Draft of February 7, 1997
In this warm-up first chapter we present, in an informal way, a sample of some of the most basic notions that must be treated in a theory of grammar. Our initial observations will largely emerge from a careful consideration of a single English sentence. That sentence is:
(1) Does she realize how heavy the big ones are?
We will limit our attention here to morphosyntactic aspects of linguistic structure: phonological matters such as the features of phonemes, syllable structure, foot structure, intonation, and the like will be out of consideration, though questions of interface with such matters do indeed belong in our domain.
Constituent Structure
The first thing we need to realize is that sentence (1) is more than a simple string of nine words, one following the other: it has a structured organization. The ultimate syntactic constituents of sentence (1) are its WORDS, but it can be shown that these words are grouped into phrases, and some of these phrases are grouped into still larger phrases, each of these in its turn being a proper constituent of the sentence. The full segmentation of a sentence in terms of such grouping and regrouping is what we refer to as its phrase structure or constituent structure.
If a sequence of three linguistic entities, call them A, B, and C, are grouped in such a way that the A and the B "belong together" in the sense that they jointly form a unit of linguistics structure, we can represent such a grouping by the bracketing shown in (2).
(2) [A B] C
To say that the constituents A and B "belong together" means that the sequence formed by an A followed by a B constitutes a unit of grammatical structure. If we also wish to show that the whole A B C sequence is itself a grammatically relevant unit, then we can add that information by augmenting the diagram as shown in (3), where the whole thing is surrounded by a pair of brackets:
(3) [[A B] C]
The bracketing in (3) can be read as claiming that the sequence A B C is itself a grammatical unit and that it has a substructure in which a unit with the form A B is one of its component parts.
We can approach the constituent structure segmentation of a sentence from the top downward, asking first what the largest constituents are and then asking whether these constituents can themselves be further segmented, or from the bottom up, looking for groupings of adjacent words or phrases. The results will be the same either way. We might notice, for starters, that the sequence how heavy in sentence (1) deserves to be regarded as a unit of some sort. That is, the words how heavy are felt to belong with each other in a certain way, and we have no such "feeling" for either realize how or heavy the in that same sentence. (Compare the phrasing suggested by (4)a with that shown in (4)b-c.)
(4) a. Does she realize [how heavy] the big ones are?
b. Does she [realize how] heavy the big ones are?
c. Does she realize how [heavy the] big ones are?
Of course if we had only our FEELINGS about how things fit together, that would not be very satisfactory. If you were to challenge us by claiming that the analysis suggested in (4)c strikes you as more reasonable than that of (4)a, we owe you better reasons for claiming the analysis in (4)a than our feelings. What we really mean when we say that how heavy makes up a constituent in this sentence, whereas realize how and heavy the do not, is that the grammar of English that we are assuming - which we take as incorporating (part of) the knowledge that speakers of English have of their language - licenses the formation of a phrase like how heavy but does not license the formation of the tentative phrases realize how or heavy the. More precisely, we say that the grammar of English is a collection of grammatical constructions, such that each grammatical or grammatically well-formed English word or phrase is licensed by one or more such constructions. A sentence or other linguistic expression is ungrammatical, or grammatically ill-formed, if there is no combination of grammatical constructions capable of licensing a phrase consisting of this sequence of words.
For the present case, we would say that in a properly drawn up grammar of English there is a principle, or rule, or construction, which defines phrases of a particular type by specifying that they can contain a constituent of the kind exemplified by how placed before a constituent of the kind exemplified by heavy to form the kind of phrase we find in how heavy. Such a construction has an internal part which specifies what kinds of constituents can combine with each other in the way that how combines with heavy, and an external part which specifies what kind of constituent results from this combination.
How do we know that how is the kind of word that can occur first in this phrase, and that heavy is the kind that can occur second? It is up to the rest of the grammar- including, in this case, the lexicon - to guarantee that words or phrases capable of occurring in phrases defined by this construction receive the necessary descriptions. How do we figure out what properties to assign to the phrase as a whole? It is the responsibility of the rest of the grammar to tell us how phrases of this type function elsewhere in the language.
Decisions about assigning properties to the whole phrase depend on the existence of grammatical constructions that permit such phrases to combine into still larger structures, or that permit them to be used alone as English utterances. The task of assigning labels to phrases is often spoken of in terms of a visibility metaphor. The constructions which can "use" the phrase we have just considered can "see" only the information that identifies the phrase as a whole and cannot "see" the phrase's innards; if it is important for knowing the distribution of a phrase like how heavy that the phrase contains the specific word how, then that information will somehow have to be a part of the description of the phrase as a whole.
