Constraint-based Syntax 2: Week 2

Parts of Speech

Our online grammar uses the following type hierarchy for parts of speech (the values of the feature HEAD):

The hierarchy differs from that in GS in containing the part of speech adverb. The following table lists the parts of speech, their features, and the function of the features:

a. The maximal part of speech types are the following:

Type	Full name	Examples
v	verb	snore, like, must, to
g	gerund	snoring, liking
c	complementizer	that, whether, if
a	adjective	big, new, red
p	preposition	to, of, with
n	noun	Lilly, she
d	determiner	the, a, any
adv	adverb	early, very, not

Note that v and g are grouped together, since verbs and gerunds behave very similarly. Perhaps surprisingly, complementizers are also grouped together with verbs and gerunds, so that all three parts of speech share the head features VFORM (verb form) and IC (independent clause). We will see later why that is useful.

Head Features

The next table lists all the head features of all the parts of speech and explains their function:

Feature	Value type	Function of the feature
PRED(icative)	boolean	Is the sign predicative or not?
MOD(ification)	mod-synsem	Can the sign modify and if yes, what kind of synsem?
VFORM (verb form)	vform	What is the verb's verb form?
IC (independent clause)	boolean	Is the sign (the head of) an independent clause?
AUX	boolean	Is the sign (the head of) an auxiliary construction?
INV(erted)	boolean	Is the sign an inverted auxiliary (construction)?
POL(arized)	boolean	Is the verb a polarized auxiliary (not, TOO, so)?
P(reposition)FORM	form	Which preposition is it?
CASE	case	What case does the nominal sign carry?
SPEC(ifying)	sem-obj	Which semantic object does the D specify?
NEG(ation)	boolean	Is the adverb the negation not?

The verb form hierarchy

The different verb forms that the framework makes available form another important aspect of the theory:

In the text, GS also talk about a verb form subjunctive which is supposed to be used in imperative and subjunctive clauses. They do, however, not include that verb form in their grammar specification. In order to analyze these sentence types, the online grammar for this course adds a verb form sbjn" (= subjunctive) to the type hierarchy above. It assumes that sbjn" is another immediate subtype of clausal", so that clausal has the three immediate subtypes sbjn", fin(ite), and inf(initive). There will be a constraint that clauses are required to have a clausal part of speech, which means that only words with VFORM sbjn", fin(ite), or inf(initive) can head clauses.

The table below gives an overview of all verb forms with examples:

Verb form	Full name	Examples
sbjn	subjunctive	be, dance, call
fin	finite (indicative)	is, must, dances, dance (non-third person singular), danced (past tense)
inf	infinitive	to (the infinitive marker)
base	base form (= bare infinitive)	be, dance, call
prp	present participle	being, dancing, calling
pfp	perfect participle	been, danced, given
pas	passive participle	invited, rumored, introduced

Valence

In Week 1 we saw that all signs carry the 3 valence features SUBJ, SPR, and COMPS, which are list-valued. Corresponding to each valence feature, there is a phrase type: hd-subj-ph, hd-spr-ph, and hd-comp-ph. As the names suggest, each phrase has a head daughter and additionally a SUBJ, SPR, or one or two COMPS daughters. The projection of these phrases is driven by the head daughters valence features as illustrated in Week 1.

The linguistic objects on the valence lists (i.e. the valents) are of type synsem. In order to simplify the representations of the valents, we use the following abbreviations:

VP = a phrase of part of speech v which is COMPS <> , SPR <>, and SUBJ <synsem>

AP = a phrase of part of speech a which is COMPS <> , SPR <>, and SUBJ <synsem>

S = a phrase of part of speech v which is COMPS <>, SPR <>, and SUBJ <>

NP = a phrase of part of speech n which is COMPS <>

PP = a phrase of part of speech p which is COMPS <> , SPR <>, and SUBJ <>

This permits us to write valence lists like the following, where the information in the brackets further specify inflectional properties of the valents which should be self-explanatory:

Example	Valence list
snores	<NP[nom]>
likes	<NP[nom],NP[acc]>
put	<NP,NP[acc],PP>
persuade	<NP,NP,VP[inf]>
afraid	<NP,PP>
of	<NP[acc]>
that (C)	<S[fin]>

Words

So far, we have introduce the following concepts:

• Our grammar deals with signs which are either words or phrases.
• Every sign belongs to a maximal part of speech, which may have features.
• Signs also have the 3 valence attributes SUBJ, SPR, and COMPS which contain a description of the syntactic and semantic information of other signs.
• These valence features drive the formation of specific phrases.

