As discussed in 3.2, the syntactic metalanguage used in the FrameNet project is intended as a framework for lexical description-i.e. to describe the syntactic valence properties of individual lexical items. It is not intended as a framework for the complete syntactic description of sentences. In choosing the phrase types and grammatical functions, the major criterion was whether or not a particular label might figure into a description of the grammatical requirements of one of the target words. Our goal is to annotate words or phrases in a sentence that are either in direct grammatical construction with the target word, where this notion is extended to include both extracted, extraposed constituents (cf. section 3.2.7), or that are dependents of higher commanding predicates that are construed as participants in the target word's frame (cf. section 3.2.5).
Initially, the emphasis of FN annotation was on what was most relevant to lexical descriptions, namely the core and peripheral frame elements of target words. Accordingly we limited ourselves, for the most part, to those phrase type labels needed for the annotation of such elements. Over time, annotation experience required broadening the range of elements we annotate to include extra-thematic frame elements. Such frame elements evoke frames that are distinct from the one evoked by the target, and typically embed the target frame in a larger scenario (cf. section 8). While we find the inclusion of extra-thematic elements in our annotations valuable from a lexicographic point of view, and in fact necessary for the annotation of full texts, the concomitant introduction of additional phrase types has led to some inconsistencies in the number and type of distinctions made between phrase types. (These will be pointed out below). In the future we may revise the inventory of phrase types to make it more compatible with what would be desirable for theoretically-oriented syntactic descriptions.
Finally, note that there are two cases in which frame element labels are not paired with grammatical function and phrase type labels. First, although certain types of noun targets can sometimes bear frame element labels, they are never assigned phrase type or grammatical function tags, as is shown in Figure 4. This policy applies mainly to relational target nouns such as brother and mother, and to artifact and natural kind-denoting target nouns such as building and forest. Likewise, frame elements annotated on any other layer than the first FE layer are never assigned grammatical function and phrase type labels (cf. 3.2.4 on frame element conflation).
What follows is a list of phrase types that are used in FrameNet, accompanied by some examples. The phrase types are discussed and exemplified in greater detail in sections 4.2 and higher. Phrase types are assigned automatically by the FrameNet desktop software during the annotation process, but may require manual correction.
Non-referential noun phrases, also called expletives, such as there in There was a row and it in It was raining, are not assigned frame element labels and consequently cannot have phrase types either. Such constituents are marked only on the Other layer.Possessive Noun Phrase (Poss)
The type possessive noun phrase includes both the possessive determiners, shown in (1), as well as noun phrases with the 's-genitive clitic, shown in (2).
Nominals that are not referentially complete, especially in compounds, are given the phrase type label N. Such nominals may themselves be internally complex (cf. (3)).
(3) [fast food] allergy (4) [car] manufacturer
Referentially complete noun phrases that could fill core verbal argument slots are assigned the label standard NP, as in the examples below.
(5) [My neighbor] is a lot like my father. (6) [John] said so, too. (7) [You] want more ice-cream? (8) [The notebook I found] 19 to Sue. (9) [Two women] came in.
Two types of Prepositional Phrases are assigned the phrase type PP.
(10) Scrape it back [into the microwave bowl].
(11) I carefully peeled the skin [off]
(12) Peter thought [about going home]
(13) He taught me not to think [about where I had been and what I had done]. (14) I worry [over why we cover this story].
(15) Who did she believe [had left]?
(16) We made him [go to the store again]
(17) What should she do [to test her hypothesis]?
(18) The twist it included in the storyline had me [tickled].
(19) Visitors don't enjoy [filling out HTML forms].
(20) Last night I learned [that surrealism isn't just a Salvador Dali thing].
(21) Could you tell me [how the Hawaiian Islands formed]?
(22) She told the BBC she did not know [whether the man who survived 7.2 parts per thousand had set a world record].
(23) My mom doesnt like [me being a vegetarian]!
(24) I'd like [you to say hi to my sister].
(25) In 1937, a friend arranged [for Reagan to take a screen test]
(26) Deborah requests [that she be allowed to live in a town nearby].
(27) [Although Smithers credits Lightnin' Hopkins and Mississippi John Hurt as early influences], he says he's not really a bluesman in the classic sense. (28) Alex considers Smithers to be one of his best friends [because Smithers is always looking out for him and making sure he's safe].
