I. Review of Assigned Readings
Pinker:
Phonology
If we defined phonetics as the study of the sounds used in natural language, then we can think about phonology as the study of the systematic ways that these sounds behave both within and across languages. Phonology, for example, looks at the way that sounds are organized in a language. Thus, as we have seen, two languages may both have aspirated voiceless stops, but these may appear in a completely predictable manner (as in Enlish) or in an unpredictable manner (as is the case in Hindi, for example).
Specifically, we've spent quite a bit of our time looking at the notion of phonemes and allophones, as well as the notion of relating the phonemes to allophones by phonological rule. As we recall, the phonemic level of representation is called the UR or Underlying Representation and the allophonic level of representation is called the PR or Phonetic Representation. Phonological rules map between these levels:
UR (the level of the phoneme)
Phonological Rules
PR (the level of the allophone)
In order to understand why we need this kind of a model for understanding how phonology works, however, we have to start with an important premise that is not necessarily obvious at first glance. This premise is the following:
"Important Premise":
Understanding the phonology of a language (and, by extension, of languages in general) is quite a bit more complex than simply memorizing the way in which each of the words of a language must be pronounced.
Imagine that we do simply memorize all of the details of pronunciation of every single word we know. For example, imagine that we simply memorize that the English word TIGHT (I use capital letters to refer to the word independent of its pronunciation) is pronounced as it really is. That is, imagine that we commit to memory the following: to say the word TIGHT, I have to:
So, big deal, you think to yourself, I can memorize that. Well, you can, but there is a deeper problem lurking here. Let's call this the problem of generalization. If we assume that we simply memorize all of the aspects of pronunciation of all of the words of a language, how come we are able to pronounce perfectly words that we have never heard before? This is, in fact, what a psychologist named Berko-Gleason first showed back in the dark, pre-disposable lighter days of 1958. She presented small children with pictures of funny looking imaginary animals, one of which was called a bik and the other of which was called a wug. She presented these pictures to kids, and when they were asked to tell her how to say more than one bik or more than one wug, the kids systematically provided the phonologically appropriate form of the plural--i.e. an [-s] after the [k] sound, but a [-z] after the [g]. And if you think about it, if someone gave you a new word "zoak" written on a piece of paper, you wouldn't sit around scratching your head over how to pronouce it. Imagine the following:
- i) first make a voiceless alveolar stop sound that is aspirated;
- ii) then make a comparatively short [ay] vowel sound (if I speak a dipthongizing dialect); and
- iii) then make a voiceless alveolar stop sound that is unreleased.
What Berko-Gleason's experiment showed, and its importance cannot be underestimated, is that knowing how to pronounce words involves knowing a system. The problem with simply assuming that we memorize all we need to know about the pronunciation of every individual word is that it fails to capture significant generalizations about how our language works. For example, in the word TIGHT, it's not simply the case that we know that the first /t/ is aspirated and the final /t/ is unreleased, as if this knowledge existed in some kind of vacuum. Rather, the aspiration of the initial /t/ is intimately related to the aspiration of the initial /k/ in a word like KICK and to that of the initial /p/ in a word like PICK. Likewise, the lack of release on the final /t/ in TIGHT is initimately related to the lack of release on the final /k/ of PICK or of the final /p/ of TIP. We know whether a /p/, /t/, or /k/ will be aspirated, unaspirated, or unreleased, regardless of the particular word in which each might occur, because we are able to more generally identify contexts in which the different manifestations of these sounds predictably appear. Assuming that we simply memorize the pronuncitation of each word in our language fails to capture any of these generalizations. Furthermore, it fails to make any predictions about the realizations of sounds such as /p/, /t/, and /k/ outside of words which we have not already commited to memory. In both of these senses, it is unsatisfactory.
- Psychologist: "Can you pronounce the word spelled 'z-o-a-k'?"
- You: Gee, I haven't previously memorized that word. Sorry.
So, does this mean that we don't memorize stuff? Absolutely not. Much of knowing a language requires memorizing idiosyncratic information. This is because the relation between sound and meaning is arbitrary. Anyone who speaks two unrelated languages can see this quickly. In Coatzospan Mixtec, a language of southern Mexico, the word meaning GOOD is /va?a/, while the English form is 'good'. These words are phonetically quite dissimilar yet mean roughly the same thing. The bottom line is that it is an arbitrary fact that English has chosen to use the sequence of sounds [g] + [high, back, lax, rounded vowel] +[d] to express this particular meaning. This is indeed something that we must commit to memory if we are to learn English, just as we must commit to memory that CAT is /k-ae-t/. But what isn't simply an unpredictable fact of English is that the vowel of 'good' is going to be realized as phonetically longer before the final voiced consonant /d/. And it isn't simply an unpredictable fact of English that the /k/ of CAT is going to be aspirated. So, there's a fundamental question here of distinguishing between what we've just got to memorize by brute force and what constitutes a systematically predictable aspect of the way sounds are realized in a language.
