Network Grammar has been showing how a sentence may be processed assuming that junctions within the sentence can be identified as it unfolds, left to right. This post looks at how that identification could be done, using the ideas about ‘progression’ introduced in NG10 together with some new ideas about activation.
Remember, Network Grammar uses ‘word’ to mean one specific P / C / M proposition. The P is usually distinguished as ‘phonological word’ and can of course occur in homonyms.
Back to basics
A good starting-point is the distinction between the parent and dependent roles that a word may play in a sentence. Analyses, notably those on coordination in LS33 et seq, depend crucially on distinguishing these roles and in particular on every word being permitted to play the dependent role no more than once. How can that be done without a ghost-in-the-machine?
NG7 proposed that different instances of a phonological word P can have different conceptual meanings and different syntactic categories. P may occur in several language-knowledge propositions, each linking it to a particular combination of M (conceptual meaning) and C (roughly, syntactic category).
The roles of the three concepts in a proposition (see NG13) are given by the arrow: QUO at its tail, SIC at its head, REL the other one.
Having selected one of these ‘words’ the first time P occurs in a junction, the same M and C must be used for any other junction in which P occurs within the sentence. Other words with P may have nothing in common; for example bear as noun and bear as verb.
Other words may share the same M; for example fire in fire wood and in log fire.
It’s also possible to share C; for example bank as noun.
The idea of using the same C for two disparate concepts that happen to allow the same syntactic behaviour may seem strange. But it is perfectly compatible with the following ideas given that one or other of the words is preferred as a result of earlier activation from the discourse.
Only once as dependent
It’s now necessary to refine the ‘same P / C / M’ principle to account for P occurring not more than once as dependent but any number of times as parent. The assumption is that the P / C / M is endowed with six units of activation as a result of the phonological occurrence of P. P / C / M words are paired as a junction subject to two conditions. First, there must be a C / R / C rule-of-combination for the two Cs. Second, the dependent P / C / M must have six units of activation; the parent P / C / M may have either six or zero units. The C / R / C takes over the six units from the dependent P / C / M, leaving the latter with zero units (and therefore able to participate in further junctions as parent). The six units on the C / R / C are then used to activate the M / R / M proposition formed from it.
This is a simplification. For one junction there may be multiples of C / R / C rule with M / R / M proposition. For example, Nero__gave generates three pairs – for AGENT, THEME and GOAL.
The conventions in LanguidSlog are that parent is QUO and dependent is SIC, and that the left-hand concepts in the diagram relate to the earlier word in the sentence and the right-hand concepts to the later word. Thus the arrows joining the Cs and joining the Ms may point either way. Here they point leftwards for Nero__gave.
Prioritising preceding words
An incoming P must be tried with each of the existing Ps, in reverse chronological sequence, to see if the pairing is a valid junction. This is enabled by chaining Ps together using links with a relation that can be called ORDER (circled O in the following diagrams).
The idea is that, for each successive P, a (null) / ORDER / PN proposition is created (or perhaps pre-exists in language knowledge for each P). The following P replaces the null to complete PN+1 / ORDER / PN, as well as bringing its own (null) / ORDER / PN+1 proposition. Thus a chain is formed. Then, positioned on any particular P, ‘progression from QUO’ leads to the immediately preceding P and so on.
For incoming PN, the newly-extended chain is followed back in order to assess whether PN1__PN, PN2__PN, PN3__PN etc are valid junctions.
The diagram is incomplete because it only shows possible pairings between neighbouring words. P2__P4, P1__P4 and P1__P3 are also possible junctions.
Sequencing parent and dependent
The nodes in candidate C / R / C rules are not labelled in the diagram. They may be arranged either way: incoming C as QUO and existing C as SIC, or existing C as QUO and incoming C as SIC. Both ways, CSIC represents the dependent in the junction and CQUO represents the parent. As already mentioned, activation of a valid C / R / C comes from the dependent, i.e. via QUO.
More to come
If an existing P / C / M has zero units of activation, it can only form a valid junction with the incoming P as its parent. But if an existing P / C / M still has six units, it can form a valid junction with the incoming P either as its dependent or as its parent. This second case is puzzling: how can aaa-then-bbb be correctly identified as parent-then-dependent or as dependent-then-parent?
See you next week!