The point of the distinction between internal and external structure can be understood by imagining the following (somewhat pointless) kind of activity. Suppose we have the job of packaging things according to their labels, and then putting labels on the packages we produce. One of our assignments is to package an X and a Y together and to label the package Z. We find an object labeled X and an object labeled Y and we wrap them up together. We then label that package Z. We receive a new assignment, which is to package an A and a B and to label the package Z too. Now if we're given the job of packaging a Z and a W together into a package labeled Q, we will be able to use either of the Z packages we have just created, or any other object labeled Z, but we don't have to remember what they contain. We can only "see" the label Z, not its contents.
One of the theoretical questions that will come up later in the text will center around the question of whether all syntactic operations are subject to a visibility constraint and what such a constraint would actually mean, In the first place, if we impose no restrictions on the content of a "label" this "constraint" can be easily fudged. It may be that there are some processes that only need to know, for the above examples, whether or not something is a Z, and so the label will have to include the "Z" specification. But the label could also include the information that a given Z contains an A and a B, for example. But furthermore, there might be some grammatical processes (say, recognizing antecedency for any number of anaphoric relations) which determine the appropriate combinability of phrases only by checking pthe nature and location of particular components. Suffice it for now to say that what the components of a phrase are is one thing, what kind of thing the phrase itself is is another.
Continuing to examine sentence (1), we might notice that the sequence consisting of the last six words - how heavy the big ones are - "goes with" the verb realize, in the sense that it expresses the proposition that is the content, so to speak, of a mental process of "realizing". Certain words impose requirements on what kinds of expressions they need for completing their job in a sentence. These requirements have both semantic and syntactic aspects: in the case of realize, we need to know what kind of semantic entity it requires, and we also need to know what grammatical form an expression of that entity must or may assume. We speak of these requirements as the word's arguments. A description of the verb realize, then, must tell us what kind of phrase it needs to accompany it in a sentence, and the description of the phrase how heavy the big ones are must likewise tell us that it has the properties allowing it to serve as an argument of the verb realize.
The decription of many aspects of a word's co-occurrence requirements is what we will refer to as the word's valence. It will become obvious later that there are close correspondences between a word's semantic requirements and its syntactic requirements: the kind of meaning that needs to be expressed is often realized in only a limited number of ways syntactically.
In any case, the grouping assumptions we have made so far lead to the structuring seen in (5):
(5) Does she realize [[how heavy] the big ones are]?
The bracketing in (5) shows that the last six words of the sentence (enclosed in one pair of brackets) make up a single constituent, and within that constituent, the first two words (enclosed in the inner pair of brackets) belong together as a constituent of the larger unit. There remain questions about the internal structure of the constituent to which we have just added bracketing, and about the way it fits into the structure of the whole sentence.
Deliberating further about the same example, we might propose that the big ones is another of our phrases. Thus, we progress to (6):
(6) Does she realize [[how heavy] [the big ones] are]?
Until now, the assignment of words to phrases has more or less coincided with the "pacing" or "phonological phrasing" a speaker would give to the sentence when uttering it. Our first intuitions of constituent structure are based on a sense of which words are pronounced together. Places in our sentence at which speakers are likely to pause, for example, are after realize, after heavy, and perhaps after ones. But in many cases, the grouping required by the grammarian will not be supported by normal ways of pronouncing the sentence. Thus, it is not likely that any normal utterance of this sentence would pause after she and pronounce the rest of the sentence in a single run, though a major sentence break between she and the rest of the sentence is the next constituency decision we are going to propose.
Consider first the sentences in (7):
(7) a. Does she [know] ?
b. Does she [see us] ?
c. Does she [want me to] ?
d. Does she [think I'm ready] ?
A generalization over all of these sentences could be captured if we introduced a constituency break between the subject (she) and what follows, and if we then united what follows into a single verb phrase constituent. An analogous structure in our sentence (1) would unite realize with its complement, as in (8).
(8) Does she [realize [[how heavy] [the big ones] are]]?
Deciding on the constituent structure of the phrase the big ones is even less obvious. Pronunciation gives us no help at all, and questions of grammatical structure are subtle and indeterminate. The choices are given in (9):
(9) a. [the big ones]
b. [[the big] ones]
c. [the [big ones]]
We will take (9)c as the one we want, since there are reasonable versions of English grammar which start out from the assumption that a determiner, like the, combines with the remainder of a nominal expression. (See chapter 2.) But we may question this arrangement later on.
Regarding the complement of realize, the structuring that would be generally agreed upon is what we see in (10), where brackets have been added to enclose the big ones are:
(10) Does she [realize [[how heavy] [[the [big ones]] are ]]]?
That is, we accept an analysis in which how heavy is one constituent and the big ones are is the other. In the case of this sentence there are neither phonological nor immediately obvious structural reasons for preferring it to one that does not connect are to the big ones. But we see the phrase how heavy the big ones are as analogous in structure to the bracketed constituents in the following sentences:
(11) a. We aren't sure [[how much] [she knows]].
b. I wonder [[which player] [they gave the ball to]].
c. Do you know [[what] [you're talking about]]?
