical” categories are the kind assumed by many theories to be the only kind.
They are modeled when
(a) a salient (well-entrenched) pattern or schema
(b) has many less salient sub-cases, and
(c) any other schemas based on those sub-cases can be reasonably ignored. As a result,
(d) The category does not overlap with neighboring categories.
(e) The result is that you can almost treat the category as if the schema were all there was to it.

31. Categories Well-behaved “classical” categories look like this:
Or, even better, like this:
Such a set of classical categories with no overlap between them is a classical categorial system.

32. Categories
But “classical categories” are just one case of the range of kinds of categories we find in language.
(The meaning of baby, for instance, is nowhere nearly a classical category.)

33. Categories
Another way to model categories is to think of the mind as slightly loose soil or swampy terrain.
A usage event is like a kid on a pogo stick: each time he jumps he tamps the dirt down a bit.

34. Categories
Each concept is at a particular location, so each time the language user activates that concept, that location gets tamped down a little further.
Every time a closely similar concept is used, a neighboring spot is tamped down a little further.

35. Categories The complex hole so formed is a category.
Wider parts of the hole correspond to schemas.
Narrower parts are subcases.
The widest part is the highest-level schema.

36. Categories
In a classical category, the hole is deep and uniform: no one particular subcase within it gets activated differently enough to draw attention to itself.

37. Categories
But of course any number of other kinds of holes (categories) are likely to develop.

38. Categories
Classical categories can occur, but more often you get near-classical or non-classical categories.

39. Phonological Categories
Phonological structures, as much as semantic structures, are cognitive in nature.
Phonological categories are of the same kinds as any other linguistic categories.
Phonology, too, is founded on usage.

40. Phonemes
Perhaps the most important thing I can say about phonemes is that they are near-classical categories.
Classical phonemic theory insisted that they were pure classical categories forming classical categorial systems.
Theorists since then have tended to point to evidence that they fail of absolute purity
and conclude that therefore they do not exist at all.
This conclusion not just logically flawed and unnecessary; it is pernicious. It distorts your conception of phonology.

41. Phonemes Phonemes exist, and are important.
Most if not all languages do in fact have a set of contrasting sounds that tend to be structured close to the norm for classical categories.
They are abstracted from millions of individual usage events. As usage changes, they too will change.
They tend to consist of a single articulatory position or gesture, though sometimes a complex one.

42. Phonemes
They are of two main types: syllable peaks, and syllable margins (onsets or endings).
Among them they cover the range of articulatory gestures of the language pretty completely, with little overlap.
These are the phonemes of the language.
The syllable peaks are vowels, at syllable margins are the consonants.

43. Phonemes
They tend to be formed around natural classes, i.e. usage (the pogo stick jumper) tends to respect the natural contours of the land.
But they don’t do so perfectly enough to be 100% predictable.
They are *not* necessarily consciously prominent in the minds of naive speakers, but they are discoverable.
The classical techniques for investigating phonemic structure, if not applied mindlessly, work well for helping discover them.

44. Phonemes
Their existence is the best explanation for the fact that reasonable alphabets can be made for most if not all languages.
It does not follow that all contrastive elements fall into well-formed classical categories.
(There are serious difficulties in some cases, notably in the analysis and writing of tone.)

45. Phonemes
They show up through examination of general patterns and particularly of alternations.
Alternations are cases where a morpheme with a constant meaning consistently changes in one small place, especially when it occurs in different environments.
The following examples are from Mösiehuali (Tetelcingo Aztec or Nahuatl).

46. A “classical” phoneme
The phoneme t consists largely of a group of sounds which, though differing slightly among each other, have in common a (non-nasal) apico-dental gesture at the margin of a syllable (and its acoustic correlates).
Although specific pronunciations (subcases) differ in detail, none of those details particularly stand out.
The highest-level schema is the prototype (most salient member).
So this is a good example of a “classical” category.

47. A “classical” phoneme
The particular pronunciations are called “allophones”.
The whole point about them is that they do not stand out much.
If they did, they would make it easier not to take the category as a whole.

48. A “classical” phoneme
Note that, even though an alveolar stop would be very similar to these other sounds, and you might expect it to be assimilated to the category, it isn’t there.
That’s just because Mösiehuali speakers don’t hear that kind of a t very often, and don’t learn how to pronounce it. Again, it is usage that dictates what’s in the category.

49. A “classical” phonemic system
t fits in with a number of other similar categories in a system which pretty well covers the territory of sounds pronounced at syllable margins. (Only the obstruents are represented here.)

