This exercise shows how difficult it is to decide in advance the outcome of a piece of research. What began as an exploration of the devices for introducing sentence constituents that give the "settings" of the scenes that are built up in the sentence's semantic core, ends up exploring a device for expressing something much more general.
So far we've been looking mainly at situations where the means of introducing frame elements into a sentence are either the valences that come "off the shelf" with the verbs we choose, or those augmented valences that are created through special argument structure constructions with their own specific accompanying semantics.
In connection with the V-Comp construction (formerly PVP) we learned that the valence of a verb is to be taken as a subset of the valence of its "mother" V-Comp construction. One of the ways in which there can be more valence elements in a VP or in a clause than are found in a lexical valence, is through "adjunct" constructions.
Whether the notion "adjunct" needs to be a primitive term in our grammar, for example as a special kind of "role" name, is still up for grabs. We do need to be able to distinguish these from other comps, but that can also be done by distinguishing those comps that carry "setting" semantics from those that do not. For the time being, at least, we can think of them as merely fitting the "comp" slots in the V-Comp construction. We can then say, at least, that those comps of a V-Comp that are semantically designated as "settings" participate in certain distributional patterns,for example the ability to occur in initial position in a sentence. (The third and fourth examples below suggest that if there are both place and time adjuncts in the same sentence, there is a preference for fronting the time adjunct.)
The most obvious setting adjuncts are those of time, place, and condition. If the rest of the sentence describes some happening, these adjuncts provide the temporal or spatial settings for those happenings, or the "hypothetical space" which "contains" those happenings.
Adjuncts of setting can be expressed as
The semantics of a setting adjunct will have to be incorporated into the semantic structure of the VP that contains it, and ultimately the clause as a whole.
It is important to notice that not every expression of time or place in a sentence is a setting adjunct; some of them are complements. The verb "live" (except in certain contexts where it means 'not to be dead') seems to require either a time specification or a place specification. Consider:
Sometimes these predicates participate in the main semantic predication of their clauses, as in the following simple sentences.
The cases that will interest us here are often called secondary predicates. In some cases the secondary predicate is a necessary part of the valence of the verb, as in the next sentence, where Pat controls or coinstantiates the subject of the NVP "a very good friend."
But sometimes a sentence could be ambiguous according to whether a particular adjunct is controlled by some argument in the verb's valence, or whether the adjunct identifies some setting feature of the scene as a whole. For a sentence like
We need to be more careful than simply to say "the rest of the clause", because a sentence can have more than one setting adjunct. In fact, we need to distinguish, in the semantics of a sentence, the "settings" from the core semantics. (This will get more complicated when we consider conditional sentences.)
The kinds of constructions we called Head+Comp (or HComp) were phrases built up out of a lexical item and those elements that count as its complements - in the clearest case, those elements that satisfy its valence requirements. The categorial types of HComps are named after their head constituent: thus
Whereas the HComp constructions have one constituent that is a lexical item, the SpecHead constructions have heads that are potentially phrasal. In other words, the Head of a SpecHead construction could be a single word, but it could also be an HComp construct.
The only SpecHead construction we have considered so far is the one we called Determination. Where we originally had [role determiner] for the first constituent in the Determination construction, we ended up substituting [role specifier].
Let us recall that
Some years ago, I believed that English had postpositions as well as prepositions. One of the words that I thought was a postposition was "ago", as in "three years ago", "a very long time ago", etc. It was a word that occurred by itself as a constituent in its phrase, and it was clearly the head of its phrase. The kind of object an "ago" phrase is is determined by the kind of word "ago" is, rather than by the kind of phrase that accompanies it.
This last fact distinguishes "ago" from, say, "else". In the case of "nobody else", "somewhere else", or "when else", the syntactic type of the phrase is determined by the first member of the phrase, not the second. So, whatever else "else" is, it is not a "head".
Although most of the words that characterized the kinds of phrase they formed occurred before their phrasal companions, there were a few that did not, I thought. In addition to "ago" there was also "away" (many miles away), "hence" (several years hence), "notwithstanding" (your unwillingness to cooperate notwithstanding).
The conclusion we will reach is that these phrases are instances of SpecHead constructions.
In examining a phrase like "two years ago" we can agree that it has two constituents, "two years" and "ago". A construction grammar analysis requires us to ask the following:
The answer to the first question is that the phrase identifies, as the time of the event expressed by the rest of the sentence, a point or period two years earlier than the time at which the utterance is spoken. There are three parts to this description: it identifies
The answer to the second question is that the word "ago" identifies a temporal LM (the present moment) and a temporal DR (in the past), and the MG phrase "two years" is the measure of the distance between the time referred to and that of the LM.
The answer to the third question is that the phrase creates a Reichenbachian temporal schema in which both the reference time ("R") and the event time ("E") are together and precede the speech time ("S"), the distance between them being a span of two years.
We can express this with the formula
where R and E are together, and both are earlier than S.
This is the same formula that was needed for the simple past tense, and so the temporal schema introduced by this phrase must unify with the temporal schema that matches the tense of the sentence. Hence something like "He bought the car two years ago" is semantically well- formed.