The construction which licenses all of these phrases is one which allows an interrogative phrase, like how heavy, how much, which player, and what, to appear to the left of a clause in the understanding that it completes the semantic and semantic requirements of the clause remnant which follows it. That is, we can compare the bracketed portions of (11) with the corresponding parts of (12):
(12) a. She knows too much.
b. They gave the ball to that player.
c. You're talking about my future.
The how much of (11)a serves as the object of know; the which player of (11)b is the object of the preposition to; the what in (11)c is the object of the preposition about. Similarly, how heavy the big ones are is analogous to the bracketed portion of (13):
(13) She knows [that the big ones are very heavy].
We will later introduce a Left Isolate construction which presents (among other things) an interrogative word or phrase to the left of a clause which has "needs" or "can use" an expression like the one to its left for completing its semantic and syntactic structure. That is, the left-isolated constituent stands for something unexpressed in the clause-remnant which follows it.
Finally, we will claim that the constituent structure of the entire example in (1) is what is presented in (14). This time we have made explicit the notion that the individual words in the sentence are themselves constituents by bracketing each one. In other words, the bracketing that we originally introduced to mean 'is a group' will now mean 'is a constituent'.
(14) [ [Does] [she] [ [realize] [ [ [how] [heavy] ] [ [ [the] [big] [ones] ] ] [are] ] ] ]?
Although we introduced the bracketing practice as a way of showing the boundaries of a unitized sequence of words, it is used here to indicate constituency, and that includes the individual words. This may be the place to introduce a potentially confusing way of using the term phrase. In general, isolated uses of the words phrase and phrasal will refer to sequences of words functioning as sentence constituents (including the sentence as a whole). Thus, when we speak about phrasal constructions we will mean only those constructions which are built up of more than one word. But when the word "phrase" is modified - as in verb phrase, nominal phrase, or interrogative phrase, it will refer to a constituent type that can often be instanced by either a phrase or a single word. In these usages we are following general practice.
Representation and Constituent Relationships
Bracketed representations like (14) are hard to read and will be replaced in later chapters by graphic representations that make it easier to tell, without the need to count left and right brackets, which strings of words go together. These representations will be of two types: tree diagrams, in which constituents are represented by nodes and the constituents are nodes hanging down from a mother node; and box diagrams in which all constituents are represented by boxes, and constituents in the internal structure of a phrase are represented as smaller boxes inside a larger box. In general, we will be relying on box diagrams for much of what we do, simply because that allows us to write into each box the package of information we need for each constituent type. (Take a quick glance at the later chapters if you can't imagine what this is all about.)
Following modern linguistic practice, we will be using female kinship terminology for speaking of the ways in which constituents are related to each other, and of the ways in which phrasal constituents are related to their parts. In the examples just considered, we will say, for example, that how and heavy in [ [how] [heavy] ] are sisters of each other. The "sister" relationship can hold not only between words, but between phrases as well. Thus, we will also say that how heavy and the big ones are in [ [ how heavy ] [the big ones are ] ] are sisters of each other. It is also possible for one sister to be a word and the other to be a phrase. This is the case for the constituents in [ [realize] [how heavy the big ones are] ].
The whole phrase how heavy can be said to be related to its component constituents, and for that the terms mother and daughter are used. The daughters of the phrase how heavy are how and heavy, and the phrase as a whole is the mother of each of these words.
It deserves repeating that the constituent structure of sentence (1) presented in the bracketing in (14) has to be taken as an indirect claim about the grammar of English. Someone who proposes such an analysis owes us an account of the grammar which assigns properties to each of the imputed constituents and which licenses combination into phrases on the basis of those properties. It will not do, for example, to say that there are some arguments for grouping the big ones as [[the big] [ones]] and other arguments for grouping it as [the [big ones]], and that therefore its structure is indeterminate. The grammar one ends up with will assign at most one of these analyses.
Features and Feature Structures
There are two aspects of linguistic structure that are different in many ways. One of these has to do with matters of constituent structure, of the sort we have been discussing so far. The other has to do with the properties of the constituents that somehow get assembled into higher-order structures.
We group syntactic and syntactic properties together, into an attribute known as synsem (abbreviating syntax & semantics), and the two subtypes, then, are syntactic and semantic, which we will abbreviate as syn and sem.
The synsem features tell us what kinds of constituents we are dealing with. There is another order of features, which we call role or phrasal role, which tell us what function a constituent has within its phrase. Limiting ourselves still to headed phrases, we find that we can assign the major components of a phrase into such roles as head, complement, filler, specifier, and modifier.
