Most intriguing (to me, anyway) is what Chaser’s achievements add to existing knowledge about canine understanding of categories. Earlier research suggests that dogs have this ability, but Chaser is the first to offer evidence that a dog can use human words to categorize things (as opposed to visual or non-linguistic auditory stimuli). As I wrote in my previous diary, Chaser understands the word toy to mean any one of the 1,022 things she is allowed to play with (and has individual names for) and recognizes the words ball and frisbee as names for mutually exclusive subcategories to each of which some of the toys belong (by virtue of their being spherical and bouncy for the former or having “disk-like qualities” for the latter), each with its own individual and distinct name.
As Pilley and Reid note:
By forming categories represented by common nouns, Chaser mapped one label onto many objects. Chaser also demonstrated that she could map up to three labels onto the same object without error. Experiment 1 demonstrated that Chaser knew the proper-noun names of all objects used in this study. Chaser also mapped the common noun ‘toy’ onto these same objects. Her additional success with the two common nouns ‘ball’ and ‘Frisbee’ demonstrates that she mapped a third label onto these objects. (192)Here's my attempt at a visual representation of Chaser's understanding of categories:
Much of the discourse surrounding the research with Chaser and other studies in dog linguistics has focused on the lexical: the extent to which the dogs understand words. But in spoken language, words are just combinations of sounds, specifically speech sounds (realizations of phonemes) that in combination become distinct and meaningful. But as we’ve seen, the phoneme is not a discrete, static thing but is fluid, variable, and relational , as are the ways in which speakers (and apparently non-speakers, including dogs) come to understand them.
Where all this is going for me is in two directions, both of which lead eventually — at least in my head — back to The Phoneme and the Many Lives It Has Destroyed, with The Phoneme functioning simultaneously in its more or less literal linguistic-terminology sense as well as in a somewhat more metaphorical sense. These two directions have to do with:
1. how sentient beings, including people and dogs, associate speech sounds with particular and distinct meanings; andSo, can dogs predict phonemic splits?
2. how we (again, people and dogs) conceptualize categories, relations among their constituents as well as across boundaries, and the boundaries themselves.
Considering the mercurial nature of The Phoneme (and of attempts to define it), alongside the ways in which we (people and, it turns out, dogs) use categories to make sense of things and language to help define them, got me wondering about whether dogs might be sensitive in ways that humans may not be with respect to phonological variation among speakers, and if so, what that might mean about their possible perceptions of sound changes in progress.
In my recent exploration of The Phoneme, I note that one way of conceptualizing it — as a (discrete) unit of sound — is complicated by the rest of its definition, that it is a class of speech sounds that a native speaker will identify as the same sound. An “individual” phoneme, then, is actually a category containing multiple sounds, somewhat like (although it is far from a perfect analogy) Chaser’s categories toy, ball, and frisbee, the first of which contains not only the other two (as well as everything the categories ball and frisbee respectively contain) but another nearly 900 words in addition. Toy also functions to distinguish for Chaser what she is allowed to play with from what is off-limits, or not-toy.
Similarly, the thing we call a phoneme is a category that contains qualitatively similar but not identical sounds that, despite their variability, native speakers of a language will interpret as "close enough" to one another so as to be (more or less) interchangeable without affecting meaning. In other words, the differences among those similar-but-not-identical sounds can be ignored. The 26 objects that Chaser knows as members of the class frisbee are not identical to one another, but if Chaser is anything like my border collie, Scarlett, she is probably willing to ignore the differences because all frisbees are good frisbees simply by virtue of their being frisbees –- flying discs that are fun to run after and catch spectacularly -- unless she is directed to select a particular frisbee for which she has learned a unique name.