In preparation of a more detailed look at phrases, look at the lexical entries of the words in the table below in the online grammar. Pay attention to the part of speech, its features, and the 3 valence feature of each word:

Online Grammar for Chapter 2: HPSG Background

Type	Full name	Examples
v	verb	snore, like, must, to
g	gerund	snoring, liking
c	complementizer	that, whether, if
a	adjective	big, new, red
p	preposition	to, of, with
n	noun	Lilly, she
d	determiner	the, a, any
adv	adverb	early, very, not

Maximal phrase types

The part of speech v

v is the part of speech of all verbs and auxiliaries, including the infinitive marker to.

The table below is to be interpreted as follows. In the leftmost column, we find parts of speech. The top-to-bottom order represents the immediate subtype relationship, i.e. verbal is an immediate subtype of subst, verb is an immediate subtype of verbal, and v is an immediate subtype of verb.

The second column lists the features introduced by each part of speech, followed in the third column by the value type of the feature, and finally a short statement of the function of the feature in distinguishing different subgroups of the part(s) of speech under discussion.

Part of speech	Feature	Value type	Function of the feature
subst	PRED	boolean	Is the expression predicative or not?
	MOD	mod-synsem	Can the expression modify and if yes, what kind of synsem?
verbal	VFORM	vform	What is the verb's verb form?
	IC	boolean	Is the expression (the head of) an independent clause?
verb	AUX	boolean	Is the expression (the head of) an auxiliary construction?
	POL	boolean	Is the verb a polarized auxiliary (not, TOO, so)?
	INV	boolean	Is the expression an inverted auxiliary (construction)?
v

Let us now ask what this means for expressions of part of speech v. v itself does not introduce any features, but since every v is also a verb, a verbal, and a subset, an expression of part of speech v has all the features introduced by its supertypes and the corresponding values:

[v

[PRED boolean]

[MOD mod-synsem]

[VFORM vform]

[AUX boolean]

[POL boolean]

[INV boolean]]

Task: go to the online grammar and parse the following words in order to see the values for their head features:

a. like, likes, liked, liking

Note that those verb forms which can serve as the complement of the verb be (the passive liked and the progressive liking) are [PRED plus], whereas all the remaining forms are [PRED minus]. Also, all main verbs are [AUX minus], from which it follows that they cannot occur in auxiliary constructions.

b. does

The forms of the unstressed auxiliary do are the only words of English which are restricted to occurring in auxiliary constructions. Thus, they are all marked [AUX plus]. Note that all the forms of the auxiliary do are finite.

You get two solutions when you parse does, one which is [POL minus] and a second one which is [POL plus]. You will notice that this contrast goes along with a difference in COMPs lists of the two verbs. The non-polarized verb has a single complement, namely a VP, whereas the polarized word selects an additional complement, namely a polarized adverb (one of the words not, TOO, or so).

c. will

This is a typical modal verb. Like the other modals, it has only finite forms. Again you get two solutions, a polarized and a non-polarized one. An intriguing difference between will and the unstressed do discussed above, is that will can occur in auxiliary constructions (and then is [AUX plus]), but it can also occur in constructions where main verbs can occur as well (and then, like the main verbs, will be [AUX minus]. In order to be compatible with both values of the feature AUX, will, it takes the neutral value [AUX boolean] in its lexical entry. All verbs that can appear in auxiliary constructions except for the forms of the unstressed do behave like will.

-->

The Argument Realization Principle (ARP)

Recall that in GS the syntactic arguments of a word are in a sense represented twice: once on the ARG-ST (argument structure) list which, among others, is used for Binding Theory, and again on one of the valence lists SUBJ, SPR, and COMPS. Rather than being arbitrary, the relationship between the ARG-S and the valence lists is governed by the Argument Realization Principle:

In words, the principle says that the ARG-ST list is the result of merging the three valence lists into one list, with the elements of the SUBJ list coming first, followed by the elements on the SPR list, followed, finally, by the elements on the COMPS list.

Besides the ARP, the valence properties of words depend on their part of speech. Thus, all words of part of speech v must have a SUBJ list with exactly one element on it and a SPR list which is empty. In combination with the ARP, verbs of part of speech v thus must have the following properties:

1. The ARG-ST must have at least one element on it.
2. The first element of the ARG-ST is also the single element on the SUBJ list.
3. If there are additional elements on the ARG-ST, then these elements also occur on the COMPS list.
4. The elements which appear both on the ARG-ST and the COMPS list, appear in the same order on both lists.

The following Exercises on the append relation have the purpose of showing that the 4 statements above must be true.

Exercises on the Argument Realization Principle

  ↑ Main page         ↑ Week 1