Home » Articles » NG42. Identifying junctions (2)

NG42. Identifying junctions (2)

 At the end of last week’s post we asked ‘How can the sequence aaa-then-bbb be correctly identified as parent-then-dependent or as dependent-then-parent?’  The answer is that aaa-then-bbb can have only one of those possibilities.  This post explains why that is so.

Evidence and verdict

An example such as saw__mill and saw__mill seems to invalidate this assertion, but aaa and bbb must each be treated as a P / C / M word (not, more broadly, as a phonological word).  The rule-of-combination linking aaa and bbb is C / R / C, not P / R / P.  In the example the two instances of saw differ in terms of both C (syntactic category) and M (conceptual meaning).

There may be exceptions to this but they seem rare.  An exception need not jeopardise the argument here because it could be explained either by the combination being lexicalised as a single entry (like sawmill perhaps) or by being dealt with in cognition as a garden-path.

There is evidence to support the assertion that aaa-then-bbb can have only one arrangement of parent and dependent.  Logically, aaa and bbb can appear in two sequences and parent and dependent can appear in two sequences, giving four possible junctions:





But empirically, no more than two of these possible junctions may co-exist for two P / C / M words:

aaa__bbb and bbb__aaa;  for example rice__pudding and pudding__rice

aaa__bbb and bbb__aaa;  for example John__gave and gave__John

aaa__bbb and bbb__aaa;  for example to__start and start__to

As noted above for saw and mill, the fourth combination is rarely if ever encountered.  Indeed it is the absence of this combination that prevents co-existence of any more than two of the four possible junctions:

aaa__bbb and aaa­­__bbb

This combination is the only one in which the sequence of the words is the same in both junctions.  The conclusion is that the sequence of a given pair of words in a junction determines the parent/dependent arrangement.


This is perhaps the single most important point so far.  There is no purely logical reason why all four possible junctions between a given pair of P / C / M words should not deliver distinct meanings.  There is no purely logical reason why, for a given pairing, only two of the four should be available.  And there is no purely logical reason why those two rarely include the two words in the same sequence.

In truth NG’s three-concept propositions would struggle to support four rules from a pair of P / C / M words but they support two rules with elegant simplicity – aaa-then-bbb or bbb-then-aaa.  Apart from the architecture that NG assumes there is no apparent reason why four rules are not available.  If ‘only two rules’ were true across all languages, the case for NG would become even more compelling.

Finding the rule

So far it has been assumed that a previous (left-hand) word for which a pairing with the incoming (right-hand) word is sought has already participated in a junction.  That being so, the appropriate P / C / M word has been identified and activated and, if the P has been a dependent, had its activation depleted.

This is a simplification.  Obviously the first word in a sentence cannot be thus when pairing with the second word is tried.  Also the second or a subsequent word (apart from the sentence-final word) might not form any junction in which it is the right-hand word.  This is illustrated by the first and second words in Olivia Nero gave to Poppaea.

The quest for a C / R / C rule for a pair of words must therefore be considered generally as multiple P / C / M words for incoming Pright trying multiple P/ C / M words for previous Pleft.

The diagram shows a large number of possible left/right pairings of P / C / M words.  (It should perhaps show what is probably more typical: a P having more possible Ms than possible Cs.)

Only one junction can be formed from Pleft and Pright.  Multiples are impossible because, if aaa__bbb were formed there could only possibly be residual activation to allow aaa__bbb; or vice versa.  But the combinations available under ‘only two rules’ do not include aaa__bbb and aaa__bbb.

It might be argued that, for a pair of Ps with multiple P / C / M words, there could be other valid pairings.  That may be true but is irrelevant: once one valid pairing has been found, only those specific P / C / M words may be used for forming a junction.

Can there be contention between two C / R / C rules for the same two P / C / M words?  The answer must be NO if LS39 is correct about how rules-of-combination are learned.  Once there is a rule for CX and CY, no other rule can be formed with the same Cs in the same sequence.

‘Only two junctions’ is expressed in NG’s own terms.  But phrase-structure theories also rely on asymmetric relations essentially the same as parent/dependent.  If the impossibility of both aaa__bbb and aaa__bbb is empirical fact then the NG-based sentence processing described in these essays can provide a theoretical explanantion.


Thus for any two words aaa and bbb, one junction is identified or none.  If one is identified, only one C / R / C rule is selected.  Therefore no ambiguity can arise from the process itself.

…or not?

First-year linguistics students are hit with sentences like:

(150) Visiting relatives are a nuisance

(151) Visiting relatives is a nuisance

There must be a junction between Visiting and relatives.  OK, the Cs for these words are different – (verbal adjective)(noun) in (150) and (gerund)(noun modifier) in (151).  But the analysis must keep open both possibilities until the verb is encountered.  That needs to be covered in a later Network Grammar post.


This site uses Akismet to reduce spam. Learn how your comment data is processed.