Among the syntactic features, we can distinguish head features and level features. Head features are those properties of a head constituent which characterize the "mother" as well as the "daughter". One way of talking about this is to say that the features designated as head features "percolate" from the lexical head to higher phrases, as long as those higher phrases do not have their own head. This accounts for the fact that we say that if a verb is passive, the clause that contains it is a passive clause; if a noun is singular, the noun phrase that it heads is also singular; if a verb is in the past tense, then the clause which contains it is also past tense; and so on. Many of these decisions are paralleled in HPSG grammars. One aspect which (still) separates CG from HPSG is that we regard the identity of the head lexical item in a phrase as a head feature.
Level features, by contrast, are features that indicate the structural level of a constituent. One major level feature is what we call maximality. A maximal constituent is one which is capable of occurring in a "major syntactic constituent" in a sentence; a non-maximal constituent is one which has to be a component of a larger phrase of the same type. The general notion of being a major syntactic constituent is that of representing an argument of a valence set.
The contrast can be most easily illustrated with nominals. A clear case of a non-maximal constituent is a singular count noun, such as bird. The simple word bird cannot occur as the subject of a sentence or the object of a verb or of a preposition; but if it occurs as the head of a specification construction, yielding a phrase like the bird, or my bird, the situation changes. Thus, bird by itself is non-maximal; the bird or my bird is maximal.
Phrases built by nominal modification constructions do not automatically acquire maximality. If bird is non-maximal, so is blue bird or favorite bird. These, too, require specification: a blue bird and my favorite bird are maximal.
Synsem features are divided into syn features and sem features. Syn features are separated into head features and level features. More will be said about each of these below. Sem ("semantic") features include simple features, such as animate, masculine, bounded, etc., but are generally built around what we will be calling frames. A frame is conceptual structure that TO BE CONTINUED
In addition to role features and synsem features, the next important feature type is valence. A valence attribute for a lexical head is a set of feature structures, each of which characterizes the semantic and syntactic characteristics of an argument of the predication represented by the semantics of the valence. The valence of a TO BE CONTINUED
Relational features
Phonological features
Basic Grammatical Categories
The constructions which license particular phrases must specify the kinds of constituents they require. This means that grammatical constituents have to have certain properties assigned to them. The "properties" of "features" (these words will be used interchangeably) that figure in grammatical generalizations are assigned to both lexical constituents (words) and phrasal constituents, and they are of several kinds. The first kind we can call categorial, giving us information about the kind of entity the thing is. The second kind are relational, giving us information about the role a constituent has in a particular context.
Consider the facts in example (15):
(15) a. Cats are lovable.
b. I hate cats.
In both (15)a and (15)b, the form cats is a noun; that characterization is of the categorial type, independent of what the word is doing in its sentence. But in (15)a cats is the subject of the sentence, and in (15)b, cats is the object. This time we are talking about how the word cats relates to its environment; and the characterization of something as subject or object is of the relational type.
Categorial Information
We begin with categorial information. The simplest and most familiar are the traditional "parts of speech" as these are assigned to individual words. But in many cases, to say that something is a verb or a noun is not enough, since there are subclasses of these parts of speech that have to be known in explaining their distribution and behavior. In the case of the nine words in our sample sentence, we might say the following:
(16) Categorial features.
does verb; auxiliary verb
she noun; personal pronoun
realize verb
how adverb; interrogative word
heavy adjective
the article
big adjective
ones noun; indefinite pronoun
are verb; copula; auxiliary
For consistency, the word "noun" is used here for any kind of nominal constituent. Traditionally, the word "noun" is the usual name for a word which is a nominal but which is not a pronoun.
The main available category labels will include verb, noun, adjective, and preposition among the clearest cases; to these we will find reasons to add others from time to time - quantifier, adverb, article, demonstrative, etc. It is clear, however, that many of these words will need to bear more than one label, i.e., will have to be assigned more than one grammatical feature. Some of the additional features are subtypes of the main features. (Thus, auxiliary is a subtype of verb, pronoun is a subtype of noun, etc.). But sometimes the categories associated with a word are cross-cutting, such as what we see in (16) like thus or slowly, how is an adverb, but unlike thus or slowly, it is also an interrogative word; who and which are interrogative words, unlike how, but they are not adverbs.
Phrasal or multi-word constituents have grammatical properties, too. At the largest level, our example is a sentence; it is also a clause; as a clause it is an interrogative clause; as an interrogative clause it is an inverted clause (an auxiliary precedes the subject).
The constituent how heavy the big ones are is also a clause, and it is also an interrogative clause, but it is a "WH"-clause (its first element is an interrogative phrase).
Since many clauses have properties that derive from their constituency, and are named with reference to the nature of their phrasal combination, we can turn to the relational notions here.
Relational Features: Phrasal Roles
So far we have been talking about features of constituents that merely represent "what they are": we turn now to questions of "what they are doing where they are". We will later return to more properties of the former kind, but especially properties that determine, or are sensitive to, the kinds of relational features we are about to examine.
