Grammar: syntax and the lexicon

An introduction to Grammar: Syntax and the Lexicon

Introduction.

We start this page with the hypothesis that the lexicon is initially underspecified: that some features may be NIL (not specified as either plus or minus). A NIL feature is represented either with no marker, or with NIL, or with 'º'. We may start our grammar either with phrase structure rules (bottom down) or rule apply bottom up. Let's start with phrase structure rules.

Let us start with a simple expansion--that of QuP (quantifier phrase, based on the syntactic project Qu) which may also be written as NP (based on the semantic projection N). In the syntactic projection of Qu, Qu is the head, NP is its argument, and the prase should be written QuP.

QuP --> Qu + NP (NP --> Qu + NP).

NP -> N (PP)

The NP expansion rule is an incomplete rule, but it will suffice here. N is the head of NP, and PP is an optional argument. Let us expand QuP and NP, choosing a noun which takes no complement:

The colour red here means that each node and expansion is the result of a process. That is, QuP is the input to the rule which expands QuP as Qu and NP, and NP as N.

Next a lexical item is selected and inserted in the appropriate nodes. Let us select 'TWO' and 'CAT', respectively and insert them as underspecified lexical items. That is, the feature Pl is not an inherently specified as plus or minus; but it must be specified by a grammatical rule. We assume here that the features of the lexical item are hard wired--that is, these features are stored directly in the mental lexicon. This is the underspecified hypothesis. In the following diagram black represents hard wired information.

Note that the nil plural feature becomes plus. This happens through agreement when Qu governs NP and N. This is a process and is marked in red above. In the underspecified hypothesis, structure expansion of feature agreement is a process each time a sentence is uttered.

Jackendoff () and Halle () and others have claimed that the lexicon is fully specified. Chomsky 1995 concurs noting that there should be some sort of rules that determine this form. It seems to us that the above processing rules are each just that sort of a rule--a rule based on predicate logic and semantic input. Let us adopt the claim that each lexical item is fully specified.

It seems reasonable to suggest that the above rules and other similar rules apply in the formation of each lexical item. The resulting lexical item is then fully specified. These forms become learned and are stored in the lexicon. Let us adopt this point of view.

Now, if the result of the expansion and agreement rules becomes stored in the lexicon, it stands to reason that the result of the process must be too.

We assume that the vast majority of lexical itaems are not hard wired to a structure, but that the structure itself is hard wired. Any lexical item can be inserted into a slot, but only those that match will pass. That is, if a verb is inserted into a nominal node, the features will class; they do not agree. In the above diagram the red arrow indicats that the lexical item that it points to is inserted as a process, but the lexicall item itself, which is in black, is hard wired.

Now if the firs diagram is in some sense true, and the second one is derived from it, it stands to reason that they are linked in the lexicon; i.e. the underspecified one feeds the specified one. To put it in another way, CAT does not occur twice in the lexicon, but once. The information listed in the underspecified entry is hard wired to the fully specified entry, thus requiring one bundle of memory cells. Only the feature listed in red is not hard wired to the underspecified entry.

Note also that if one is chosen to modify cat, the feature of plurality is minus. A fully specified lexical entry is inserted in each node:

Both the fully specified lexical items are connected through derivational hard wiring to the initial underspecified lexical item. That is, CAT,[-Pl] and CAT,[+Pl] are linked directly to CAT. The only items not fully connected are the values of plurality:

The two surface lexical items (word forms) are hard wired to the basic lexical entry; there is no processing that takes place between them.

Let us return to the issue of phrase structure rules. Although it is rarely mentioned if ever at all, phrase structure rules seem to be regarded as process rules. I was certain they were. But we are now suggesting that phrase structure rules may be presyntactic ones--ones that develop structures through which features must agree. That is, the structures derived by the phrase structure rules become hard wired and remain so. This means that the figure at the top of the page is now hard wired. It is reproduced in black:

Obviously, we are encroaching on the field of syntactic and lexical acquisition. We are not trying to claim how, when, or how long it takes a child to acquire the above figure, but he must do so if our approach is correct. Once acquired and burned into the grammar, it remains there until death (in most cases).

In an encoding process the speaker of English would select TWO (or ONE) and place it in Qu, and select CATS and place it in N. However, before he can do so, CATS must occur there. So far, all we have generated is {CAT,[+Pl],... ,} the lower right hand lexical item in (5) above. {CAT,[+Pl], ...} is spelled out as cats, but how. We propose how to do this in Deriving the Number of the Noun . There, we spell the feature bundle out as:

'a features' means that whatever features that in one place marked with a must occur in another place also marked with a. This is the final step of the lexical rule deriving cats It is the hard wired form.

This page updated 6 SE 2000

Top of Page