(29) [economic] policy (30) [educational] excellence
(31) Philip has [bright green] eyes. (32) The light turned [red].
(33) All items at [greatly] reduced prices! (34) I've been doing that all night and, [quite frankly], my jaw aches.
(35) I have [two] bottles of correction fluid on the stand beside my favorite seat.
(36) ["Could his performance tonight make or break the campaign?"] exhaled John Gibson of Fox News Channel. (37) ["Hush, dear,"] Ruth whispered, ["I know, and I'll tell you some time, but I don't want her to know."]
The distinction between referential and non-referential NPs warrants attention. Expletive it and there are the two kinds of non-referential NPs. As pointed out above, these constituents are given neither FE labels nor GF and PT tags, although we record their presence on the Other Layer. In such cases, FrameNet syntactic tagging cannot be directly mapped onto ordinary syntactic parses.
Some examples are given below.
(38) [It] is clear that we won't finish on time. (39) [It] is odd that George is winning. (40) In the same year [there] arrived from France the Rev. Louis Maigret. (41) [There] ensued a great controversy, as some of the users didn't particularly want to be called losers to their faces every time they used the computer.
Referential NPs are either possessive NPs (marked Poss) or standard (non-possessive) NPs (marked NP). Possessive NPs, which may either be possessive pronouns or noun phrases marked with 's, often express frame elements of predicating nouns. For example, in the Communication frame, possessive nouns express the Speaker role when they are the determiners of target nouns such as claim, remark, reply, etc.:
(42) I question [your] claim that the car was already damaged. (43) [The President's] remarks surprised the reporters. (44) [Leslie's] reply was well-timed.
The phrase type Poss is always paired with the grammatical function Gen(itive).Note: The label Poss is not restricted to NPs denoting
actual possessors. It is a morphosyntactic type rather than a semantic type. In this connection, note that of-PP's whether or not they denote Possessors never get the phrase type Poss. It should further be noted that given our lexicographic purposes, there is no reason to have a special category covering the complex NP type "a remark of the President's" or "a friend of mine". That is, we believe no frame-bearing word will specifically identify such phrases among their valence members.
In some situations it is necessary to tag nominal expressions which are not complete (i.e. maximal) noun phrases. For example, consider nominal modifiers of target nouns, as in examples (45) and (46) below, or the modified nouns in sentences showing target adjectives used attributively, as in the second pair of examples, (47) and (48).
(45) The judge dismissed the [forgery] allegations. (46) [Cancer] treatments are advancing rapidly. (47) Allergic [patients] benefit from this medicine. (48) The senator gave a polemical [acceptance speech].
These non-maximal nominal expressions are given the phrase type N (for `nominal').
In contrast, head nouns that are frame elements of post-nominal modifiers are not treated as non-maximal nominals. Instead they are treated as if the post-nominal modifier was actually predicated of the nouns in a copular clause. Thus, they are labeled full NPs with respect to Phrase Type, and as External arguments with respect to Grammatical Function.
(49) The problem seems to affect [people NP/External] sensitive to primulas.
We treat as standard Noun Phrases all nominals that are not excluded as non-referential noun phrases--recall that these latter are not assigned frame elements, phrase types, and grammatical functions at all--or assigned the phrase type labels possessive (Poss) and non-maximal nominal (N). Standard Noun Phrases are marked with the phrase type NP.
(50) I heard [an interesting story] today. (51) I dropped [the lid] on my foot.
Since we have a commitment to tagging full constituents rather than only their headwords with frame element labels (cf. section 3.2), modifiers and complements are included in the tagged noun phrases, as shown below. Notice in particular, as shown by examples (52) to (57), that we include both restrictive and non-restrictive relative clauses as well as appositives in the NP's we tag.
(52) [The cat in the corner] likes celery. (53) [The cat running down the hall] has the catnip. (54) Stop [that cat with orange stripes running down the hall]. (55) [The cat that's sitting on your lap] sure sheds a lot of hair! (56) [My father, who worked as a clown all his life,] refuses to laugh at any of my jokes. (57) [My uncle, Hollingworth Bowler III,] likes to tell stories from his sea-faring days.