The simple present tense has the formula S/R/E. Since this conflicts with the E/R < S of the "ago" phrase, a sentence like "*She lives here two years ago" is not licensed.
The tempora schema for the present perfect is E < R/S, which conflicts with E/R < S, and so "*He has seen her two years ago" is also not licensed.
The temporal schema for the pluperfect is E < R < S, which also clashes with the E/R < S of the "ago" phrase, and so "*He had met her two years ago" is also not licensed.
(The last example is bad in ordinary conversational English. The combination is possible, however, in certain kinds of third-person narrative that expresses a central character's "point of view". The form is also permitted in the so-called counterfactual conditional: "if you had been here two days ago".)
But now we must consider the other two questions. We can approach them by recalling that our phrase has two constituents - let us call them "A" and "B". The "A" part is what we can call the "magnitude (MG) phrase" (two years, a long time, several hours, etc.); and the "B" part is the "LM and DR" part.
To ask about "A" is to ask what other phrases can have the same function. The first examples one might thing of are those which, like our starting example, is a quantifier plus a plural noun. The nouns most likely to occur in such phrases are the names of conventional temporal units (month, week, year, day, hour, minute, millenium, decade, etc.); the quantifier can be a numeral (three, seventeen, etc.) or an informal quanitifier of countable elements (several, many, just a few etc.) Thus, several years, many months, four hours, two centuries, etc. The heads of these NPs are units of measurement for temporal magnitudes.
Many of the nouns in question have two uses, which we can refer to as calendric and metric. In particular, the words week, month, year, decade, century and a few others, can refer to anchored temporal cycles or to measures of time lengths. The difference can be seen in the different ways in which we interpret the phrases "within the month" and "within a month". If on the 20th day of a month I promise to finish some project "within the month", it will be completed before the first of the next month; but if I promise to finish it "within a month", it will be completed before the 20th of the next month. The former is the calendric use; the latter is the metric use. In our "A" phrases, these words are given their metric senses.
In addition to phrases with quantifiers and nouns, the "A" constituent could also be satisfied with simpl plural nouns (that happened years ago, hours ago, centuries ago, etc.). The implication is that the number of years, hours, or whatever is great.
Let us refer to the above types as basic temporal magnitude expressions.
When serving as the MG phrase in our construction, with "ago", the interpretation is that of designating an extent of time, whereas in the alternative contexts, they do not. Thus, if I claim that I spent three summers in Banff, they do not need to be successive summers, but if I say that something happened three summers ago, the period in question contained just three summers.
As a kind of figurative extension from examples of the last type, the construction also permits nouns that are metonyms of event types that are construable as occurring with a certain cyclicity. Thus
Lastly, there are certain words or phrases which on their own identify spans of time in this context. The phrase "a long time" (as in "a long time ago") can be used in other contexts as well, but the word "long" is idiomatically associated with this construction. ("long ago")
(In other contexts, the word "long" is a negative polarity item: "were you there long?", "this won't take long", etc.)
The semantic role of the "B" part is to identify the landmark and the direction. In the case of "ago" and "hence" the LM, built in to the meaning of the word, is the present moment, the "S", the speech time, and the DR is toward tne past for "ago" and toward the future for "hence".
The "B" part can also be phrasal, in which a preposition indicates the direction and the complement of the preposition indicates the LM. In the simplest case, the prepositions are "before" and "after". The complement can either be a noun-phrase designating a time point, or a clause indicating an event whose time of occurrence is taken as the LM. Hence "three days before the general's death" and "three days before the general died".
In some cases the LM can be pragmatically given; for this we can use the label prag. For certain single words, the LM can be some time point understood pragmatically in the context. this is true, for example, of "before", "afterwards". That is, in interpreting a phrase like "two years before", or "some time afterwards", the interpreter needs to recover the LM in the preceding discourse. (In the case of "before", since it can also occur with its object, we can speak of a "dni" interpretation; since "afterwards" does not allow an explicit complement, we can either say that there is an obligatorily omitted (under "dni" conditions) object, or - more reasonably - that the LM, with "prag" interpretation, is built into the semantic structure of the word.
The word "long" as a specifier in this construction has more than one property that makes it idiomatic. In particular, as we remarked earlier, in other contexts temporally construed "long" is a negative polarity item (I won't be long, *I'll be long), but in the context of our construction it is not. As an "ordinary" temporal extent phrase its appearance is quite idiosyncratic. For example, it's welcome after "for" (We won't be at it for long") but not after "in" (*We won't finish the project in long"; but it occurs after "take" (This won't take long") but not "spend" (*We won't spend long on the project).
AGO: max -, no comp, LM now, DR past
By making "ago" nonmaximal, we guarantee that it has to occur with a specifier. It has no comp, since the LM ("S") is built into the meaning of the word.
RECENTLY: max +, no comp, LM prag, DR past, MG small
SOON: max +, no comp, LM prag, DR future, MG small
BEFORE: max [ ], dni-omissible
comp, LM prag,
By making maximality unspecified with "before", we make sure that it can occur either with or without a temporal extent specifier; by making the comp dni we allow the comp to be omitted under conditions of pragmatic givenness. If there's no specifier, then distance indefinite.
