Cristofaro, Sonia (2008). A constructionist approach to complementation: Evidence from Ancient Greek. Linguistics 46.3: 571–606. DOI: 10.1515/LING.2008.019

Cristofaro argues against the notion that the meaning of a sentence with complement clause can be view as arising from the lexical entry of verb of the matrix clause by itself. Rather, complement clauses provide their own contribution to the overall meaning of the whole. Ancient Greek allows for three types of complement clauses. These are infinitive complementation, participle complementation, and complementizer +indicative clause complementation.

(1) Περσέων μέν νυν οἱ λόγιοι Φοίνικας αἰτίους φασὶ γενέσθαι τῆς διαφορῆς
‘The Persian learned men say that the Phoenicians were the cause of the dispute’ (Herodotus, Histories 1.1.1)

(2) ἄρξομαι δὲ ἀπὸ τῆς ἰατρικῆς λέγων
‘I will start my talk with [lit.‘I will start talking from’] medicine’ (Plato, Symposium 186b)

(3) εἶπον ὅτι νῆες ἐκεῖναι ἐπιπλέουσιν
“They said that ships were sailing against them.” (Thucydides, Peloponnesian War 1.51.2.4)

Relevant for Cristofaro’s purposes, many complement taking predicates also take NP complements as well. For example, when the perception predicate in (4) takes an NP complement, the sentence describes sensory perception rather than mental perception.

(4) ἤ που ἔτι ζώει καὶ ὁρᾷ φάος ἠελίοιο
‘Whether he still lives somewhere and sees the light of the sun’ (Homer, Odyssey 4.833)

This is equally true in English and Greek, shown in the English translation of (4) with example (5):

(5) a. [He sees the light of the sun.
b. I see what you are saying.

Such phenomenon is easily accounted for in ruled based approaches to complementation, since there is a one-to-one correspondence between the syntactic alternation of complement/NP and the semantic alternation between sensory/mental stimuli.

The problem with this approach is, according to Cristofaro, that the one-to-one correlation is not quite so clear cut. Consider the utterance predicate class in Ancient Greek in Table #1.

Table #1
Complement-taking predicate class	Complementation Pattern	Semantics
Utterance	Infinitive clause	Report of statement to which the speaker is not committed
		Report of statement w/subject co-referent
		report of command/suggestion
	ὡς+Indicative	Report of statement to which the speaker is not committed
	ὅτι, ὡς+Indicative	Report of statement to which the speaker is committed
	NP	Production of Speech

With such predicates, two problems appear for approaches that assume lexical rules. First of all, the semantics of the construction can no longer be derived from the subcategorization of the verb itself. Infinitive clauses and complements introduced by ὅτι+indicative constrain the level of commitment of the speaker. In contrast, complements introduced by ὡς+indicative are not constrained by such an interpretation (ὡς+Indicative and ὅτι+Indicative clauses, then, can be said to be in an asymmetrically marked relationship). Additionally, ὡς+Indicative and ὅτι+Indicative clauses diverge in their usage based on the information structure of the event when the speaker is committed to the statement. Cristofaro argues that ὅτι+Indicative clauses is used with utterance predicates where the information communicated by the complement clause is part of the focus: new, asserted information. Conversely, ὡς+Indicative is used with utterance predicates where the information communicated in the complement clause is presumed by the speaker to be already known to the audience, as in example (7).

(7) ὡς τοίνυν τῶν ἐφόρων ἐγένετο, μάρτυρας ὑμῖν παρέξομαι
‘Now, as for the fact that he was one of the overseers, I will offer you witnesses’ (Lysias, Against Eratosthenes, 46.1)

In such cases, the focus of the clause falls elsewhere rather than the content of the clause introduced by ὡς+Indicative, which is the normal expectation for ὅτι+Indicative.

Ancient Greek utterance predicates create a problem for lexical rules because these different complementation patterns cause differences in the semantics of the complement itself, rather than the complement-taking predicate. The result is that these Ancient Greek utterance predicates would result in “implausible verb senses” (582-3). Thus, a verb such as λέγω according to the lexical rules approach, would have a number of lexical rules that would result in derived verb sense, for each of these complement types and their semantics. The problem is that the different means cannot be attributed, directly or indirectly, to the main clause predicate itself. This is especially true in the context of information structure influencing complement type or the degree of commitment a speaker has toward a given statement proposition is entirely dependent upon the complement itself and has nothing to do with the lexical semantics of the relevant verb. The syntax of complement constructions in Ancient Greek is constructional in nature, with the total meaning of a given clause being more than simply the sum of its individual parts.