Put it this way, I can't tell you if it's going to rain next April 18th, but I can be pretty sure that if it rains, the ground will get wet. Similarly, I can't tell you if a particular word is going to begin with a voiceless stop followed by a stressed vowel, but I can tell you that if it does, the voiceless stop will be aspirated. And that's the heart, the first step, the initial point of departure in our looking at how phonology works.
The Big Picture
Once we recognize the importance of distinguishing the predictable from the unpredictable, there are a number of significant consequences. The first is that we become committed to a view of phonology that involves two levels of representations. One of these will have to provide us with all the information that is idiosyncratic (i.e. unpredictable), while the other will provide us with a representation of what we actually say when we produce a word (i.e. the unpredictable information plus all of the predictable aspects of pronunciation).
These two levels are what we've been calling the levels of Underlying Represenation (UR) and Phonetic Representation (PR), respectively. Another way to think about this is to say that the UR provides us with the minimal amount (and only the minimal amount) of phonological information that we have to know about about a given word, while the PR provides us with all of the aspects of the pronunciation of that same word. If you think back to phonetics, you'll see that these two levels are actually what lie behind the idea of broad vs. narrow transcription. When you do broad transcription, you are transcribing the phonemes that make up the underlying representation of a word. But when you do narrow transcription, you are transcribing the allophones that make up the phonetic representation of the word.
Let's go back to our example of the word TIGHT to make this point more concretely. We simply can't know a priori that TIGHT is composed of an initial voiceless alveolar stop, a diphthong /ay/ (or a monophthong /a/ in some of your dialects), and a final voiceless alveolar stop. I reiterate: WE CAN'T KNOW THESE FACTS A PRIORI. So, in order to learn this word in the first place, we've got to simply commit these facts to memory. (Note: don't be confused by spelling. There is no /g/ and no /h/ in TIGHT.) Our UR must thus encode this information as part of the lexical entry of the word TIGHT. For this, we use the convention of forward leaning slashes enclosing the relevant phonetic symbols thus: /tayt/. But, of course, we also know that when we say TIGHT, our pronunciation will not be exactly the same as the UR, because the initial /t/ will be predictably aspirated and the final /t/ will be predictably unreleased. These facts are encoded in the PR of the word, which, by convention, we enclose in square brackets thus: [thayt7]. (Note: the h = aspiration, the 7 is standing in here for the little right angle that should be superscripted to show that the final stop is unreleased.)
So, the UR of TIGHT has four phonemes: /t/+ /a/ + /y/ +/t/
The PR of TIGHT has the allophones: [th]+ [a]+ [y]+[t7]
Okay, we've got these two levels of representations. Now what do we do? Can we stop here? Can we go home and watch TV? Not yet, because just having two representations (a UR and an SR) isn't enough. What's missing is the way in which we relate these to each other. This, of course, is the role that phonological rules play.
Phonological rules take underlying representations and map them to their phonetic representations. For the English word TIGHT under discussion here, two phonological rules are relevant in the mapping from underlying to surface forms: 1) a rule accounting for the aspiration of the initial /t/ and 2) a rule accounting for the lack of a release on the final /t/.
Setting aside the issue of the precise formulation of these rules, it is nevertheless crucial that we see that they generalize beyond the single sound /t/. The aspiration rule is true of ALL voiceless stops in English that appear as the first sound in a syllable that has a stressed vowel. And the unrelease rule is true of ALL word-final stops in English (unless we make a point of releasing them for clarity). Here's the big picture with its 3 parts:
1) Underlying representations. These encode what isn't predictable about the way we pronounce words. Underlying representations are comprised of phonemes. They are abstract in that they may but need not correspond to what we actually hear when a word is pronounced. The reason they need not correspond to what we hear is that in order to arrive at the underlying representation of a word we abstract away from the predictable aspects of its pronunciation.Any phonological analysis, then, must explicitly include these three levels because they are mutually dependent. Once we commit to the idea that there is a level of representation that is MORE ABSTRACT than what we hear (the level of the phoneme or UR), we are also committed to expressing the generalizations (or rules) that map this abstract level to the surface phonetic level (our PR).2) Phonological rules. These supply the predictable information to underlying forms in order to map underlying representations to their surface representations.
3) Phonetic representations. These reflect the output of phonological rules. They can also be characterized as the way that words are actually pronounced. (I'm actually fudging a bit here, but for our purposes, this is true enough.) Unlike URs, which are comprised of phonemes, PR's are comprised of allophones of the underlying phonemes.