The phrasal constituents that we encounter can in the first instance be divided into headed phrases and non-headed phrases. Non-headed phrases are more or less symmetrical, in the sense that there is no daughter constituent that determines the character of the phrase as a whole. In the phrases that we will consider here, one daughter constituent more closely determine the phrasal type of the mother than the other or others. That is the constituent which we will call the head of the phrase.
The head of the verb phrase [ [realize] [how heavy the big ones are] ] is the verb realize, and the remainder is its complement. We can refer to this as a head-complement phrase, and the construction which licenses it as a head complement construction. Other head-complement constructions place noun complements after nouns (fear of lightning), complements after adjectives (afraid of lightning), and complements after prepositions (in the rain). We will see later that heads determine what kinds of complements they welcome, and the mechanism by which that is achieved will be discussed under the heading valence. In the realize example, there is only one complement and the phrase has a binary structure; but a lexical head can require more than one complement, as in sentence (17), where there are two complements in the verb phrase headed by showed.
(17) The plaintiff [[showed] [the photographs] [to the judge]].
We will temporarily regard the basic structure of the whole sentence as a three-part verb-headed phrase in which does is the head. Thus, [[does] [she] [realize how heavy the big ones are]].
Returning to the conventions of using the word phrase, we can notice that there are verb phrases which are not head-complement phrases, as in (18)a, and there are verb phrases which are not headed phrases, as is the case with the bracketed constituent in (18)b.
(18) a. The defendant [disappeared].
b. The lawyer [stood up and screamed].
(In the case of (18)b, the substructure of the verb phrase contains two verb phrases, but the constituent as a whole is a conjunction, hence non-headed, rather than a headed phrase.)
The phrase [ [big] [ones] ] is a noun phrase, and its head is ones, which we have classified as a noun. This is an instance of a modifier-head phrase, licensed by a modifier-head construction. Other types of modifier-head phrases include adverbs modifying adjectives, such as very cold. Since the naming convention we are adopting for English phrasal constructions follows the left-to-right order of the constituents, we could add head-modifier constructions to our list of modification patterns; examples of head-modifier phrases would be noun phrases like man in the moon or milk that she drank.
Not all noun-phrases are phrasal, as in (19)a, and not all phrasal noun-phrases are headed, as seen in (19)b.
(19) a. [You] were wrong.
b. [You and I] have got to talk.
The phrase [ [how heavy] [the big ones are] ] is a filler-head phrase, from the fact that the initial adjective phrase how heavy "fills" or satisfies a requirement of its head, since the verb are does not have its complement locally satisfied. That is, the verb be (here in the form are) "asks for" something to complete its function; to say that what it needs is not locally satisfied means that it does not appear as a right sister of are in a head-complement phrase.
In the phrase [ [the] [big ones] ], the word the is a determiner and big ones is the head. Following a common tradition, we will generalize the notion of determiner to a broader notion, specifier, and will describe the phrase as a specifier-head phrase, in which the specifier is the and the head is big ones.
Eventually, certain claims will be made about the differences between the specifier-head and the modifier-head patterns. Until then we can leave unsettled the question about whether the phrase [ [how] [heavy] ] should be treated as a modifier-head phrase or as a specifier-head phrase.
Within phrases, then, we can identify types of relationships holding between the constituent members of a phrase. These can be represented by features indicating phrasal roles. All of the phrases we have discussed so far have one constituent whose role is head. The others are modifier, specifier, filler, and complement.
Further Relational Properties: Grammatical Functions and Semantic Roles
It is important to repeat that the role features just discussed are necessarily context bound. To say that something is a head is not to say what it is, but what its function is in a particular linguistic context. Nothing is a head, or a complement, in isolation. To call something a head, or a complement, is to speak of its role within a phrase.
There are two additional kinds of relational notions that we need to recognize in grammatical theory, and these are grammatical functions and thematic roles. The way we use these words, they will be assigned by properties of valence sets. In a sentence like (20)
(20) Jimmy broke the balloon.
we will describe Jimmy as the subject and the balloon as the object, but OF WHAT? They are the subject and the object of the verb break (here broke), but it would also be common to say Jimmy is additionally the subject of the whole sentence.
The grammatical functions of a verb are represented in something we call a valence set; a valence set is initially associated with the verb, but since it shapes the basic syntactic and semantic structure of the whole sentence, it is "transmitted" or "projected" to the sentence as a whole. For a simple clause like the one in (20), to say that Jimmy is the subject of the verb is the same as to say that it is the subject of the sentence, since Jimmy as subject is a member of the valence set which belongs to both.
Repeating: when a verb is found in a sentence, it has brought with it, as a part of its lexical entry, a valence set identifying and controlling the constituents that may or must co-occur with it in the sentence. This valence set is projected to the "mothers" and "grandmothers" of the verb, up to the level of the sentence.