My girl Scarlett has a large collection of frisbees of various materials (e.g. canvas, plastic, rubber) and different colors. She doesn’t seem to care which one she plays with, as long as someone will throw it for her. (Quinn, our Australian shepherd, on the other hand, has developed a distinct preference for the firm but flexible plastic Hyperflite Jawz Hyperflex™ brand frisbees , although he has not signed any endorsement deal that I am aware of, so this shout-out is on the house.) While it is sometimes possible for member-sounds (i.e. variant realizations or allophones) within a category or class of sounds (i.e. a phoneme) to sound less like some of the other members of its own class and more like sounds considered to be members of other classes, it seems much less likely that any frisbee could have more in common with any ball than with other frisbees. Based on her past behavior, I think there is pretty much no chance that Scarlett will bring back a ball when she is instructed to “go get the frisbee.” She’s never done that. She always brings back a frisbee.
The apparently essential (to Scarlett, at least) frisbee-ness of flying discs notwithstanding, all categories — linguistic and otherwise — are going to be arbitrary to some extent. One of my favorite linguistic examples of this arbitrariness is the way certain realizations of speech sounds are considered members of separate categories (different phonemes) on the basis of qualities that are language-specific, such as the quality in English that distinguishes [i] (the vowel in seat) from [Ι] (vowel in sit): the so-called tenseness (in the case of [i]) or laxness (in the case of [Ι]) of the vowel, a quality that is phonemic in English but noncontrastive in Spanish, in which [i] and [Ι] are close enough for speakers to ignore the differences between them.
So, what does all this have to do with whether dogs can predict phonemic splits? Well, since Pilley and Reid present good evidence that dogs have the cognitive ability to determine and comprehend categories and that they can even understand and interpret language (in the form of meaningful combinations of human speech sounds) to do so, I am wondering about whether they understand phonemic categories in the same ways that human speakers do. Would a dog who grew up in an English-speaking household interpret a meaning distinction between sit and seat? Or would a dog who has been responding since puppyhood to the human-articulated instruction to “sit!” by doing just that hear the two words as close enough phonemically (and/or semantically) to respond the same way to an instruction to “seat!”? Would a dog raised in a Spanish-speaking family to come running when given the instruction “ven!” (come!) -- pronounced [ben], which is roughly similar to the English word bane -- respond in the same way to an American English-accented version of the same command that sounds like [vεn]? 
In Spanish, [b] and [v] (and also [β], although it is less commonly articulated at the beginning of a word) are allophones -- variant realizations -- that belong to the same class of sounds, namely the phoneme [b]. In many varieties of Spanish, word-initial v is often pronounced [b] and rarely [v]. In English, [b] and [v] are two distinct classes of speech sounds (and words spelled with v are always pronounced [v] and not [b] by native speakers). The difference between [b] and [v] is thus phonemic in English, which can be demonstrated by considering a minimal pair like ban and van. Since the only difference between ban and van in terms of sound is in the initial consonant of each word, the difference in meaning – in English, ban and van are of course distinct words with different meanings – shows that the initial consonants are members of different classes of speech sounds, which is another way of saying that they are phonemically distinct.
In Spanish, the difference between [e] and [ε] is also not phonemic; both sounds are allophones (variant realizations) of [e]. In English, however, the distinction between [e] and [ε] is phonemic, as evidenced by the meaning difference between the English words main [me:n] and men [mεn].
The other day, my husband and I were talking about whether dogs can distinguish between variant pronunciations of unstable sounds like [æ], the vowel that is more or less usual in American English pronunciations of the word cat but which is highly unstable (i.e., variable) in American English, partly as a result of the Northern Cities Shift (NCS), a set of apparently interrelated changes in vowel pronunciations. The epicenter of NCS is the Great Lakes region, where we have lived since 2004 after having spent most of our respective lives at various points along the east coast, from Florida (me) to Virginia (him) to Washington, D.C., North Carolina, and Georgia (both of us).
Even after nine years in Michigan, the differences between NCS and non-NCS pronunciations are still so salient to my ears that NCS speakers sound to me like they are articulating a different phoneme from the one I use to pronounce words with the [æ] sound. To me, their pronunciations of cat sound like [kejæt] (‘KAYat’) or [kiæt] (‘KEEat’), with the [æ] raised noticeably (to me, anyway) as well as tensed and diphthongized. This is very different from my pronunciation, which has a slightly backed [æ] that is also low in contrast to their raised version. My pronunciation is farther back than [æ], sounding more like the low-central [a] than a full-on low-front [æ]. Native Michiganders with NCS features in their own speech who hear the difference (and they don’t all hear it) describe my low-central pronunciation as sounding to them more like [kat] and articulate something that sounds like [kɑt] ('kaht') when they try to imitate it.