AFTERWARDS: max +, no comp, LM prag,
The feature [max +] means that the word "afterwards" can be an "AB" construct on its own. It has no comp because the [LM prag] feature is built into the meaning of the word.
EARLIER, LATER: max [ ], no comp? ("than
that" is okay),
LM prag, DR past & future resp.
These words can be accompanied by specifier phrases ("much later", "two years earlier"), but do not need to be. There is no comp because [LM prag] is built into their meaning.
The constructs just reviewed can be referred to as temporal setting constructs. But there are spatial setting constructs that have the same essential syntactic and semantic structure. That is, they have an initial phrase which gives the distance in space (MG) and a second phrase which identifies a spatial LM and a DR.
The directions, in time, are limited to past and future (earlier than vs. later than), but in three-dimensional space there are possible ways to indicate directions. Here is a sample of the variety of expressions of the "AB" type, paired with parallel temporal phrases.
Spatial specifiers go with spatial heads, temporal specifiers go with temporal heads, and so the unification process by which their combinations are licensed is going to be sensitive to a "dimension" feature: [dimension space] vs. [dimension time]. Thus we would predict that something like "*a few minutes below our feet"and "*a mile after Tuesday" are not licensed.
But it's not that simple. There is one conceptual area in which time and space are coordinated, and metaphors are possible in both directions.
These phrases can, of course, be interpreted if we are talking about travelling. Notice that if there is conflict between specifier and head, the head wins out. That is, "two miles ago" refers to a (past) time, and "three hours north of Minneapolis" refers to a place. Thus, e.g., "*My house is [present tense] two miles ago" is no good: it can't mean that my house is located at the place where we were in this journey two miles back.
First we thought that we were dealing with a temporal setting construction. Then we noted that the same pattern is found with spatial setting expressions, so we assumed that we were dealing with a more abstract construction, inherited by each of these, that could be thought of as a dimension- free "setting" construction. But now we are going to see that there are other constructions in English which have nothing to do with settings, but which do have the internal semantic structure of LM-DR-MG, and we may need to devise an inheritance system which is more general than what we earlier had in mind.
One of these is the comparative construction. This construction can bring any scalar dimension at all. A comparative phrase introduces a LM (the object of "than"), a DR (determined by the choice of adjective in the comparison), and optionally includes a MG. Examples:
Again, the units of measurement in the "A" constituent have to be in the same "domain" as that of the adjective: units of time for age, units of spatial extent for linear dimension adjectives, units of temperature for warmth, etc.
There is one more construction type capable of being the "B" constituent in for this Magnitude Specification Construction. The positive, or "unmarked", member of a pair of polar adjectives can be used for indicating extent from zero. Thus, "three feet tall", "six years old". Some adjectives do not welcome this: "*twenty pounds heavy", "*thirty degrees warm". LM is the "zero" end of the scale.
First, if the attribute "domain" can be given such values as time, space, temperature, age, intelligence, i.e., any of the dimensions which support scalar relations, we will have to say that the domain must be shared across all constituents. The units of measurement for the MG must be of the same type as that of the phrase as a whole. Thus,
Figure 1: "domain" unification
We need to note that the "A" constituent is the specifier and the "B" constituent is the head.
Figure 2: "Spec/Head"
Next, we need to notice that the construction is free to be of any of several categories (especially PP and AP). We could represent this, if necessary, by a kind of feature decomposition of grammatical categories; but since the semantic descriptions will never be appropriate for verbs and nouns, we can just as easily simply say that the category is unspecified. Further, the mother constituent is [max +] and the head daughter is [max -]. The reason for this is that (1) the construction does not reiterate, (2) some heads are specified as [max -], meaning that they must be accompanied by a specifiers (e.g., "ago", and a temporal "from" phrase, as in "two years from now"), (3) some words or phrases operating in this domain can occur either by themselves as expressions of the type of the mother constituent, and they will be [max [ ]] ("before the revolution"), and (4) some words having the semantic features of this construction are marked [max +], meaning that they cannot occur with MG specification (e.g., "recently").
Figure 3: Category and Maximality
The semantics of this construction has to specify the three components we have been observing, since these are expressions that depend on recognizing a landmark, a direction, and a magnitude. The "A" constituent provides the MG, and the "B" constituent provides the LM and the DR. The name LOCATION for the frame is not felicitous: it is supposed to cover the cognitive act of identifying the position of something on a scale, given a landmark.
Figure 4: The Phrasal MG:LM+DR Construction
The "B" constituent can be filled with a single word (e.g., "ago", "tall"), but it can also be phrasal. The two types we have noticed are those seen in "before the election" and in "older than Harry". I will ignore the structure of comparative construction and mention only the PP version.
Figure 5: The Phrasal LM+DR Construction
But if you get bored waiting and watching, you can try to draw a box, following the feature-architecture proposals you learned about (click here for details), which puts all of the material in Figures 1-4 together.