Note further that standard NPs do not have to be headed by nouns. We treat free relative clauses (also called headless relative clauses) as NPs also. Likewise, we consider bare numerals as complete NPs.
(58) Harry will eat [what Sally is eating]. (59) I want [two].
PP is assigned to ordinary prepositional phrases with nominal objects and to particles, the latter under the assumption that particles can be regarded as prepositional phrases which lack objects.
phrases with gerundial objects rather than nominal ones.20 Here are some examples:
In addition, we assign the phrase type PP to the second piece of certain discontinuous degree phrases, as exemplified below.
(64) Billy Jeidels is so wicked [as to be beyond redemption]. (65) Stealing paper clips from work is less immoral [than taking them from a store PP].
These are closely related to the than- and as-phrases in the following sentences, which are also tagged as PP.
(66) Claire is as wicked [as John PP]. (67) Ada is less moral [than Dave PP].
The PT-label PPinterrog is used for prepositions that govern finite or non-finite wh-clauses and verb phrases.
(68) That depends [on who did it PPinterrog] (69) He asked [about how she was doing PPinterrog] (70) Are you confused [about when to start meds PPinterrog]?
Particles like those in the following examples are treated as prepositions without objects and are assigned the Phrase Type PP.
(71) Did the police escort the streaker [out]? (72) He put his hand [in] and felt for the ring.
Note that particles of this kind may occur before NPs and therefore give the appearance of being the heads of regular PPs with NP objects.
However, given that they are separable, as shown in the earlier examples, they cannot plausibly be treated as the heads of PPs in these contexts: out the streaker and in his hand are not constituents in (73) and (74), respectively. Therefore, they are assigned a separate label PP.
Whether or not a word W is to be treated as a particle can be established by this criterion: if verb V and particle W can be used as acceptable paraphrases of each other both in verb phrases of the form V W NP and of the form V NP W, then the word W is a particle rather than a preposition governing an NP.
While some particles, like in in (74) above, are equivalent in form to prepositions and may therefore misleadingly appear to head PPs in certain contexts, other particles do not resemble prepositions and are therefore less likely to be mis-analyzed that way:
There is no syntactic reason to distinguish the particles in (75) and (76) from the ones which resemble prepositions, and they are therefore given the same label (PP).
Finally, note that a verb-particle combination may be either a productively formed combination or a multiword expression (multilexeme lemmas). What is crucial in this connection is whether the verb could have the same meaning when the particle is either missing or when it is replaced by a different particle or preposition. Take off, for instance, is found in the Undressing frame; take up in the Activity_start frame; and think up in the Invention frame. In all these cases, the particle cannot be omitted with the frame-appropriate meaning intact. These combinations are thus to be treated as multiword expressions where no FE/GF/PT triple is assigned to the particle since it is part of the target and accordingly marked with the Target label on the Target-layer.
Some verb-preposition combinations are clearly conventional, as shown here.
(77) The passengers looked at the information monitors. (78) Let me know if you come across that reference I asked you about.
We analyze the prepositions in such expressions as heading PPs.
(79) The passengers looked [at the information monitors]. (80) Let me know if you come [across that reference].
Though these verb-preposition combinations are units in the lexicon, we do not capture their unitary status in terms of constituent structure. That is, we do not analyze look at and come across as syntactic constituents.
In accordance with the Construction Grammar analysis of these expressions, their unitary status is captured in the valence representations of lexical entries. For example, there will be a lexical entry for look at which states that the verbal head look requires a PP headed by the preposition at.
Some prepositions function as individual lexical units relative to a target verb but orthographically consist of more than one word (complex prepositions shown in italics).21
(81) Put the birthday cake next to the other desserts. (82) He mowed the lawn instead of me.
Expressions of this kind are treated as single complex prepositions which head normal PPs. The PPs in the above sentences should be tagged in the following way:
(83) Your birthday cake was put [next to the other desserts]. (84) He mowed the lawn [instead of me].
A preposition and its complement may be separated from each other, with the preposition appearing in a canonical post-verbal position and its complement noun phrase appearing pre-verbally in the clause, in a position that is not syntactically licensed by the verb.
(85) [John] we laughed [at]. (86) [Who] did you refer her [to]?
