More problems are thrown up by coordination. This time the tricky sentences are undeniably grammatical and an extension to NG’s range of mechanisms is needed.
New mechanisms?
What if a later clause has no ellipsis at all?
(134) John gave Mary apples and Ken sold Nancy bananas
At Ken the analysis has to allow for GIVE / GOAL / KEN, GIVE / THEME / KEN (Ken being a slave of course), SELL / GOAL (or THEME) / KEN (Ken being a fronted object) as well as SELL / AGENT / KEN which becomes clear as the speaker’s meaning once sold is encountered.
A mechanism is needed to allow options to be kept open. One possibility is that nothing is delivered until the clause is finished (let’s just assume that the sentence processor can know when that happens). Then the syntactic jigsaw puzzle is solved all at once. This approach was already hinted at in LS36 to address a ‘loose end’ in the analysis of sentence (133).
A less radical approach would be to extend a mechanism which has already been used for many give sentences: allowing a noun immediately following the verb to be either GOAL (if a double-object sentence) or THEME (if a to-dative). Premature delivery to cognition is avoided by creating two propositions and splitting the noun’s activation between them.
The most likely continuations after Ken are as in (134) above and in (128):
(128) John gave Mary apples and Ken bananas
For these an even simpler mechanism could suffice.
Simplest option
First it must be assumed that a gave__and junction alters the C concept (see LS26) for gave. (This is slightly at odds with LS34 which proposed that and forms a junction with the nearest preceding word in order to deliver the COORD / HASP / AND proposition.) With the different C, subsequent junctions with it can be treated differently if necessary. For and__Ken, the proposition (null) / (null) / KEN could be generated with six units of activation.
Then, for (128), gave__bananas could generate GIVE / THEME / BANANA and GIVE / GOAL / (null). The incomplete propositions have nulls in complementary positions (see LS12) and so they combine to form GIVE / GOAL / KEN.
For (134), gave__sold could generate SELL / AGENT / KEN, SELL / THEME / (null) and SELL / GOAL / (null). Note that gave__sold is specified. Ken__sold is not possible because Ken has already occurred as dependent in and__Ken. However its activation has not yet been expended delivering a complete proposition.
(134) John gave Mary apples and Ken sold Nancy bananas
Without Ken__sold there remains a problem accounting for KEN and its activation (a) not attaching to one or other of the brought-forward give propositions, and (b) attaching to the SELL / AGENT proposition rather than to THEME or GOAL. (Usually the order of noun and verb in the junction is the distinguishing factor.) Perhaps and__Ken doesn’t count – it only generates a vestigial proposition – and Ken__sold is allowed after all.
Other continuations
The to-dative variant of (134) presents no further problems:
(135) John gave apples to Mary and Ken sold bananas to Nancy
A later clause need not follow the same pattern as the first:
(136) John gave Mary apples and Ken sold bananas to Nancy
Another possibility is no continuation. Mary receives a some fruit and a slave:
(137) John gave Mary apples and Ken
The attempt to coordinate unmarked and marked nouns (see LS34) surely makes this a garden-path sentence. We need not explain how the (null) / (null) / ken proposition delivers the intended meaning.
An interesting one
More interesting is what happens if the later clause has a fronted object. Sentence (132) in LS36 showed that a fronted indirect-object in a later clause is not analysed correctly because the three successive nouns are interpreted as AGENT, GOAL and THEME respectively. What if the verb is explicit?
(138) Mary John gave apples and Nancy Ken sold bananas
Sentence (138) seems OK and analysis of its second clause is straightforward. The indirect-object following immediately after and can be treated the same a the subject in (134). Note that there can be no Nancy__sold junction because and__Nancy has already occurred. In this case, assigning nancy to goal is unproblematic.
Anomaly?
Sentence (139) also seems acceptable but the analysis would be like (138) with BANANA (and APPLE) as GOAL and NANCY (and MARY) as THEME.
(139) Apples John gave Mary and bananas Ken sold Nancy
If to is included, the preposition phrase ensures GOAL and therefore THEME get the correct concepts.
(140) Apples John gave to Mary and bananas Ken sold to Nancy
The acceptability of (139), despite sentence processing delivering incorrect propositions, could mean that cognition interprets the meaning from the semantics.
Verb-less clauses again
That (138) is both analytically and intuitively correct while (132) is not shows the importance of an explicit verb in the later clause.
(132) Mary John gave apples and Nancy Ken bananas
Variants of (132), still without a verb in the later clause, must all be similarly misanalysed but seem slightly less bad.
(141) Mary John gave apples to and Nancy Ken bananas to
(142) Apples John gave Mary and bananas Ken Nancy
(143) Apples John gave to Mary and bananas Ken to Nancy
Compared with (132) they do include some clues to help cognition – preposition to or inanimate first noun or both. Drawing reliable conclusions would need research into the acceptability of these sentences. Pending that, the best guess is (132), (141), (142) and (143) are garden-paths with (132) the least likely to be correctly re-interpreted by cognition.
Sentence (132) cannot be rescued by prosody and none of (134) and (138) to (143) is affected by prosody.
Other points
The approach can also handle a verb in a later clause that has a different ‘valency’ from the verb in the earlier one.
(144) John gave Mary apples and Ken peeled bananas
The approach doesn’t interfere with the analysis of sentences like (126) with nothing in the later clause before the verb.
(126) John gave Mary apples and sold Nancy bananas
Not finished yet
Perhaps one more long post like this one will finish off coordination. Sorry it’s such a big topic.
Also the caveat following the analysis of (134) is a real weakness in the argumentation. Any suggestions for how to resolve it?