A bit more of the nitty gritty
So, that's the big picture. Phonology is about two levels: a UR that is comprised of phonemes and a PR that is comprised of allophones of those phonemes. Phonemes are unpredictable. Allophones are predictable realizations of phonemes. How, though, do we go about doing phonological analyses? That is, how can we tell what the phonemes of a language are and what the allophones of those phonemes are? This is where we get into the terrain of the nitty gritty of doing phonology. Your textbook, Language Files, actually does a good job of walking you through how to work out phonological problems.So you should make sure to familiarize yourself thoroughly with Files 41, 4.2, and 4.3 in order to be able to work on problems such as those in File 4.4. Here, I'll list a number of concepts/terms that you want to make sure you understand and control for the exam.
Terms to make sure you understand
predictable vs. unpredictable
Predictablity is crucial to understanding how phonology works. The basic issue is this: some aspects of how we produce the words of a given language are entirely predictable, while others are entirely unpredictable. We've just talked about this above, so I won't repeat all of that here. But, keep in mind the distinction between broad and narrow transcription. Broad transcription notes only the unpredictable aspects of a word. Narrow transcription then adds all the predictable details. Look about to our example of TIGHT to review this.
distinctive vs. non-distinctive
If you understand predictable vs. unpredictable, then you are on your way to understanding distinctive vs. non-distinctive. The main point here is that if a sound is completely predictable, then it is non-distinctive. So, for example, aspirated voiceless stops in English are completely predictable: they show up if a voiceless stop is the first sound in a syllable with a stressed vowel, as in 'pin'. And, we know that we also have unaspirated voiceless stops, as in 'spin'. However, though these two kinds of voiceless stops are present in English, the difference between them is non-distinctive.
To understand this better, it is useful to consider another language that has both aspirated and unaspirated voiceless stops, such as Hindi. In Hindi, in contrast to English, the differences between aspirated and unaspirated stops make differences in meaning. So, Hindi has pairs of words like the following:
kaphi 'meaningful'
kapi 'copy'
The only difference between these words is that one has an aspirated [p] while the other has a plain [p]. Yet, in this case, that difference makes one word distinct from another. So, in Hindi, we say that aspirated voiceless stops are distinctive, or to put it another way, we say that aspiration is distinctive for voiceless stops.
Note also that we thus can't predict a priori whether in Hindi a voiceless stop will be realized with or without aspiration. In this sense it is important to realize that non-distinctiveness goes along with predictability, while distinctiveness goes along with unpredictability.
minimal pair
In the discussion of aspiration in Hindi, I actually used an example of a minimal pair: [kaphi] vs. [kapi]. Minimal pairs are very important to phonology in that they are instrumental in determining whether or not a sound or class of sounds appear in a predictable, hence non-distinctive fashion.
Simply put, a minimal pair is a pair of words that are identical in every way except for one sound. Here are four golden rules for minimal pairs:
1) they must have the same number of sounds
2) they must be identical in every sound except for one
3) the sound that is different must be in the same position in each word
4) the words must have different meanings
Let's keep on the comparison between Hindi and English for a moment. Note that in Hindi, [kaphi] and [kapi] are a good minimal pair. Why. Well, let's go through our golden rules:
1) they must have the same number of sounds
Each has four sounds: [k] [a] [ph] [i] / [k] [a] [p] [i]2) they must be identical in every sound except for one
They are!3) the sound that is different must be in the same position in each word[k] [a] [ph] [i]
[k] [a] [p] [i]
Yep!4) the words must have different meanings[k] [a] [ph] [i]
[k] [a] [p] [i]
And indeed they do!By contrast, note that we NEVER find minimal pairs in English in which the only difference between the two words is that one has a voiceless unaspirated stop and the other has a voiceless aspirated stop. We do find words such as "spin" and "pin", but note that these are not minimal pairs. The word "spin" has four sounds, while the word "pin" only has three. Golden rule one is violated!kaphi 'meaningful'
kapi 'copy'
overlapping vs. complementary distribution
What we're doing when we look for minimal pairs is that we're looking at the distribution of sounds in the words of a language. When we find minimal pairs, in fact, what we've found is what we call overlapping distribution. What do we mean by overlapping? Simply, we mean that two different sounds (in our Hindi example, aspirated and unaspirated "p"), show up in the same context or environment. How do we know it's the same environment? Because everything else about the words in the minimal pair is identical. That's the beauty of minimal pairs. They allow us to focus our attention on the importance of the two sounds that make the pair of words distinct from one another. We say that these sounds are overlapping because each is not restricted to appearing in a place that the other one can't. Instead, they can appear in the same place. To put it visually, if you put the two words in the minimal pair on top of one another, all of the sounds would be identical, except for where the different sounds were overlapping with one another.
