Network Grammar has slogged for nearly 10 months analysing sentences with one-word arguments – Nero, apples etc. We now look at multi-word noun phrases and solve some problems.
In a noun phrase, words may precede the ‘head’ noun:
(157) Darnley kissed the young queen
Neither the nor young can be in a junction with a preceding word in the sentence. But the__queen and young__queen should both be possible and, although queen is parent in those junctions, kissed__queen is still possible. However, …young the queen is not possible and so the__young is the relevant junction in (157), not the__queen.
Some grammars replace ‘noun phrase’ with ‘determiner phrase’, DPs having articles, demonstratives, possessives etc as head. NG prefers to have verbs in junctions with nouns and will have to solve by other means the problems for which the ‘DP hypothesis’ was mooted.
Of course NG easily handles anomalies such as:
(158) Darnley imposed martial law
(159) Darnley faced a court martial
(160) Darnley had faced several courts martial
In English, qualifiers following a noun are not limited to oddities like (159) and (160):
(161) Darnley kissed Mary the young queen
Let’s ignore for now the possibility of ambiguity about the number of ladies/kissings (perhaps Mary is someone other than the queen and and occurs somewhere to the right). Anyway a more familiar phrase seems to avoid any hint of that:
(162) Darnley kissed Mary Queen of Scots
In both (161) and (162) it might appear that kissed__Mary can be processed as soon as Mary is encountered. Immediate processing is implied in earlier LanguidSlog examples. But it cannot be the case:
(163) Darnley kissed Mary Queen of Scots’s maid
Now MQoS is not an argument of kissed but a qualifier of maid. Processing has to deal with a bunch of nested noun phrases before reaching the word giving kissed__maid.
The principle seems to be that a phrase must be fully processed before its head word can be joined elsewhere in the sentence.
To control the sequence in which the words in a sentence are joined, NG can use two things. One is activation, particularly the principle that the activation brought by a word is depleted when that word participates for the one-and-only time as dependent in a junction. The other is the mutability of the C (category) concept for a word; this idea was first introduced in LS26.
Activation is important because, when a phrase has been processed by NG, only the head word retains its activation; every other word has been a dependent somewhere. That is obviously true in (163) because the last junctions processed are Mary__’s and ’s__maid, but it is true wherever in the phrase the head word is located.
Category mutability is important because it prevents junctions being formed prematurely.
Still using (163) as the example, the proposal is that the category for kissed is initially CA. There is a rule for CA and CX, the category of Mary:
The rule doesn’t generate a proposition because there is no R concept. The activation from Mary is not depleted but the leftwards scan for Mary is stopped. Crucially the category for kissed is changed from CA to CB because there is another rule for CB and CX. When the end of the noun phrase is reached, two words with activation remain and these can form a junction:
The CB / R / CX rule can’t be invoked when Mary is first processed because at that point kissed has CA as a result of John__kissed.
In the diagram, the P and M concepts on the right have not been labelled because they can be for either Mary in (162) or maid in (163). The assumption is that Mary and maid share the same C – at least for kissing. In (163), Mary has acted as dependent in Mary__’s and so cannot be the dependent in a junction with kissed.
This may all seem like hocus-pocus but there are only two new ideas. First, there can be a rule whose sole purpose is to change the C of one of the words in the junction. Second, in a situation like (162) two still-active words to the left of the current position in the left-to-right scan can form a junction and deliver a proposition.
We’ve only looked at noun phrases as objects immediately following a verb. The same treatment can apply to an NP complement of a preposition:
(164) Darnley sang for MQoS’s maid
And it can apply to an NP as the direct object in a double-object sentence:
(165) Darnley gave the maid MQoS’s jewels
The DO case needs explaining. At maid in the first NP, the category of gave will have changed from CE to CF (despite the simplicity of the phrase). At Mary the category must change again – from CF to CG. This is not a problem because the rules are different: CF / GOAL / CX and CG / THEME / CX.
Actually the story is a bit more complicated because the first post-gave NP can be for either GOAL or THEME. At maid in (165), the two incomplete propositions would be created (see LS8) using a rule with CF. The proposition from jewels would use a rule with CG. CF and CG are syntactically identical, existing only to enable correct processing of NPs.
An NP in subject position is straightforward:
(166) MQoS’s maid gave Darnley satisfaction
The relevant rule is CE / AGENT / CX.
One objection to this proposal might be doubt about how the C-changing rule could be acquired. This is indeed worthy of detailed study beyond the scope of LanguidSlog.
Provisionally we can say that acquisition as described in LS39 and LS40 looks feasible. The infant must initially infer that the first noun after a verb is its object. But the infant using such a rule will soon learn that a language-delivered M / R / M proposition doesn’t always match what they perceive. A C-changing rule wouldn’t be difficult to acquire by trial-and-error. There’s no implication of innateness.
There are some loose-ends in the above. One is the question of what REL concepts are invoked by determiners. Another is the treatment of possessives. I’ll try to cover these things before ending the blog on schedule at NG57 – but don’t bet on it. By now the avid Network Grammar follower should be able to work out some answers at least as well as I can.