We will also say, with respect to the sentence's meaning, that Jimmy is the agent of the action of breaking the balloon, and that the balloon is the patient of that action. These terms, indicating the semantic role of each element with respect to the action, are also specified in the same valence set; and since the action of breaking is associated with the verb as well as with the entire sentence, once again we have this apparent ambiguity in respect to what these things are agents and patients OF.
(In referring to the subject as "Jimmy" and the agent as "Jimmy" reflects the idea that in the one case we are talking about a linguistic entity, the noun-phrase Jimmy; in the other case we are talking about the person Jimmy as the agent of the activity.)
The essential difference between phrasal roles on the one hand and the grammatical functions and semantic relations on the other hand is that in the former case, the relations are between sisters in a phrase, the latter are relations between a valence set and elements which might be represented anywhere in the sentence, or, as we will see later, elements which are not actually physically present in the sentence at all.
Auxiliaries
An auxiliary requires a subject and a complement. Two of the words in sentence (1) are classified as auxiliaries, does and are. In the case of does in our sample sentence, the subject is she, and the complement is realize how heavy the big ones are. But from what was said earlier, the word she functions in one sense as the subject of does and in another as the subject of the whole sentence. The subject of are is the big ones, which is simultaneously the subject of the inner (and incompleted) clause the big ones are. and of the whole (interrogative) clause how heavy the big ones are.
But she in this sentence, while directly instantiating the subject of does, is also indirectly the subject of the verb realize. We will say that she directly instantiates the subject requirement of does (this is the verb onto which she induces grammatical agreement - we get does rather than do), but that she indirectly instantiates (or co-instantiates) the subject requirement of realize.
(Eventually we will want to recognize the big ones as the indirectly instantiated subject of how heavy, but questions about the subject relation for adjective phrases can be postponed.)
The grammatical functions that we will need to recognize, for English, are subject (already discussed), but also object. For valence elements which are neither subjects nor objects, we will use the term oblique to cover them all. In a sentence like
(21) I showed the photographs to the judge.
we will say that I is the subject, the photographs is the object, and to the judge is an oblique. A reason for saying "the subject" and "the object" but "an oblique" is that a grammatical English simple clause contains exactly one subject, at most one object, but any number of obliques.
In the following sentence the obliques are bracketed:
(21) We drew the picture [yesterday] [on the blackboard] [with colored chalk].
REWRITE NEXT PARAGRAPH, OR RESHAPE THE REST OF THE CHAPTER WHICH GIVES DISTINCT TERMS FOR COMPLEMENT ROLE, VALENCE ELEMENT, VALENTIAL OBLIQUE
We have introduced the term complement as the name of a phrasal role, for cases in which we have a lexical head followed by a complement, where the word stands for a phrasal role. In addition to the remarks made above about sentence (21), we can notice that the verb phrase drew the picture yesterday on the blackboard with colored chalk is a head-complement construction with head drew and four complements. There are three (3!) other uses of the word complement that we need to be careful about. Some people use the word to refer to anything in a word's valence set, thus including subjects (which generally do not appear with the verb in a head-complement phrase); and some people use the word to refer specifically to those valence elements which are obligatory, reserving the word adjunct for those that are optional; and there is a use of the word that is limited to verb phrases or clauses, rather than noun phrases or preposition phrases, that serve as members of valence sets.
Semantic Relations
The third important kind of relational notion has to do with semantic relations, of which we have so far considered agent and patient. This part of grammar, and the problems connected with it, will be dealt with in detail in the chapter on valence. Suffice it for now to make it clear that semantic relations do not correspond one-to-one to grammatical functions; that is, they are not definable in terms of the semantic interpretations of specific grammatical functions. This contradicts some traditional definitions of subject and object, whereby the subject is defined as the name of the individual who does something and the object is the name of the thing that has something done to it.
The lack of correspondence between grammatical functions and semantic roles can be illustrated by considering contrasts and distinctions in the following two sentences
(22) a. Fred delights in his children.
b. Fred delights his children.
There is a semantic role called experiencer, identifying the locus of a psychological state or process, which is instantiated by the subject of (22)a and the object of (22)b. (In our example sentence, she is the experiencer of a "realizing" predication.) We also notice in (22) that the subject in (22)a does not have the same semantic role as the subject in (22)b.
We will reserve until later a discussion of the semantic roles of the non-experiencers in (22).
Other Morpho-Syntactic Features
Once we recognize the existence of grammatical relations we are ready to see the need for further descriptions of the grammatical properties of the words in our sample sentence. Many of these new features render the difference between categorial and relational features a bit confusing. Characterizing she as nominative is saying something true about what kind of a word she is; but at the same time we have to recognize that she is the kind of a word which is limited to particular linguistic contexts. We are dealing here with categorial properties which are reflections of relational properties.