Since allowing for preposition stranding is not lexically relevant information, FN avoids annotating such sentences. If, however, sentences with preposition stranding have been annotated, then the two parts are assigned the same phrase type value, namely PP. (The two pieces also share the same grammatical function value, Dep.)
If the target word is inside the relative clause, we mark the prepositional phrase containing the relative pronoun or relativizer as a frame element as usual. The noun phrase antecedent to the relative pronoun or relativizer receives the identical FE/GF/PT triple. Ant and Rel labels are applied on the Other layer to the antecedent and the relative pronoun, respectively, as shown earlier.
Notice that both in (87) and (88) the second bracketed constituent is treated as a PP.
If preposition stranding occurs within the relative clause, we proceed in the way described earlier in section 53. The stranded preposition carries a frame element label and shares its phrase type (PP) and grammatical function (Dep) both with the relative phrase (if there is one) and with the antecedent. The Antecedent and any relative word present carry the Ant and Rel labels on the Other layer. The stranded preposition, of course, carries neither of those labels, as is shown in Figure 88.
Every verb phrase has at least a head verb, which may be a main verb or an auxiliary. VPs headed by main verbs may also contain one or more auxiliaries. A verb phrase may also have a negative marker, an infinitive marker, a pre-verbal adverb phrase, one or more complements of the verb, and one or more post-verbal adjuncts. A VP may be headed by the main verb in a sentence or it may be embedded as a complement under another verb. The following examples show a variety of VPs, where the VPs are italicized.
(89) I have. (In response to "Have you taken out the trash?") (90) This book really stinks. (91) I didn't expect you to eat your sandwich so quickly.
Any VP containing a verb (including auxiliaries) which (1) expresses information about tense and (2) is not in a separate embedded clause is tagged as a finite VP. Finite VPs are not generally subcategorized for, but it is nonetheless necessary to tag them in certain contexts, as illustrated here.
(92) Who do you think [ate the sandwich]? (93) What did you say [fell on your hat]?
This pattern seems to be limited to a fairly small number of verbs of belief and assertion which subcategorize for clausal complements: think, believe, say, claim, assert, etc.
In contrast to the irregular verbs eat and fall in the above examples, note that finite past tense verb forms are frequently identical in form to past participial forms, e.g. played-played; stated-stated; etc. The participles are, however, non-finite and they are not covered by the tag VPfin (cf. section 220.127.116.11).
Among non-finite VPs, it is necessary to recognize bare stem infinitives (VPbrst), to-marked infinitives (VPto), verb phrase relatives marked by to (VPtorel), past participial phrases (VPed), and gerunds (VPing).
Bare stem infinitives are non-tensed verb phrases headed by verbs in the bare stem form without the infinitive marker to. Examples of bare stem infinitives (VPbrst) are given below.
(94) We made the children [take naps]. (95) Management let the employees [set their own hours].
Note that the children take naps and the employees set their own hours are not treated as so-called small clauses in the FrameNet project, though that is how they are sometimes analyzed.
marker to. Otherwise they are identical to bare-stem infinitives. Examples of to-marked infinitives appear below.
(96) The cat wants [to go outside]. (97) The mayors persuaded the President [to support the cities]. (98) It is hard for children [to tie their own shoes]. (99) I wish John Edward all the luck in the world and hope there is some element of truth in his claim [to be able to speak to the dead].
Note that to-marked infinitives that occur with noun targets are not always instances of VPto as in (99) above. Some of them serve as restrictive relative clauses and are then given the phrase type VPtorel, as discussed in section 18.104.22.168.
Noun-modifying non-finite relative clauses headed by a to-marked VP are marked as VPtorel, regardless of whether they include a wh-word or not. (There is a separate phrase type label Srel for finite relative clauses.)
(100) Towels [to dry yourself with] can be found in the closet on the left. (101) Nietzsche insists that there are no rules for human life, no absolute values, no certainties [on which to rely].
As a relative clause type, the VPtorel phrase type is, of course, restricted to dependents of noun targets.22
Notice that when verb phrase relatives (and also clausal relatives) are annotated relative to a head noun target, no marking of Ant and Rel on the Other layer takes place.
Participial Verb Phrases are VPs that begin with a past participle, typically ending in -ed. These phrases usually occur as post-nominal modifiers of target nouns, as in (102) and (103), but are also found as complements of certain control verb targets such as have, as in (104).