50. A “classical” phonemic system
These sounds, by and large, contrast clearly with each other, each potentially occurring between vowels, for instance.
This looks like a good “classical” categorial system. If the phonemes didn’t overlap, it would be perfect.

51. A “classical” phonemic system
Schemas can be extracted to group or categorize these phonemes, but they are not likely to be as useful or prominent as the phonemes themselves are.
These are CG counterparts to the traditional “archiphonemes”, or in some cases to phonological features or feature combinations.

52. A messy phoneme
Not all phonemes present such a pretty picture, however.
For example, the phoneme w in Mösiehuali includes a range of pronunciations including [w], [w̥], [β], and [ɸ].
These “allophones” are clearly salient in people’s minds, to such an extent that /w/’s unity can be (and certainly has been) questioned.

53. A messy phoneme
Yet when the allophones are compared a schema is easily extractible.
They clearly have a lot in common.
Yet they are different enough that one could easily separate them, and many languages do.

54. A messy phoneme
How do Mösiehuali speakers learn not to separate them? Through usage.
The usage they hear gives them two very important kinds of evidence; pattern evidence and alternation evidence.

55. A messy phoneme
In fact, there are other sub-generalizations that can (and should) be included in the picture.
(The relations between these sub-generalizations embody CG’s analogue of phonological rules)
This phoneme is a much messier category than t, č, and the others.

56. A messy phoneme
In fact, w is already such a messy category that adding h to it may not seem like that big a change.
This extension is well-established by alternation evidence: e.g. kʷaβɪƛ = kʷaw-iƛ ~ kʷahme = kʷaw-me ‘tree~trees’, or kɪhtowa = ki-htow-a ‘3p says it’ ~ okɪhtohkɪ = o-ki-htow-ki ‘3p pl said it’.

57. A messy phoneme
This is where the “submarine wiring diagram” fits in: linkages outside of the ovals tend not to have a coherent schema embracing all and only the relevant (allo)phones.
Still, it is less than clear what schema (if any) one should posit as extracted from the extension.
Certainly that schema would not fit well into the nice set of phonemes in Mösiehuali.

58. Neutralization to h
More importantly (perhaps), this h “allophone” (or particular subcases of it) is shared by other phonemes.
These phoneme categories overlap at this point.

59. Neutralization to h
What is more, these h “allophones” fit quite well among the non-distinguished allophones of another phoneme, namely h.

60. Neutralization to h
What had looked like a nice neat classical system suddenly isn’t—it allows massive overlap.

61. Neutralization This sort of thing shows up in many languages
even though not quite on such a scale in most languages

62. Neutralization
The conclusion, however, should not be that the phonemes do not exist.
They just aren’t perfect. We shouldn’t have expected them to be.

63. Consonant Neutralizations in Mösiehuali
Under at least some other theoretical models this sort of thing is essentially inexplicable.
If purely classical categories are the basis of phonology, why would you ever get this kind of a mess?
But, it is in fact quite normal, empirically, to get this kind of mess
(even though not usually on this scale).

64. Summary: Phonemes as categories
The main points I would draw from all this are:
Phonological categories are like other categories we find in language.
Like other categories, they are usage-based.
It is usage (along with , not instead of brute “etic” similarity) that lets us know which sounds (or other concepts) are to be compared and categorized together.
What we call phonemes are a kind of phonological categories.

65. Phonemes as categories
Like other categories, phonological categories may be nice, neat and orderly (“classical”), or they may be messy. Phonemes are near the classical end of this spectrum.
However, most of the time they are at least a little bit messy.

66. Phonemes as cognitive categories
From the CG perspective, aberrations from the classical phonemic model are perfectly normal and to be expected.
What needs explaining is not why the phonemic model isn’t perfect, but rather why it works so well (comes so close to the facts) in so many cases.

67. Phonemes as cognitive categories
(fwiw, though we haven’t shown it) The CG perspective on categories allows us to characterize, in natural and naturally related ways, such traditional structures and mechanisms as:
Morphemic alternation rules
Morphophonemic and other phonological rules
Rule ordering
Phonemes and allophones
Archiphonemes
Features
Neutralization

68. Phonemes as cognitive categories
I think this is important.
If a model obliges me to say that older thinkers were all wrong, something is probably wrong with the model.
If it helps me understand what those older thinkers were talking about, that it is a point in its favor.

69. Phonemes as cognitive categories
And, finally,
Aren’t languages like Mösiehuali beautiful?
Isn’t it admirable how a community of language-speakers can allow, and even encourage to the point of enforcing, such massive neutralizations,
and still be able to tell perfectly well what they are talking about?

70. Power Point to be available at