The needed kinds of additional information can be suggested by the following table:
(23) Further properties of words.
does present tense (compare with did) singular (compare with do)
third person (compare with (I) do)
she singular (compare with they) nominative (compare with her)
feminine (compare with he) third person (compare with I,
you)
realize bare infinitive (compare with realized, realizes)
the definite (compare with a)
ones plural (compare one)
are present tense (compare with were) plural (compare with is)
third person (compare with am)
Many of the properties just illustrated are reflexes of the word's context, but they are nevertheless properties revealing something about what the words are. The properties belong to grammatical feature sets called case (the nominative feature of she), tense (past versus present in verbs), number (singular versus plural), person (first person [I, we], second person [you], third person [the rest]), verb form (finite, infinitive, gerund, etc.), and a few that we'll be making up on the run.
Lexemes and Word Forms
What we have been referring to so far simply as the "words" in sentence (1) are mostly variants of more abstract entities, and for this we use the term lexeme. The concept lexeme is an abstraction that allows us to say that the lexeme be shows up in sentences in many word forms. The word forms that realize be are am, is, was, were, be, and being. The lexeme do shows up in our sentence as does. The lexeme realize shows up as realize.
In English the name we conventionally give to a lexeme is one of its word forms; in the case of verbs, it is the infinitive or bare-stem form. In the case of nouns, it's the singular form; in the case of pronouns, the nominative form, etc. (The lexeme she has word-forms she, her, hers, for example.)
A lexical constituent will have to specify both a lexeme and a word-form. We will associate the valence sets, the meanings, and the major categorial features to lexemes. Then when lexemes are realized as particular words, the word-forms chosen will bring with them information of the kind we have just reviewed. The result is that any word in a sentence will simultaneously represent a lexeme and a word-form, each bringing its own requirements. (In some cases, namely the invariant words, there will be no distinction between lexemes and word forms.)
Combinatorics and Argument Structure
In connection with what we can call the combinatorial properties of words, we have already spoken of valence. The combinatorial requirements of a word will often specify constraints on the grammatical realization of the arguments of the word. Such constraints can be expressed in terms of grammatical functions and thematic roles of its arguments, as well as their grammatical form. Of the verb realize, for example, we might say that it requires the presence of a constituent identifying the experiencer of a particular kind of mental state or process and the content of that state or process; the experiencer is capable of realization as the subject of a sentence headed by realize, and the content is capable of realization as its object. The object can be expressed as a noun phrase (he realizes his limitations) or clausally, either as a that-clause (that she loves me in she doesn't realize that she loves me) or as an interrogative clause (what time it is in do you realize what time it is?).
(In the final story we will need one more distinction in the lexical realm, treating the fact that words can be used in multiple meanings, and that different senses of a word might have different valences. For example, there is a second meaning of realize seen in an expression like to realize one's potential; in this sense the word has no clausal complements.)
The valence of a word, then, is the description of the combinatorial potential of the word. Since there is some redundancy or predictability in respect to the co-occurrence of grammatical functions, semantic roles, and syntactic form, it is possible to take advantage of that by distinguishing the minimal valence of a word, that specification of valence information from which the rest can be predicted, and the full valence, a description which identifies the relational features of all of the valence elements in a given sentence. The generalizations which account for the pairing of semantic roles with grammatical functions make up an area of grammatical description known as linking.
The Auxiliary DO
The valence of the auxiliary word does in our sentence is fairly complex. A form of the verbal lexeme do, it is an auxiliary which requires a subject and a complement, the subject is assigned no semantic role by this verb, and the complement must be a verb phrase headed by a non-auxiliary verb, in its bare infinitive form.
DARN. SOMEHOW I SHOULD HAVE MADE IT POSSIBLE FOR ME TO USE THE WORD "COMPLEMENT" APPROPRIATELY IN THIS DESCRIPTION. IN OTHER WORDS, I SHOULD REWRITE THE EARLIER DISCUSSION ABOUT THE WORD SO THAT I CAN FREELY REFER TO COMPLEMENTS AS BEING (NON-SUBJECT ONLY?) VALENCE ELEMENTS.
That is, it shares with auxiliaries like will or can in taking the bare infinitive form in its complement, but will and can are not limited to complements headed by non-auxiliary verbs. Thus we can say
(24) a. Does she understand? [understand: bare-stem of non-auxiliary]
b. Will she understand?
c. *Does she have understood? [have: bare-stem, but auxiliary]
d. Will she have understood?
It shares with the passive auxiliary the requirement that its complement has to be headed by a non-auxiliary verb, but the passive auxiliary requires a complement headed by a verb in past participial form. Thus:
(25) a. Did you break it?
b. Was it broken?
c. *Did you broken?
d. *Was it break?
And do is necessarily finite: it can only be the first verb in a verb chain.
(26) a. She will be eating it.
b. *She will do eat it.
c. She has eaten it.
d. *She has done eat it.