(102) The witness believes that the man [shown on the photograph] is the bank robber. (103) By discussing the events [covered in the news], parents can help their children gain a better understanding of the world in which they live. (104) The pastor interrupted the service in the middle of his sermon and had the man [forcibly removed].
Gerundive VPs are VPs headed by verbs in the -ing form. They often occur in syntactic contexts in which nominal expressions also occur. Examples of Gerundive VPs are provided here.
(105) My friend likes [running barefoot]. (106) [Inhaling pepper] makes most people sneeze. (107) We watched the dogs [playing].
Gerunds present a challenge because they are sometimes verb-like and sometimes noun-like. FN annotators consider both syntactic and semantic criteria to determine if the automatic classification of gerundive verb phrases is correct. In particular, if the -ing form takes the same arguments as the related verb, e.g. if it takes an object or is modified by an adverb, we tag it as VPing rather than NP, as in (106) above or as in He kept [singing the Albanian national anthem]. Also, if the context makes it clear that the -ing form refers specifically to an action, we use the VPing tag: We were thinking [about dancing tonight], but I twisted my ankle. In this example one could not substitute the noun dance and keep the same meaning. By contrast, Let's discuss [dancing] is adequately re-phrased as Let's discuss dance and one can therefore tag dancing as an NP. In addition, any -ing form that is determined like a noun by, for instance, the, this, that, or a (the killing), or modified by a possessive (my dancing), an adjective (quick thinking), or by a following of-PP (sounding of the alarm) is treated as a noun.
FrameNet syntax treats certain expression types as combinations of smaller constituents in contrast to some syntactic theories which treat them as clauses. For example, the sequence Pat leave in a sentence like They made Pat leave is sometimes analyzed as a `small clause,' but in the FrameNet metalanguage it is treated simply as an NP followed by a bare stem infinitive VP.
This strategy has been adopted for two reasons. First, it simplifies the lexicographers' task of annotation, making it unnecessary to decide in certain cases which combinations of constituents should be treated as clausal and which should not. Second, it makes the lexical descriptions produced by FrameNet relatively theory-neutral. While the question of which verbal complements are clausal and which are not is answered differently in different syntactic theories, the analysis of clauses into their major constituents is uncontroversial in most cases.
As the reader will notice, there sometimes is no parallel between verb phrase types and clausal phrase types. This is true, for instance, for phrase types that figure in main clause and embedded questions. The label Sinterrog (cf. 4.5.3) covers both finite clauses and non-finite verb phrases because there are no predicates that specifically select either finite wh-clauses or non-finite verb phrases. Another case where there is a lack of parallelism involves `small clauses'. Small clauses that are arguments of a target predicate are divided up into an NP and a secondary predicate, except for cases tagged as Sing (cf. 4.5.5). By contrast, small clauses that modify NPs or clauses are assigned to the single category Sabs. (These modifying small clauses are said to figure in absolutive constructions, hence the name Sabs, and they are typically tagged as the extra-thematic frame elements Depictive or Event_depictive.) Not all kinds of small clauses that can appear as arguments can also appear in absolutive constructions: compare *[Bill to arrive], John hid the money and I want [Bill to arrive]. Thus, we lose some formal information by not recording the specific subtypes of absolutive clauses that occur in the data. However, the form of an absolutive construction is not lexically selected, in distinction to the kinds of `small clause' that a predicate can take as an argument: I saw him leave v. *I saw him to leave. From a lexicographic point of view, our treatment is therefore adequate.
In particular for the annotation of Depictive FEs, we introduced a PT Sabs for small clauses modifying either a participant of the main clause as in (108)-(109) or the frame instance evoked by the main clause predicate (110).
(108) [His hands in his pockets Sabs], he shuffled back out of the room to wait until Unca had time to talk to him again. (109) Purring loudly , Cas padded towards her, [tail erect Sabs] , [bright green eyes unblinking Sabs]. (110) [Both sections smashed to flinders Sabs], he could not put it back together before Mom got home.
Note that this label also applies to absolutive constructions that are introduced by with. In cases such as (111), with functions as a subordinator rather than as a preposition: notice that the predicate out of the window is required, which would be unusual if the phrase were a normal with-PP.
(111) [With both feet sticking out of the window Sabs], she evinced great surprise when the officer pulled along side.
