Speaker's Mind → Speaker's Mouth → Listener's Ear → Listener's Mind
We will write rules to connect the Mind and Mouth.
The amazing discovery is that people systematically ignore certain properties of sounds. They perceive two different sounds as the same sound. We call the stored versions of speech sounds phonemes. Thus phonemes are the phonetic alphabet of the mind. That is, phonemes are how we mentally represent speech; how we store the sounds of words in our memory.
Though the phonetic alphabet is universal (we can write down the speech sounds actually uttered in any language), the phonemic alphabet varies from language to language. For example, English has no memorized front rounded vowels like German or French, and French has no [θ]. This leads to seemming contradictions when we consider both actual productions of speech sounds as well as their memorized representations. English has no memorized nasal vowels, but English speakers do make nasalized vowels when vowels and nasal consonants come together in speech. The changes between memory and pronuciation are what we will be discovering in this section of the course
How do we find out what's in someone's mind?
How do we figure out how people store the sounds of words in their memories?
One trick that we can use is to look for minimal pairs of words. A minimal pair is a pair of words that have different meanings and which differ in only one sound. Since the difference between the two sounds is meaningful, the words must be stored differently in memory. Since the words differ in only one sound, this difference must be stored in memory. Thus the difference in sounds is significant, and so the two sounds must both be phonemes.
Here is an example from English:
These two words aredifferent words of English. But they differ only in their initial sound. Therefore, the [s]/[z] difference is significant for English speakers. Therefore both [s] and [z] are stored in the memory. Thus, [s] and [z] are part of the English mental alphabet. We notate elements in the memory by putting them in-between slashes / /. In this case /s/ and /z/ are part of an English speaker's alphabet for memorizing words.
Another example from English:
These three words are all distinct words of English. Therefore, the speech sounds (in the mouth) [m], [n] and [ŋ] are all significant to the mind. And therefore, English includes the phonemes /m/, /n/ and /ŋ/.
Sometimes it isn't possible to find minimal pairs for all words. But speakers can also tell when a contrast would yield a distinct possible word, even if this is not an actual word.
The phonetic context (or frame) [b_t] can be used to find minimal pairs for many English vowels:
This minimal set establishes all these vowels as mentally distinct, and therefore phonemic.
Near minimal cases can be found for the other two vowels:
And additional minimal pairs can be constructed to justify all 11 vowels, pair by pair as necessary. Since other minimal pairs can be found ([lʊk] "look" versus [luk] "Luke"), ([kɑt] "cot" versus [kɔt] "caught"), these vowels are also mentally significant, and are therefore phonemes.
We will use the articulatory phonetic descriptions to write pronunciation rules.
We will add additional features as necessary.
The fact that speakers have a mental representation of what they say, and that this can be different from what they actually do when they speak, shows us that speakers do not memorize every aspect of speech sound production. Only the essential (contrastive, phonemic) features are stored in memory. Other features (specifics of pronunciation) are added during speech planning and production. Predictable information about speech is not memorized. Predictable features are added by rules of pronunciation (phonological rules).
English speakers pronounce vowels either with the velum closed (oral) or with the velum open (nasal). By careful listening or experimental investigation, we can determine that the velum is open during the entire production of the word "man":
In contrast, speakers do not have the velum open at all in the production of the word "bat":
Most important, however, is that English speakers perceive that both "man" and "bat" have the same vowel. That is, English speakers are ignoring the difference in nasality between the two words. English speakers feel that this difference in nasality is unimportant for recognizing a word in their memory. We can understand this behavior through understanding that speakers memorize vowels without the feature [nasal]. English speakers believe that there are no nasal vowels in English, at least for the purpose of memorizing words.
The reason for this is that in English nasality in vowels is predictable. In English, nasal vowels only occur before nasal consonants. Everywhere else English speakers use oral vowels. Therefore [nasal] is predictable for English vowels, and is governed by a rule of pronunciation:
We can contrast the situations in the Memory and in the Mouth:
| Memory: | /æ/ | ||
| Mouth: | [æ̃] only before nasal consonants |
[æ] Everywhere else (Elsewhere) |
We call the mental representation a phoneme, and we call the distinct pronunciations allophones.
The predictable aspects of pronunciation (here [nasal] in vowels) are added by the rules in the phonology of the language. The rules of pronunciation determine the variants in speech sounds. This particular rule makes one sound (the vowel) more similar to an adjacent sound (the following nasal consonant), by making the vowel [nasal]. Rules that make sounds more similar are called assimilation rules. Rules that make sounds less similar are called dissimilation rules. Assimilation rules are much more common than dissimilation rules.
