• No results found

A Problem Concerning the Underlying Form

In document Introduction To Linguistic (Page 67-71)

Regular verbs form the past tense by adding one of the following three suffixes:

[t], [d] or [d]. The choice between these forms is determined by the root.

(83)

[t] [d] [d]

licked [lıkt] bugged [b2gd] mended [mEndd]

squished [skwıSt] leaned [lind] parted [pAtd]

kept [kEpt] buzzed [b2zd] feasted [fistd]

laughed [læft] played [pleıd] batted [bætd]

We ask: what is the source of this difference? Surely, it is possible to say that the past has three different forms and that depending on the verb a different form must be chosen. This, however, misses one important point, namely that the choice of the form is determined solely by the phonological form of the verb and can be motivated by phonological constraints of English. The facts can be summarized as follows.

À [d] is found if the last sound of the verb is voiced but unequal to [d].

Á [t] is found if the last sound of the verb is voiceless but unequal to [t].

 [d] is found if the verb ends in [d] or [t].

We mention here that it is required that the verb is regular. Thus, /run/ and /catch/ are of course not covered by this rule. We may think of the latter as entered in the mental lexicon as unanalysed forms. Thus, rather than seeing /caught/ as a sequence of two forms, namely /catch/ plus some past tense marker, I think of the form entered as a whole, though otherwise functioning in the same way. It is, if you will, an idiom. (Compare this with earlier discussions of the plural formation.)

It seems that the choices can be accounted for solely by applying some general principles. First, notice that in a coda, with the exception of sonorants ([l], [m], [n]), all consonants agree in voicing. An obstruent is a consonant that is either a stop (or an affricate), or a fricative.

Voice Agreement Principle

Adjacent obstruent sequences must either be both [ : +] or both [ : −] at the end of a word.

(The principle is less general than possible: the constraint is valid not only at the end of a word but in any coda.) This goes half way in explaining the choice of the suffix form. It tells us why we see [d] after voiced consonants. But it does not tell us why it is that we get [lind] rather than [lint], because either of them is legitimate according to this principle. Furthermore, we do not know why we find the inserted schwa. The latter can be explained as follows: suppose there was no schwa. Then the present and the past forms would sound alike (/mendd/

would be [mEnd]). Languages try to avoid double consonants (although they never completely manage), and English employs the strategy to insert schwa also in the plural. We find [b2sz] /busses/ (or /buses/), not [b2s:] (/buss/, with a long /s/).

(Another popular strategy is haplology, the dropping of one of the consonants.) It is possible to recruit the Voice Agreement Principle if we assume that there is just a single form not three, and that variants arise only as the result of a repair.

The repair is performed by applying some rules. Various analyses are possible.

Analysis 1. We assume that the underlying form is [d]. There is a rule that de-voices [d] right after a voiceless obstruent. There is a second rule which inserts a schwa right before [d]. For the purpose of the definition of the rules, two conson-sants are called similar if they differ at most in the voicing feature (for example, [t] is similar to both [t] and [d], but nothing else).

" +voice

The symbol ∅ denotes the empty string. The first rule is actually two rules in our feature system. I reproduce here the reproper formulation:

(85)

The second rule effectively says that it is legal to insert schwa anywhere between similar consonants. Since we have two rules, there is a choice as to which one shall be applied first. We shall first schedule (84a) before (84b). This means that

the rule (84a) is applied to the original form F0, giving us an output form F1, and then we apply rule (84b) to get F2. Each rule applies only once, so the output is F2. This gives the following result:

(86)

root /b2gd/ /lıkd/ /mEndd/ /statd/

(84b) /b2gd/ /lıkd/ /mEndd/ /statd/

(84a) /b2gd/ /lıkt/ /mEndd/ /statd/

Notice that when we say that a rule does not apply this does not mean that no out-put is generated. It means that the form is left unchanged by the rule. Textbooks sometimes indicate this by a hyphen as if to say that there is not output. But there is an output, it just is the same as the input. So I do not follow the practice. You may figure out for yourselves in which cases this has happened! Notice, too, that sometimes rules do apply and do not change anything.