Scarlett, who had been napping on the kitchen floor, immediately jumped up, dashed to the living room window, snapped to attention with her paws on the window sill, and looked for the cat.
You may be thinking that this incident does not bode well for my hypothesis that dogs may be able to predict phonemic splits since it was clear that Scarlett completely disregarded what to me is a highly salient pronunciation difference that I can’t believe is not going to end up phonemically distinct sooner or later in American English. Fair enough. I would point out, though, that Scarlett has lived in Michigan for four years, since the age of 16 weeks (and she came from a rescue in northern Ohio), so she is as good as native to the region.
I will acknowledge that this anecdote, while charming, is nowhere near enough evidence to go on. But this post is getting long, so I will just say that in light of Scarlett’s enthusiastic response to my NCS articulation of cat (and the complete disregard it reveals on her part for what I consider a substantial difference from my usual pronunciation), it may be that dogs cannot predict phonemic splits, that they may not be any more sensitive to slight (cough) pronunciation differences than people are. However, we should remember that even American dogs are native speakers not of American English but of barking, growling, yelping, yipping, tail-wagging, and other vocal and nonvocal means of expression that categorically do not include American English or any other human language.
The other night I did another little experiment with Scarlett and found that she will respond to my instruction "besos!" (Spanish for kisses) exactly as she does when I say "kisses!" (i.e. by running over to me and licking the tip of my nose four or five times in rapid succession). I think maybe all those alveolar fricatives (and I hear myself definitely devoicing the one on the end of kisses, just as the one on the end of besos is voiceless in Spanish) and similar prosody in my pronunciation of both words in the context of giving her that particular instruction are all she needs to hear to figure out what she’s being asked to do. It might also help that she and I don’t have any other commands that have a similar combinations of sounds, so she might be running the combinations through her language processor and, not finding anything closer, selecting kisses as the closest in terms of sound and therefore most reasonable interpretation of besos.
And of course she is probably also reading all kinds of nonverbal cues. About 10 years ago, my family fostered a deaf Australian shepherd puppy for several months to help prepare her for adoption, and in the process of working with her, I learned that even hearing dogs don’t necessarily need verbal commands because they learn the nonverbal behaviors that accompany each particular command. Those behaviors (gestures, facial expressions, etc.) can be enough to transmit the command. Some trainers say that is really all they ever respond to and that the verbal commands are all to make life easier for the human, but I think the example of Chaser and her lexicon of 1,000+ words may present a significant challenge to that theory.
And so even if dogs cannot predict phonemic splits (although I am not ready to rule out the possibility), even if they pay attention to what we pay attention to and ignore what we ignore (linguistically, anyway), it is really quite remarkable that somehow they have learned to do that. Even if they are no more sensitive to phonological variation than the people who love them (or maybe they are but learn not to be), in a sense they are still demonstrating an impressive capacity for a level of understanding that is not necessarily available to humans when we learn languages non-natively.
This capacity may have to do with their cognitive and linguistic abilities, and it may also be that they have other ways of getting information from us, including the kind of information that can compensate for phonological variability and instability, in yet one more way that dogs have learned and evolved to adjust themselves to human limitations. They are probably better than humans are at discerning differences in sounds, but they also seem to be really good at adapting to human ideas of what’s “close enough” to the point that they may even be more flexible on this than we are. All I can say for certain at this point is that however they do it, it is clearly another example of the all-around awesomeness of dogs.
 As I have discussed previously, the idea of the phoneme as the American structuralist linguists conceptualized it is not discrete but relational, which is to say it exists only in relation to other sounds. The linguistic anthropologist Edward Sapir (1884-1939) was among the first to articulate the relationality of speech sounds in a 1925 paper, "Sound Patterns in Language," which is included in The Collected Works of Edward Sapir, full text available here via the Internet Archive.