Which features are predictable varies from language to language. In French speakers must memorize [+nasal] for vowels because in French this is important for the meaning of the word. That is, French has minimal pairs for nasality in vowels.
Within a single sound some of the aspects of speech sound production (features) may be predictable from the other features. Here are some examples from English:
These particular rules are rules of English. Other languages may or may not have these rules. So, for instance French has frount rounded vowels, but no high back unround vowels. Russian has the reverse: high back unround vowels but no front round vowels. Turkish has both front round vowels and high back unround vowels.
All languages impose certain restrictions on the sequences of sounds in the language. Some languages like to alternate consonants and vowels. These language do not allow sequences of consonants nor sequences of vowels.
There are two possible responses a language can make to an unwanted sequence. One is to change one of the sounds, through a rule. This is what we observed with English nasal vowels. Oral vowels are not allowed to be followed by nasal consonants, so the vowel is changed to be nasal.
The other possible response is to simply ban the sequence from words as they are stored in memory. In English there is a general ban on words beginning with *[tl] and *[dl], even though words starting with [pl], [bl], [kl] and [gl] are fine. But there is no general rule to repair these bad sequences.
We have seen the case of one mental sound having two mouth pronunciations, repeated here:
| Memory: | /æ/ | ||
| Mouth: | [æ̃] only before nasal consonants |
[æ] Everywhere else (Elsewhere) |
It is also possible to have the situation where sound that are memorized differently are nevertheless pronounced identically under certain circumstances. Consider the pronuncation of the vowels in these two words:
But since both of these words involve the same morpheme, meaning "telegraph", this morpheme must have the same memorized representation, namely,
Therefore the changes in pronunciation are insignificant for memory here, and must be due to a rule of pronunciation. The rules is very simple, unstressed vowels reduce to schwa in English.
This means that in this case we have two different sounds as far as the mind goes, and yet when unstressed, only one pronunciation:
| Mind: | /æ/ | /ɛ/ | |
| Mouth: | [ə] when unstressed |
The plural marker in English has several different pronuncations:
The choice of zero or [ə̃n] is handled in the Morphology, because it is not predictable from the speech sounds. The choice of [s], [z] or [əz] is handled in the Phonology, because it is predictable from the speech sounds.
So we need to figure out what conditions the appearance of the various pronunciations of the plural. Where do we get which pronunciations?
Pronunciations: [s] [z] [əz]
Examples: cats dads bushes
beliefs sleeves roses
cups cabs kisses
tacks tags churches
paths wreathes judges
dams
cans
songs
fears
peels
bees
bays
boos
bows
boys
We want to predict the pronunciations, so we need to pick one of the pronunciations to be stored in memory. We should pick the least predictable one to store in memory, here [z]. Therefore, the memorized representation for the plural morpheme is /z/.
Now we need to write rules to get the other pronunciations.
What is similar about all the stems in the first column, the ones that take [s] in the plural? They are end in voiceless sounds! So the rule for these must be:
/z/ becomes [s] when it comes after a voiceless sound
But we can do better than this. What is the difference between /z/ and [s]? /z/ is [voiced] and [s] is [voiceless]. So we can write the rule as:
/z/ becomes [voiceless] when it comes after a [voiceless] sound /z/ → [voiceless] / [voiceless] _
Thus, this is a rule of assimilation.
What is the characteristic shared by the words in the third column, where they say [əz] for the plural? All of these words end in sibilant (see FRH p. 300) sounds. So we can write the rule:
Insert [ə] in between a sibilant and /z/
Again, we can ask if we can do better. Is /z/ sibilant? Yes! So we can write the rule:
Insert [ə] inbetween two sibilants ∅ → [ə] / [sibilant] __ [sibilant]
The use of the zero here for "no sound" or "nothing" is similar to the concept of zero morphemes. For insertion, ∅ → something; for deletion, something → ∅.
Consider the distribution of [r] and [l] in Korean in the following words. Some additional details of Korean pronunciation which are not indicated in the book are given here.
rubi 'ruby' kiri 'road' saram 'person' irɯmi 'name' radio 'radio' mul 'water' pal 'big' səul 'Seoul' ilgop 'seven' ipalsa 'barber'
Are [r] and [l] allophones of one or two phonemes? (That is, are these sounds stored differently in the memory of Korean speakers?) State your reasons, and give the rule to derive the surface phones if you conclude that they are allophonic.