Now, suppose we had instead scheduled (84a) before (84b). Then this would be the outcome:

(87)

root /b2gd/ /lıkd/ /mEndd/ /statd/

(84a) /b2gd/ /lıkt/ /mEndd/ /statt/

(84b) /b2gd/ /lıkt/ /mEndd/ /statt/

If the last consonant is /t/, the rule (84a) would first assimilate the past tense marker, and we get the suffix /t/, contrary to fact. Thus, the order in which the rules apply is relevant here. There is, however, nothing intrinsic in the system of the rules that tells us in which order they have to apply. This has to be stipulated.

In the present analysis we see that the second ordering always gives us an output form, but sometimes the wrong one.

Analysis 2. The underlying form is assumed to be [t]. In place of (84a) there now is a rule that voices [t] right after a voiced obstruent or a vowel. There is a second rule which inserts a schwa right before [d] or [t].

" −voice +obstruent

#

"

voice : +

:stop

#

/[+voice] # (88a)

∅ →[] /C C

(88b)

(C and Csimilar)

(89)

root /b2gt/ /lıkt/ /mEndt/ /statt/

(88b) /b2gt/ /lıkt/ /mEndt/ /statt/

(88a) /b2gd/ /lıkt/ /mEndd/ /statd/

If we schedule (88a) before (88b), this will be the outcome:

(90)

root /b2gt/ /lıkt/ /mEndt/ /statt/

(88a) /b2gd/ /lıkt/ /mEndd/ /statt/

(88b) /b2gd/ /lıkt/ /mEndd/ /statt/

Once again, we see that schwa insertion must take place first.

Analysis 3. The underlying form is [d]. There is a rule that devoices [d] right after a voiceless obstruent. There is a second rule which deletes schwa in between dissimilar consonants.

" +voice +obstruent

#

→" −voice +obstruent

#

/[−voice] # (91a)

[] → ∅ /C C

(91b)

(C and C dissimilar)

(92)

root /b2gd/ /lıkd/ /mEndd/ /statd/

(91b) /b2gd/ /lıkd/ /mEndd/ /statd/

(91a) /b2gd/ /lıkt/ /mEndd/ /statd/

If we schedule (91a) before (91b), this would be the outcome:

(93)

root /b2gd/ /lıkd/ /mEndd/ /statd/

(91a) /b2gd/ /lıkd/ /mEndd/ /statd/

(91b) /b2gd/ /lıkt/ /mEndd/ /statt/

We conclude that schwa deletion must preceded voice assimilation.

In principle, there are many more analyses. We can assume the underlying form to be anything we like (say, even [ð] or [Z]). However, one clearly feels that such a proposal would be much inferior to any of the above. But why?

The principal difference between them is solely the extent to which the rules that transform them can be motivated language internally as well as language ex-ternally. And this is also the criterion that will make us choose one analysis over

the others. Let’s look carefully. First, let us go back to the Voice Agreement Prin-ciple. It says only that adjacent obstruents agree in voicing, it does not claim that obstruents must agree with the preceding vowel, since we do actually find forms like [kæt]. Analysis 2 incorporates the wrong version of the Voice Agreement Principle. Rule (88a) repairs some of the forms without need, while Analysis 1 repairs the forms if and only if they do not conform to the Voice Agreement Principle. Now look at Analysis 3: it does not conflict with the Voice Agree-ment Principle. However, it proposes to eliminate schwa in certain forms such as ["lıked]. There is however no reason why this form is bad. There are words in English such as /wicked/ that have such a sequence. So, it repairs forms that are actually well-formed. Thus the best analysis is the first, and it proposes that the underyling fom is [d].

Let us summarize: an analysis is preferred over another if it proposes laws of change that are widely attested (schwa insertion is one of them, and final devoicing is another). Also, an analysis is dispreferred if its rules change representations that are actually well-formed. Thus, rules of the kind discussed here are seen as repair strategies that explain why a form sometimes does not appear in the way expected.

What we are looking at here is, by the way, the mapping from deep phonological form to surface phonological form. The last bit of evidence that makes us go for Analysis 1 is the following principle:

Not-Too-Similar Principle

No English word contains a sequence of subsequent similar obstru-ents.

Namely, Rule (84a) devoices an obstruent in coda if the preceding consonant is voiceless. And in that case, the Voice Agreement Principle is violated. After the rule has applied, the offending part is gone. Rule (84b) applies if there are two subsequent similar consonants, precisely when the Not-Too-Similar Principle is violated. After application of the rule the offending part is gone.

In document Introduction To Linguistic (Page 67-71)