Here again is the doctrine of "Lexical Atomism."  I do not know of anyone who has articulated this picture in its entirety, though it is (rightly) associated with the works of Davidson, Hintikka, and Montague.  It is a surpassingly lovely picture, and an important one too.

(A) [A strong form of the compositionality assumption]  The meaning of a sentence is entirely determined by the atomic meanings of that sentence’s individual component words (really morphemes) together with the syntactic rules  and correlatively the semantic rules  according to which the relevant morphemes are combined and arranged into well-formed sentences.

(B) The atomic morpheme-meanings can be given by individual entries in a definitive dictionary of some sort  preferably clauses in a truth-definition either extensional or intensional.  These meanings are fixed, by tacit convention, prior to any syntactical combination of the morphemes into longer constructions.

(C) There are only finitely and manageably many morpheme-meanings underlying any single natural language, or for that matter underlying the totality of all extant natural languages, primarily because any single morpheme-meaning in a natural language must be learned individually by any human speaker of that language in a determinate chunk of real time.

(D) Lexical ambiguity in a natural language is neatly limited.  Of course a word may have more than one sense, but the several senses it has may be crisply captured in a short, discrete list and treated by linguistic semantics as brutely homonymous or equivocal (otherwise its separate uses could not have been learned in so short a time by speakers).

(E) In addition to its ordinary literal meaning or meanings, a word might have some figurative  particularly metaphorical  uses as well.  But although both philosophers and linguists rightly count it a great mystery how nonliteral meaning is derived from real meaning, figurative meaning is not only derivative but nearly negligible from the viewpoint of current  active linguistic theory.  Semantics proper  studies literal meaning, though we would all admit under pressure that philosophy of language taken more broadly  comprehends the question of how nonliteral meanings are generated from literal ones (and how they are related to literal truth-conditions).  Metaphor especially is a surd in language that will need very special and arcane explanation once the more straightforward theory of literal meaning or "literal propositional content" has been straightened out.

(F) A word is either univocal or equivocal (ambiguous, homonymous), though perhaps most words happen to be ambiguous.

(G) A sentence (of a natural language) has at least one determinate logical form.  Actually, due to manageable lexical ambiguity as well as to equally manageable syntactic or structural ambiguity, almost any sentence has a multiple range of individual "readings" corresponding to distinct "logical forms" ascribable to that sentence in different contexts, but that range is not often large.

(H) The meaning of a sentence consists in the literal truth-condition(s) expressed by way of its logical form(s); truth-conditions are determined compositionally by literal morpheme-meanings and syntactic-semantic rules (cf. (A) and (B)).

(J) Atomic morpheme-meanings are a matter of important if conventional relations between individual words or morphemes and (predominantly) nonlinguistic items in the world.  The paradigms here are reference or denotation for singular terms and extensions or satisfaction-classes for predicates.

    Lovely and important, as I said.  But, alas, Lexical Atomism is also inaccurate.  It goes wrong by neglecting the analogy mechanisms at work in lexical semantics, and in particular the volatile polysemy induced by those mechanisms.  But let's see what survives and what doesn't.

    Thesis (A) still comes out true, so far as I can see; even if every word in a sentence has "differentiated" all over creation, the meaning of the whole sentence is still determined by the (however analogical) meanings of the words together with their mode of compounding.  But this affirmation of (A) must be qualified by a denial of (B).  Only the most common atomic morpheme-meanings can be listed in a dictionary; other perfectly possible meanings will always be left out, and as Kittay says, the listings will probably be only useful abstractions or prototypes.  And most ordinary meanings cannot be fixed prior to syntactical combination of the morphemes into longer constructions.  (Though if there are original, "absolute" meanings, as there must once have been on pain of regress, those can be so fixed.)
    (C) becomes equivocal.  If polysemy is infinite, then a fortiori there are not only  finitely many morpheme-meanings underlying a natural language.  On the other hand, the polysemous meanings are derivative, being projected in regular and systematic ways by way of the analogy mechanisms, and so learnability is saved; the mechanisms must work upon a finite base, for all the familiar Davidsonian reasons.  (This constitutes a second argument for the existence of absolute meanings.)
    (D) and (E) are just false, however nice a world this would be if they were true.
    (F) is true in letter, but false as intended.  Univocality is extremely rare if it exists at all, and the differences between the different types of equivocity are more interesting and important than is the difference between any of them and univocity.  (G) needs amending: the range of readings will be large, and the use of truth-conditional methods in the actual study of texts may be seriously hampered by the ubiquity of the analogy phenomena, though possibly the "readings" associated with an paronymous sentence could abstract away from word meanings entirely if all that is under investigation is logical form.  (H), of course, omits mention of the analogy mechanisms, and so is false as it stands, since the analogy mechanisms almost invariably figure in what we would unthinkingly have called the "literal" meanings of sentences.
    (J), finally, is immune to the analogy mechanisms themselves, so long as we withhold our assent to Ross’ particular explication of dominance and we do not go off the structuralist deep end and deny that meaning has anything to do with reference.
    Thus there is not a lot left of Lexical Atomism, even if truth-conditional semantics itself is still aloft.  I believe contra Kittay that the two are largely independent, even though most truth-conditional theorists have tacitly accepted Atomism because they did not particularly care about lexical semantics.  But the apparent falsity of (B) and (E), at least, should make semanticists take notice.