There are none. Therefore, probably allophones of one phoneme. Two sounds in the mouth, one sound in the mind.
There are none.
(Zoom in on the things of interest. Get rid of extraneous information.) Work from the things closest to the sounds of interest.
| [r] | [l] | sounds immediately before | sounds immediately after | sounds immediately before | sounds immediately after |
| # (beginning of word), i, a | u, i, a, ɯ | u, a, i | # (end of word), k, s |
Does the "before" environment predict which sound? NO! The sounds [i] and [a] occur before both [r] and [l]. Therefore we cannot predict [r] versus [l] on the basis of the immediately preceding sound. In the case of [i] or [a] both [r] and [l] can occur.
Does the "after" environment predict [r] versus [l]? YES! The "after" environments do not overlap!
What is the difference between the two environments? [r] occurs only when a vowel follows; [l] occurs when either a consonant follows or when nothing follows (i.e. at the end of the word).
Which is the simpler one to predict? [r], "vowel" is simpler than "consonant or nothing". Therefore the [r] is predictable, and /l/ (the unpredictable one) must be the phoneme.
Mind: /l/ Mouth: [r] / _ [vowel] [l] / Elsewhere
Koreans pronounce /l/ as [r] when a vowel immediately follows. /l/ → [r] / __ [vowel]
Can we do better? What's the difference between [l] and [r]? [l] is [lateral]; [r] is [alveolar].
[lateral] → [alveolar] / __ [vowel]
Consider the following data from Finnish:
a. [ku:zi] 'six' b. [kudot] 'failures' c. [kate] 'cover' d. [katot] 'roofs' e. [kade] 'envious' f. [ku:si] 'six' g. [li:sa] 'Lisa' h. [maton] 'of a worm' i. [madon] 'of a rug' j. [ratas] 'wheel' k. [li:za] 'Lisa' l. [radan] 'of a track'
a. [ku:zi] and f. [ku:si] differ minimally in pronunciation in the right way: a. has [z] and f. has [s].
BUT they both have the same meaning: 'six'. Because the difference in pronunciation does NOT cause a change in meaning, this is not a minimal pair. Therefore these two words do not establish a mental memorized distinction between [s] and [z].
We already found one: a. [ku:zi] or f. [ku:si] 'six'. And there's another: g. [li:sa] or k. [li:za] 'Lisa'.
| [s] | [z] | sounds immediately before | sounds immediately after | sounds immediately before | sounds immediately after |
| u:, i:, a | i, a, # (end of word) | u:, i: | i, a |
Can we predict [s] versus [z] from the "before" environment? NO! The "before" environments overlap, both [s] and [z] can occur when the immediately preceding sound is [u:] or [i:]. Therefore, knowing the immediately preceding sound won't predict which of [s] or [z] we get.
Can we predict [s] versus [z] from the "after" environment? NO! The "after" environments overlap, both [s] and [z] can occur when the immediately following sound is [i] or [a]. Therefore, knowing the immediately following sound won't predict which of [s] or [z] we get.
Moreover, in the cases of the false minimal pairs,
a. [ku:zi] and f. [ku:si] 'six'
g. [li:sa] and k. [li:za] 'Lisa'
the environments are exactly the same.
(They look like minimal pairs, but they aren't because the
change in pronunciation doesn't cause a change in meaning.)
Since the environments are exactly the same in these
alternative pronunciations of 'Lisa' and 'six', nothing in
the speech sounds themselves can predict which one of [s] or
[z] will be used.
This case is not phonologically predictable.
It might be the case that looking at social or dialect
factors would tell us how Finnish speakers choose between
[s] and [z].
At this point all we know is that they don't make the choice
between [s] and [z] based on the surrounding sounds.
However, we can notice that [s] occurs in more environments than [z] does. For example, j. [ratas] 'whell' has an [s] at the end of the word. We don't see any words ending in [z]. So it would be best to memorize the sound as /s/, and then sometimes pronounce /s/ as [z]. For this case we just can't predict exactly when /s/ is pronounced as [z].
Mind: /s/
Mouth: [s] [z]
Overlapping environments--unpredictable
/s/ becomes [z] unpredictably /s/ → [z] unpredictably /s/ → [voiced] unpredictably
[s] and [z] are in free variation
There are minimal pairs:
[madon] and [maton] differ minimally in their pronounciation, AND this change in pronunciation goes along with a change in meaning: [madon] means 'of a worm' and [maton] means 'of a rug'. Because we found minimal pairs, both sounds are phonemes.