Declarative finite complement clauses are full sentences that may begin with the complement marker that. In this PT, the entire clause, including the complement marker, is tagged.
(112) Pat knew [Kim would never agree]. (113) Pat knew [that Kim would never agree].
Structurally, a wh-interrogative clause may be a sentence or a verb phrase. Although not full clauses, the interrogative verb phrases only occur in constructions which allow a full Sinterrog as well, and therefore a single PT is used for both. Note that we treat how as a wh-expression. Wh-expressions are included in the tag for the clause.
(114) I heard [what you said]. (115) I forgot [what to say]. (116) I know [how you feel]. (117) I don't know [how to react]. (118) I asked [who came]. (119) She told me [who to invite].
Structurally, a Whether-if interrogative clause may be a sentence or, in the case of whether, a verb phrase. Although not full clauses, these phrases only occur in constructions which allow a full Whether-if clause and therefore a single PT is used for both.
(120) I wonder [whether the Indian restaurant delivers]. (121) He wondered [whether to turn back]. (122) Kim didn't know [if Pat liked the show].
With certain predicates, sequences of object-form noun phrase and gerundive verb phrase are treated as single clauses by FrameNet. The reason for the analysis as a clause is that with the predicates in question the noun phrase cannot be separated from the gerundive verb phrase, for instance, in passivization. Compare like, which takes an Sing phrase, to see, which takes an NP and a VPing phrase.
(123) I don't like [him being here all the time]. (124) [*He] wasn't liked [being there all the time] versus: (125) You could see [a muscle] [jumping in Hubert Molland's cheek]. (126) [A muscle] could be seen [jumping in Hubert Molland's cheek].
which may look similar to Sing clauses are treated as noun phrases:
(127) I don't like [his being here all the time].
The label Srel is used for all finite relative clauses regardless of being introduced by a wh-word, that, or zero, and whether the clause is interpreted as restrictive or non-restrictive.
(128) The guy [who I bumped into on the train Srel.Dep] was Herbert Kornfeldt. (129) The guy [that I bumped into on the train Srel.Dep] was Herbert Kornfeldt. (130) The guy [I bumped into on the train Srel.Dep] was Herbert Kornfeldt. (131) My neighbor, [who I keep bumping into], gives me angry looks these days.
It should be noted that the label Srel does not apply to so-called free or headless relative clauses such as He lives off [what he can sell at the farmer's market]. Such clauses are tagged with the phrase type label NP since they are distributed like noun phrases.
(132) I'd like [you to meet my mother]. (133) Certainly, but I should hate [you to forget that he has scored more runs in Test cricket than any other Englishman].
In sentences like the above examples, you cannot be the subject of a passive and therefore is treated as part of the non-finite clause.
(134) *[You] would be liked [to meet my mother]
Some nouns and verbs take clauses consisting of a for-marked subject and a to-marked infinitival verb phrase.
(135) I'd like [for you to meet my mother]. (136) I would prefer [for John to stay in the 250 class]. (137) The problem you've got is your claim [for him to contribute retroactively to that expense].
juxtapositions of for-PPs and to-marked verb phrases, as in (138).
(138) We can throw a party [for him] [to show him that we are his friends].
Some verbs and nouns, typically with a semantics involving ordering or commanding, take a clausal complement with the verbal head in the bare stem form, that is, it is identical to the infinitive but lacks marking with to.
(139) The manager demanded [that employees be on time]. (140) The conversation resulted in my insistence [that I be shown at once the place where Jones and the others had met their fate].
Certain clauses introduced by subordinators can instantiate frame elements and consequently need to be tagged. Such clauses receive the phrase type value Sub (Subordinate Clause) rather than Sfin (finite complement clause). Typically, frame elements that are assigned the phrase type Sub have peripheral or extra-thematic status (e.g., Time, Reason). Note that we do not distinguish between the lexical categories of predicates heading the subordinate clauses. That is, examples (141) to (144) all count equally as instances of Sub even though the subordinate clause is headed by a finite verb in (141); by a VPing in (142); by an adjective in (143); and by a preposition in (144).
