Does lack of a common (morpho-)phonological alternation make a word a lexical exception?

Question

I am trying to understand the following passages from The Sound Pattern of English, by Noam Chomsky and Morris Halle.

Convention 1: : Every segment of a lexical matrix is automatically marked [+n] for every rule n. [...]
Convention 2 : Every segment of a lexical matrix μ is marked [αK] for each category [αK] to which μ belongs.

Thus, in particular, if a formative belongs to the lexical category [-n], each of its segments will be marked [-n] by Convention 2, after having automatically been marked [+n] by Convention 1. Thus every time a formative is an exception to a rule, there is a certain "cost" associated with this fact, namely, a certain category assigment must be given the lexical entry. But an item that does undergo a rule need not be specially marked. Thus only exceptions to a rule contribute to the complexity of the grammar in this connection.

(page 173—it's page 93 of the PDF I linked to)

Alongside of the partially systematic class of exceptions to rule (6), we also find purely idiosyncratic exceptions. For example, consider the rule that makes vowels non-tense before certain affixes (e.g. compare serene and serenity, obscene and obscenity). There are exceptions to this rule (e.g. obese-obesity, in most dialects) which simply must be categorized as such in the lexicon, these lexical features becoming segmental features by Convention 2. Each such example contributes to the complexity of the grammar, but there is obviously no question of rejecting the rule. Doing so would amount to treating each item as an exception, in the manner of item-and-arrangement grammars (see Chapter Three, Section 16), and there is surely no point to such a decision.

(page 174/page 94 in the PDF)

Chapter 3 section 16 compares such situations to regular and irregular verbs; irregular verbs are exceptions, but the fact that exceptions exist doesn’t negate the rules that apply to regular cases. It outlines three ways of dealing with phonological irregularities: 1) abandoning general rules and lexically specifying the irregular feature for each particular item; 2) classifying words as (+) regular and (-) regular; 3) reinterpreting the underlying form to make the rule apply (page 147/page 80 in the PDF). (This last option is the one Chomsky and Halle prefer; this discussion is a leadup to their analysis of forms like “ellipse” as having underlying segmental “e” at the end.)

Anyway, I guess I don’t understand the motivation for the analysis made in the quoted material. In what sense should we consider words like “obesity” an exception? Obviously, it is one in terms of frequency (there are more items in the lexicon that alternate than ones that don’t). But Chomsky and Halle seem to be saying that it is mentally classified as an exception by native speakers. If this is the case, I don’t understand why this extra complexity in the lexical representation of the word originated historically (the word used to be pronounced with /ɛ/, and over time came to be pronounced with /i/). My default intuition would be to consider the words with historical phonological alternations as marked, in the sense that they're harder to process psychologically (for example, I have to consciously think to apply them sometimes, and avoid saying stuff like /səˈriːnɪti/), and the non-alternating forms like "obesity" as unmarked. I would not reject the rule entirely, but I don't understand why the rule is taken to be the default situation.

Is Chomsky and Halle's viewpoint current in phonology, or do most people now have a different interpretation? I know there are other approaches to phonology, such as Optimality Theory, but I am only an amateur and not familar with them in any depth.

Answer 1

In their analysis, there is a rule of Trisyllabic Laxing which should apply to obesity, but does not. It is therefore a lexical exception. There are various other things that could be said about their theory of exceptions, but that's really all there is to it. Frequency isn't what determines exceptionality in their theory. I think the issue is that you understand "exception" to mean something else, maybe "not the most frequent".

Theoretical practicioners of generative phonology generally hold to that meaning of "exception", though the formal interpretation varies (e.g. how do you have exceptions to constraints -- what does that mean?). Other phonologists (functionalists for example) may have other viewpoints.

[EDIT]

There is a technical distinction between an exception and a diacritic. An exception is where a morpheme is marked [-rule n], but morphemes are ordinarily marked [+rule n]. In this situation, that analysis expresses the generalization that most words undergo the rule, but a few don't, and the latter are representationally more complex due to bearing the marked value of the rule feature. However, there can be instances where only a few words undergo a rule, meaning that most words in the language are apparently exceptions. In that case, an alternative device is available, the diacritic, where a rule only applies to segments that are [+D], and only select morphemes bear the marked specification [+D]. So the theory can express both notions, it just uses different mechanisms.

Recall that rules are rules, period. So if every rule requires that the formatives that undergo the rule be specially marked, then a huge number of formatives of English must be specially marked to undergo aspiration and flapping. It would be a kind of absurd result to say that all words in English have the special permission feature.

For the specific case of Trisyllabic Laxing and "obesity", the vast majority of phonologists would reject the SPE analysis. So while we agree on what it means to be an exception, most phonologists would now reject the earlier assumption that serene and serenity are derivationally related via the root /serēn/ and the specific TSL rule given in SPE. Lexical Phonology ameliorated the problem with TSL a little bit, but it really comes down to questions of what kind of evidence would force us to derive the supposed TSL-related word pairs in a synchronic grammar.

Answer 2

I can't claim to understand the passages you quote from SPE. However, I do think I have enough sympathy with their approach to help you decide whether you want to go to the effort of treading their thicket of formalisms. (I don't think you want to.)

I think you capture the overall structure of the argument very well in this part of your question about how to describe exceptional forms:

1) abandoning general rules and lexically specifying the irregular feature for each particular item; 2) classifying words as (+) regular and (-) regular; 3) reinterpreting the underlying form to make the rule apply.

Using your numbering, I'll restate what I take to be the C&H argument but (for clarity) in a rather biased way: We understand that the abstract lexical forms we propose for "ellipse", "giraffe", and so on, seem absurd. But we will show that a generative description must adopt one of the alternatives 1, 2, or 3, and though 3 is absurd, 1 and 2 are even worse. Therefore, we must adopt 3.

Now here are some notes intended to clarify:

A. Since a generative grammar is supposed to capture the underlying regularities of a language, and since the words under discussion are admittedly exceptional, why bother with them at all? one might ask. Well, the "generative" in "generative grammar" means explicit. We must be entirely explicit about every last detail, or else we're just not doing generative grammar.

B. In what sense are 1 and 2 "worse" than 3? They are worse because they require more symbols to write out in fully explicit form. This is how alternative proposals about grammar are compared. You write them out, you count the symbols in each, and the proposal with fewest symbols wins. In SPE, this is called "the evaluation measure"; it is a minor modification of Halle's earlier "simplicity metric" (see "Phonology in Generative Grammar" from 1962).

C. Why can't 1 and 2 be formulated in a more compact fashion, marking entire words or (at least) entire morphemes as to whether they are exceptional to some phonological rule? There are arguments due to Stephen Anderson that individual segments must be marked for exceptionality to Turkish Backness Harmony. (I don't find those arguments persuasive, personally, but anyway ...) And if it's good for Turkish, it's good for English, of course.