There are no alternations for [t] and [d].
Because they are both phonemes (both used in memory) and there are no alternations, we do not need to calculate the environment, because there is nothing to predict.
In Finnish, /t/ is pronounced as [t] and /d/ is pronounced as [d].
There is nothing to predict here. Both /t/ and /d/ are phonemes, and they maintain separate pronunciations (there are no alternations).
Mind: /t/ /d/ Mouth: [t] [d]
No rule here, they're both phonemes with no alternations.
Consider the phonetic forms of Hebrew words:
[v]-[b] bika 'lamented' mugbal 'limited' ʃavar 'broke masculine' ʃavra 'broke feminine' ʔikev 'delayed' bara 'created' [f]-[p] litef 'stroked' sefer 'book' sataf 'washed' para 'cow' mitpaxat 'handkerchief' haʔalpim 'the Alps'
Assume that these words and their phonetic sequences are representative of what may occur in Hebrew. In your answers below, consider classes of sounds rather than individual sounds.
There are no minimal pairs for [b] versus [v].
The morpheme for 'broke' shows different pronunciations for the masculine and the feminine: [ʃavar] 'broke masculine' and [ʃavra] 'broke feminine'. But these two forms do not illustrate a [b]-[v] change. Therefore they don't immediately let us figure out what causes either [b] or [v] to occur. So we have to go on to listing the environments.
| [b] | [v] | sounds immediately before | sounds immediately after | sounds immediately before | sounds immediately after |
| # (beginning of word), g | i, a | a, e | a, r, # (end of word) |
The "after" environments overlap. The sound [a] can come immediately after both [v] and [b]. Therefore we cannot predict which of [v] or [b] would occur just given the information about what sound comes immediately after.
The "before" environments do not overlap. Therefore the before environment is sufficient to predict which of [b] or [v] will occur. Therefore for speakers of Hebrew, [b] and [v] are stored as one sound.
But which sound do you store in memory? We figure this out by asking which environment is harder to predict. In this data, [v] only occurs after vowels. [b] never occurs after vowels. [b] only occurs after consonants, or at the beginning of the word. The environment "after vowels" is simpler than "after consonants or at the beginning of the word", so the environment for [b] is harder to predict. Therefore, for Hebrew speakers [b] and [v] are different pronunciations of the memorized sound /b/.
Mind: /b/ Mouth: [b] / Elsewhere [v] / [vowel] _
/b/ becomes [v] when there is a vowel immediately before it /b/ → [v] / [vowel] __ [voiced oral bilabial stop] → [fricative] / [vowel] __
What do we want to compare? [b] with [p] or [b] with [f]? We want to compare stops with stops and fricatives with fricatives. Therefore, we want to compare [b] with [p]. Therefore, we are considering whether the rule should be extended to include:
/p/ --> [f] / [-consonantal] __
How would we know? The rule says that we get [f] if and only if it is preceded by a vowel. Therefore, there can be two kinds of counterexamples:
So we look. All [f]'s are immediately preceded by vowels. No [p] is immediately preceded by a vowel. Therefore the rule works here too.
Now we need to write one rule that covers both cases. What characteristics do [p] and [b] share? They're both labial stops. Therefore the rule must be:
A labial oral stop becomes a fricative when a vowel comes immediately before it [labial oral stop] → [fricative] / [vowel] __
What determines the choice of [b, p] versus [v, f]? The sound immediately before. What sound is immediately before the blank? [d]. What kind of sound is [d]? A consonant. So fricatives cannot appear here, because fricatives only appear when there is a vowel immediately before them.
So "(1) [b] but not [v] could occur in the empty slot" is the only correct statement.
No vowel preceding, so fricatives cannot occur in this position.
So "(2) [p] but not [f] could occur in the empty slot" is the only correct statement.
laval surva labal palar falu razif
If these words actually occured in Hebrew would they:
Let's check to see if we need to rethink anything.
The new data introduces a minimal pair: [laval] versus [labal]. This would lead us to conclude that the distinction between [b] and [v] is significant, because it goes along with a change in meaning. Then we would conclude that Hebrew speakers memorize both [b] and [v], giving two phonemes /b/ and /v/. This is clearly a different conclusion than what we came up with, so the right answer is (1), we would be forced to revise our conclusions.