Home » Articles » NG40. Language acquisition (2)

NG40. Language acquisition (2)

Teddy bearWith its word count pushing beyond 1000, last week’s post had to finish with a question about how a word proposition gets the appropriate C (category) concept.  We now try to answer that question without assuming a logically dubious process. Two possibilities are considered.  One, a general C is elaborated into a set of Cs each shared by words with the same syntactic behaviour.  Two, for each set of words with the same behaviour, unique Cs are generalised into one C.

Acquiring a word – beyond infancy

There are two scenarios for acquiring a word.  The first applies typically to very young children: the word is presented alone but associated with something the child can perceive.

The second scenario applies to any language user: the word is encountered in some sort of sentential context, so the word forms a junction with another word – or possibly forms more than one junction.  LS39 gave the example of the hearer acquiring Poppaea.

The second is perhaps the easier to explain.  As shown in LS39 the hearer acquires (temporarily at least) the word if all the other propositions are in place: the M / R / M intended by the speaker, the C / R / C rule-of-combination and the P / C / M for the other word in the junction.  The new P / C / M naturally gets the C from the rule and the meaning presumably from what the hearer perceives directly or already knows.

Uncertain category?

The newly encountered P is not guaranteed to become associated with the most appropriate C.  For example, two valid rules could already exist:

Both CX and CY form junctions with CA in the same way.  But CX and CY must be separate because they don’t both form junctions with CB:

If the P is first encountered in a junction with a CA word, then it is possible to assign it either to category CX or to category CY:

This would matter if the assignment wrongly allowed the new word to combine with a CB word or wrongly disallowed it from combining with a CB.

But even if wrongly allowed (i.e. the word is assigned CX but should have been assigned CY), our language acquirer should never hear a CB / R2 / (this word) junction from a competent speaker.  And no problem with CB words should arise in production: why would the language acquirer ever want to express something that must be nonsensical?  As for CA / R1 / (this word) junctions, CX works as well as CY.

If wrongly disallowed (i.e. the word is assigned CY but should have been assigned CX), our language acquirer will sooner or later hear CB / R2 / (this word) junctions.  This will have the effect of adding CX to the possibilities the word can have.  (Remember, one phonological word can belong to many categories.)  It’s difficult to imagine how, with no ‘ghost in the machine’, the CY for the word would be specifically deleted.  But it is plausible that it loses long-term activation as a result of disuse.

Acquiring a word – as an infant

Infants are more puzzling because they start with no ‘word’ and ‘rule’ propositions.  It’s easy enough to see how the infant acquires a P / C / M word proposition if they have the M concept.  But is the C concept (MEANS…) completely general – for example the same one would be used for beautiful and for Poppaea.  Or is the C specific to this P and to no other?  You may recall that LS7 claimed ‘Cs are content-less’.  That being so, either way could be possible.

A third way would be that the infant has some innate sense that a concept it has is used as a predicate or as an argument or as a qualifier.  If that were the case the infant could have a small number of Cs rather than just one.

It’s safe to assume that pretty soon the infant must have a wider range of Cs to be able to deal with the subtleties of language (but far fewer than one for each word).  So the process the infant goes through to achieve the right set of Cs is either elaborating from the too-general or generalising from the too-specific.

Need to rationalise

One of the points already made in this post is that a phonological word may adopt a new C and an existing C for that P may fall into disuse (and perhaps expire altogether).  That should be a key feature of the solution – whether from the too-general or from the too-specific.

But that could only happen as a result of a C / R / C rule-of-combination in which the infant hears the previously acquired word being used.  How does that rule come to exist?

It’s reasonable to suppose that the infant must acquire some words before acquiring any rules.  LS39 showed that rules are formed when the infant already knows the real-world proposition and can infer the C / R / C.

For the ‘C is completely general’ case, it’s impossible to see how any rule is ever formulated without the one-and-only C.  That would mean no P could ever adopt a different C.  Therefore C cannot be completely general.

For the ‘a few innate Cs for parts-of-speech’ case, the same would apply except that a C could be superseded in the event that it was mis-assigned initially. That seems far too flimsy to enable wholesale refinement of C allocation.

Here’s how

We’re left with the ‘C is specific to this P’ case.  You may like to reflect on how generalising from the too-specific could work.  Be careful to avoid a solution involving evaluation of the syntactic and semantic behaviour of words.  That would need a ‘ghost’, i.e. a supervisory function performed by a stored program. And that would compromise the foundations of NG.

A newly-encountered P picked up as an individual word gets a new C.  Two different words get different Cs even if they have identical syntactic behaviour:

Each of these two can combine with a third word and two C / R / C rules are inferred:

This situation presumably arises when the infant is at the two-word-utterance stage.  But C3 / R / C1 and C3 / R / C2 would ideally be only one rule.  How can the two Cs be rationalised?

Consider this situation:

The P4 / C4 / M4 word and the C4 / RY / C1 rule exist and the infant also knows the real-world proposition M4 / RY / M2 to relate to an utterance of the teddy__gone junction.  There is no C4 / RY / C2 junction so P2 / C1 / M2 is inferred.  Thus the phonological word P2 and its conceptual meaning M2 acquire C1 – in addition to C2, which was acquired when P2 was first heard as a free-standing word.

Repeated sufficient times, this process has the desired effect.  Yes, there is some redundancy but it doesn’t affect production or comprehension.  In any case, redundant Cs may disappear (as suggested earlier).

NG scores again

Language acquisition is a big subject.  We’ve made a satisfying start.  Let’s park it for now and see if there’s room for more within Network Grammar’s 57-post target.

Next week we’ll look at how junctions are identified in the one-pass, left-to-right processing of a sentence.

Comments