The opposite of overlapping distribution is complementary distribution. In a nutshell, here's the key. If two sounds are in complementary distribution in a language, you'll NEVER find minimal pairs in which they make the crucial difference. So, in English, you'll never find minimal pairs like [kat] vs. [khat], where the only difference is the aspiration on the [kh] of the second word. Instead, it turns out that aspirated and unaspirated voiceless stops are in COMPLEMENTARY DISTRIBUTION. This means that they never show up in the same context. Two golden rules of complementary distribution:
1) Minimal pairs are NEVER found for sounds in complementary distribution
2) Sounds that are in complementary distribution never show up in the same context
The reason we use the term complementary is because the sounds basically complement each other with respect to where they can appear. One can appear in context X, while the other can appear in context Y. Remember what I said in class about platooning right-fielders on a baseball team. Where one plays, the other doesn't. That is, they are non-overlapping because they never show up in the same context. Aspirated and unaspirated voiceless stops are like that in English.
contrastive vs. non-contrastive
With one small complication that we'll address below, what we're doing when we establish whether two sounds are in overlapping or non-overlapping distribution in a language is that we are looking at the role that they play in that particular system. In particular, we're looking at whether substituting one sound for the other can make a meaningful difference or not. So, substituting [m] for [n] can make an important difference in English--as in the difference between "team" and "teen". For the most part, when two sounds are overlapping in their distribution, we say that they are in contrast (as [m] and [n] are in English). And when two sounds are in non-overlapping or complementary distribution, we say that they are not contrastive. This takes us all the way back to where we began the section on phonology: to the idea of the phoneme and the allophone.
allophone vs. phoneme
Sometimes linguists are sloppy and they say things like this: "if two sounds are in contrast, then they are different phonemes." The reason that they are being sloppy is that we don't actually "hear" phonemes. Phonemes are abstractions. What we "hear" (for our purposes at least) are allophones. So, what linguists are really saying when they find two different sounds in contrast in a minimal pair is that these two sounds belong to different phonemes. And when linguists find two sounds (like aspirated and unaspirated "p" in English) that are in complementary distribution, they say that these two sounds are allophones of a single phoneme.
If we go back to the big picture, what we see from our continuing saga of Hindi vs. English is that the same sounds play different roles in these two languages. In one, Hindi, aspiration on voiceless stops is contrastive (a.k.a. unpredictable, a.k.a. distinctive), and thus, the voiceless aspirated stops of Hindi are allophones of underlying aspirated stop phonemes. By contrast, in English, aspiration is non-contrastive ((a.k.a. predictable, a.k.a. non-distinctive) on voiceless stops. So in English, aspirated stops are allophones of underlying unaspirated stop phonemes.
free variation
One concept that sometimes confuses people when they are just starting to learn phonology is the concept of free variation. The problem is that free variation is tricky because different sounds can have overlapping distributions but NOT be allophones of different phonemes. This is the case, for example, of stops at the end of words in English. We know from our class discussions and reading in the book that at the ends of words in English, stops can be optionally released or unreleased, as in:
[baet] OR [baet7] 'bat' (remember, I don't have phonetics fonts here)
The problem is that these two LOOK like a minimal pair. That is, it looks like plain [t] and unreleased [t7] are overlapping and thus in contrast and thus allophones of two different phonemes. BUT, remember golden rule (4) about minimal pairs. The two words that make up a minimal pair MUST have different meanings. These two words don't. Ergo, they are not a minimal pair. Once we know that the two stops never make a meaning difference on the basis of whether they are released or not at the end of a word, we know they are not distinctive. At this point what we see is that free variation involves two sounds that can show up in the same context but that aren't distinctive. So, we have a case of overlapping distribution that is NOT contrastive.
This is what makes them tricky. It's because of free variation that we can't simply say that overlapping distribution always means that two sounds are allophones of different phonemes.
Final Notes about phonology
I've tried to review the conceptual aspects of phonology to clarify the big issues that we covered in class. It's up to you to practice practice practice on the phonology problems that are given to you in Language Files. Review File 4.3 especially with regard to solving phonology problems. Also make sure you pay attention to two things:
1) types of rules (listed for you in File 4.2)
2) the importance of natural classes
With respect to natural classes, for example, be aware that aspiration
in English isn't a matter of three rules, one for /p/, one for /t/, and
one for /k/, but rather a single rule that affects a whole class of sounds:
the class of voiceless stops. Other rules may affect classes such as high
vowels, or voiced fricatives, etc... In this regard, it is important
to have your basic articulatory phonetics terms down cold. Because
you can't talk about classes of sounds unless terms like stop, fricative,
liquid, nasal, glide, high vowels, low vowels, lax vowels, round vowels,
and so forth don't mean anything to you.