In short, the valence requirements of do are absolutely unique. It cannot be preceded or followed by another auxiliary.
Some distributional constraints on words are not typically thought of in terms of valence. The adjective heavy is interesting. It is a scalar adjective, like old, large, long, tall, etc., and is thus able to occur with specifiers that indicate the scalar extent of the thing being measured. All of them, for example, can use extent indicators of the kind so, this, and how: how tall, how heavy, how old; so large, so long, so heavy; this tall, this heavy; etc, But adjectives for the other unilinear scales also allow qualification of the extent with a precise measurement expression, but heavy does not. That is, we get (27)a-c, but not (27)d.
(22) a. three inches long
b. six feet tall
c. four years old
d. *two pounds heavy
Semantics
Semantic information can also be built up from the semantic descriptions of the constituent elements of a sentence. The key word in this sentence is realize. This word has the property of being factive in that its use presupposes the truth of a that-clause complement, or it presupposes the known answer of an interrogative clause. Furthermore, it conveys a sense of the importance of the presupposed information. Thus a sentence like (23)a is merely a statement of my anonymity in this environment; but adds more than that - a presupposition that I am important, or that their reaction would be quite different if they knew who I was.
(23) a. Nobody here knows who I am.
b. Nobody here realizes who I am
"Flat" semantics
((just stabbing in the dark))
What's involved in the interpretation of this sentence?
Speaker is asking hearer
about a female person SHE
knowable in the context
whether the hearer knows as true or false
the proposition that SHE knows
the value of the function
the weight of THE BIG ONES
which are plural members of a set of things
whose nature is knowable in the context
which exhaust the complement in the set of a subset of things
knowable in the context
under the speaker's assumption
that it would be important to her
for her to know
the weight of THE BIG ONES
terminological and notational preliminaries
What conceptual and notational tools do we have for doing it?
structure, construct, construction
licensing of constructs with constructions
feature structures and unification
unification of constructions
syntax, semantics, pragmatics
role (head, etc.), rel
grammatical relations
thematic relations
head features
level features
maximality
srs
regularity vs. idiomaticity
generalizations and exceptions
examples wrt determination
introduction to NP
determination
modification
quantification
problems
exceptions vs. counterexamples
another 3 pages, this many pages
a lot of people, lots of mud
semantics
flat semantics
constituency-semantics discrepancies
frames, etc.
feature architecture
principles
valence
combinatorics in general
sets, lists, avm's
linking
valence variation
construction types
headed, non-headed
types of non-heads
non-binary constructions
inheritance and hierarchies
representation of constituency
basic constructions of English
phrasal:
SP, VC, DN, MN, NC, PC, AC
LI, Inv
valential:
linking constructions
extraposition
complex dependencies (wxdy, etc.)
Given that (1) must have "two-story" constructions for, e.g, count >> mass and for phonologically marked derivation (and inflection), e.g, present-participle, (2) many languages need two-story constructions for linking (passive, applicative, causative, etc.), because these valence constructions are morphologically marked, and (3) English *doesn't need* two-story constructions for valence filling out (linking) because these processes are not morphologically marked, should we have 2 kinds of lexical-item-affecting constructions in English (one- and two-story) or just make them all two story?
We talked about this a little last time we were together and I believe you favored not having two-story constructions in English unless you had to override something or add morphology. I kinda liked Orhan's argument for his kind of CG morphology that it treats zero morphology just like non-zero morphology, or rather as a special case of non-zero morphology, in which phon = input-phon. Not having the 2-story constructions except when you need to override something or change the phonology smells a little like technical artifice to me. In fact in the ABC lecture notes, we've gone all two-story. (In the intro we talk about the French inchoative: eteindre 'extinguish' >> s'eteindre 'go out'.)
It seems to me the idea of restricting overriding to the lexicon is most directly reflected formally by having all lexical constructions 2-story. But this is just a feeling for which I've provided no argument, I admit.
pAS you draft a new chapter 1, I've been trying to wonder about what to do with existing chapter 2. As part of this attempted wondering it has occurred to me that existing chapter 1 breaks naturally into two parts, the part beginning with 1.3 Grammatical constructions begins a story which is continued in Chapter 2. So, I'm thinking about a new chapter 2 which starts with existing 1.3 and goes through existing chapter 2.
Don't feel like you have to respond to this now, I just thought I'd mention it in case it could improve your weekend.
Are you going to put some version of the new feature geometry in the new chapter 1? Of course it wouldn't have to be put in AVM format; that would be introduced later. But if the feature geometry could be used as a way of introducing the basic feature notions, I think it would be nice.
p
How about this as a solution to the old MN problem?
Since we anyway have negation in "~DA", "~subj", etc., just let the MN const. stipulate that its right da is "[max ~+]".
[If you say that you suggested this eight years ago (and once a year since) I'll believe you.]
p