(141) I admire her [because she is an actress who can also sing Sub.Dep]. (142) [When considering such abominations Sub.Dep], we must be concerned for our precious bodily fluids. (143) [When not quite sober Sub.Dep], Mila likes to sleep on the floor. (144) [Though of noble lineage Sub.Dep], the Count liked to work as a video store clerk.
Certain adjective targets welcome discontinuous Degree phrases, as in the following example, where we assign the phrase type Sub to the than-phrase.
(145) That wine is more delectable [than I could imagine Sub.Dep].
Adjective Phrases typically occur as pre-nominal modifiers, as non-Subject complements of be and a small number of other copular verbs (seem, become, etc.), and as predicate complements of verbs like find, consider, etc.:
(146) They were eating [very large] sandwiches. (147) The house is [empty]. (148) You seem [sad] today. (149) The company considers these documents [extremely valuable].
An Adjective Phrase may consist of just a single adjective, an adjective with some modifying expression (such as an adverb or an intensifier), or a conjunction of adjective phrases:
(150) We found the play [dull]. (151) We found the play [extremely dull]. (152) We found the play [extremely dull and too long].
Some tagged adjectival expressions are not treated as complete (i.e. maximal) adjective phrases. This is typically the case with relational modification, which is semantically very similar to noun-noun compounding (cf. 3.4.2) in not allowing for degree modification, as shown by (153) and (154), and in not allowing for a paraphrase in which the modifier is predicated of the head, as shown by the unacceptable contrasts in (155) and (156).
(153) [marital] bliss v. *[very marital] bliss
(154) [fruit] juice v. *[very/high fruit] juice
a. The Party of the Institutionalized Revolution (PRI) devised a new [economic] policy.
b. *The policy that the Party of the Institutionalized Revolution (PRI) has devised is [economic]. (156)
a. I got stuck in [rush hour] traffic.
b. *The traffic I got stuck in was [rush hour].
Relational adjectival modifiers are given the phrase type A, paralleling the treatment of nominal modifiers of nouns, which, as non-maximal nouns, are assigned the phrase type N.
Some adjectives take complements other than the nouns they modify, and these are included as part of the Adjective Phrase.
(157) When did you first become [interested in dinosaurs?]
(158) I got [scared of the typing sound].
In addition to adjectives like interested, fond, afraid, scared etc., the comparative forms of gradable adjectives account for a large number of complement-taking adjectives.
(159) Leslie got [taller than Kim AJP.Dep].
(160) Bart turned [angrier than he'd felt in a very long time AJP.Dep].
An adjective and its complement may form a discontinuous constituent when they modify a head noun. This is very common with morphologically comparative adjectives, as in (161), but also occurs with other adjectives that semantically denote comparison, as in (162).
In such cases, both the adjective and its complement have the same frame element label applied to them, and they are assigned the same grammatical function label. However, the two pieces do receive separate phrase type values.
Adverbs, too, may express frame elements of a target verb, as illustrated in examples (163) to (165). They are assigned the phrase type AVP.
Note that some words that at first glance may not appear to be adverbs are assigned the phrase type AVP. For instance, home as used in He went home already is treated as an adverb.
When annotating in respect to target nouns, the preceding number or quantifier is given the phrase type QUANT. For example,
(166) Bob poured [two QUANT] cups of coffee. (167) Bob poured [thirty seven QUANT] cups of coffee at the brunch. (168) Bob drank [a QUANT] glass of wine.
Note that we treat cardinal numbers and a (= 1) in the same way.
Some verbs of communication take quoted material as a complement that is assigned the phrase type QUO (and the grammatical function Dep). For example:
Quoted material can be of any syntactic form, or syntactically ill-formed, for that matter. Because the distribution or `external syntax' of quoted material does not depend on its internal syntactic structure, we use a separate phrase type to tag it. Only direct quotes as in (169)-(170) and interior monologue as in (171) are given the phrase type QUO. Indirect quotes always take the form of some other kind of specific phrase type, as shown here.
(172) They asked us [what we were doing there].(Wh-clause) (173) The President said [that he would support the inner city].(That-clause)
Quoted material is easy to identify because it almost always appears in quotation marks, which is included inside the brackets marking the Quote constituent.
Sometimes quoted material forms a discontinuous constituent :
(174) ["Cities,"] he said, ["are a very high priority."]
In such cases, both portions of the quote should be assigned the PT QUO.