Conjunction

Word-level conjunction

Constituents that are labeled only at the word level (notably, verbs and particles) conjoin at that level.

(IP-INF (TO to)
        (VB (VB read) (CONJ and) (VB write))
        (NP-OB1 (NS letters)))

( (RP (RP in) (CONJ or) (RP out))

For simplicity, categories that project phrases also conjoin at the word level, provided that all conjuncts consist of a single word. Since the constituenthood of the conjunction structure is already indicated by a phrasal tag, no additional word-level tag surrounds the structure, unlike in the case of verbs and particles.

( (NP (N chess) (CONJ or) (N checkers)))

( (ADJP (ADJ red) (PUNC ,) (ADJ yellow) (PUNC ,) (CONJ or) (ADJ green)))

( (ADVP-LOC (ADV here) (CONJ and) (ADV there)))

( (QP (QR  more) (CONJ or) (Q less)))

This convention is extended to conjoined prepositions.

( (PP (P (P before) (CONJ or) (P after))
      (NP (N dinner))))

CONJP with conjoined words

Word-level conjunction structures generally include at least one overt conjunction. In the rare exceptions, CONJPs are added to clarify the structure (notably, to distinguish these cases from compound nouns).

( (NP (NS Friends)
      (PUNC ,)
      (CONJP (NPRS Romans))
      (PUNC ,)
      (CONJP (NS countrymen))))

( (ADJP (ADJ red)
	(PUNC ,)
	(CONJP (ADJ white))
	(PUNC ,)
	(CONJP (ADJ blue))))

Phrase-level conjunction

When any of the conjuncts consists of more than a single word, all conjuncts are labeled as phrases. The non-first conjuncts are labeled CONJP, and they are adjoined to the first conjunct at the phrasal level.

(XP (XP (X ...))
    (PUNC ,)
    (CONJP (CONJ conjunction)
           (YP (Y ...)))
    (PUNC ,)
    (CONJP (CONJ conjunction)
           (ZP (Z ...) (W ...)))))      ← multi-word conjunct forces phrase-level conjunction

(XP (XP (X ...))
    (PUNC ,)
    (CONJP (YP (Y ...)))                ← implicit CONJ
    (PUNC ,)
    (CONJP (CONJ conjunction)
           (ZP (Z ...) (W ...))))

The general schema applies to all phrasal categories, including clauses. But clausal conjunction raises some special issues, so it is discussed separately.

( (NP (NP (D the) (N duke))
      (CONJP (CONJ and)
	     (NP (PRO$ his) (N lady)))))

( (ADJP (ADJP (ADV quite) (ADJ expensive))
	(CONJP (CONJ but)
               (ADJP (ADV very) (ADJ durable)))))

( (ADVP (ADVP (ADV comfortably))
	(CONJP (CONJ and)
               (ADVP (QR more) (ADV conveniently)))))

( (PP (PP (P for)
          (ADJP (ADJR better)))
      (CONJP (CONJ or)
             (PP (P for)
                 (ADJP (ADJR worse))))))

All clear cases of phrasal conjunction, including ones with implicit conjunctions, contain CONJP. Implicit conjunctions are ordinary in cases with more than two conjuncts, where an overt conjunction generally appears only before the final conjunct. But there are also cases without any overt conjunction at all.

( (NP (NP (N life))
      (PUNC ,)
      (CONJP (NP (N liberty)))                  ← implicit CONJ in second conjunct
      (PUNC ,)
      (CONJP (CONJ and)
	     (NP (D the) (N pursuit)
		 (PP (P of)
		     (NP (N happiness)))))))

( (NP (NP (NS walls))
      (PUNC ,)
      (CONJP (NP (NS ceilings)))                
      (PUNC ,)
      (CONJP (NP (FP even) (NS floors)))))      ← implicit CONJ even in final conjunct

In general, sets of constituents (NPs, PPs, etc.) acting as a unit are instances of either conjunction or apposition.

( (NP (NP (ADJ great) (N joy))		       ← ill-formed structure; should be conjunction
      (PUNC ,)
      (NP (ADJ great) (N sorrow))))

( (NP (NP (ADJ great) (N joy))		       ← corrected structure, with CONJP
      (PUNC ,)                                   (separate phrasal brackets for each conjunct)
      (CONJP (NP (ADJ great) (N sorrow)))))

( (NP (NP (SUCH such) (N joy))		       ← ill-formed structure; should be apposition
      (PUNC ,)
      (NP (SUCH such) (ADJ pure) (N joy))))

( (NP (SUCH such) (N joy)		       ← corrected structure, with NP-PRN-ELAB
      (PUNC ,)
      (NP-PRN-ELAB (SUCH such)                   (separate phrasal brackets only for parenthetical)
		   (ADJ pure)
		   (N joy)))) 

Recursive conjunction structure

Conjunction structures can be recursive. We attempt to represent the intent of the text, using various clues such as the distribution of conjunctions, punctuation, syntactic parallelism, and alliteration.

( (NP-OB1 (NP (N gold) (CONJ and) (N seluer))
          (PUNC ,)
          (CONJP (NP (NP (ADJ precious) (NS closes))             ← recursive CONJP 
                     (CONJP (CONJ and)
                            (NP (ADJ vertuous) (N stones)))))
          (PUNC ,)
          (CONJP (CONJ and)
                 (NP (Q al) (D +te) (N pompe)
                     (PP (P off)
                         (NP (ADJ wordliche) (N blisse))))))
  (ID CMAELR3-M23,28.70))

( (NP (NP (NP-POS (ADJ blacke) (NS$ shepis))
          (NS skynnes))
      (PUNC ,)
      (CONJP (CONJ and)
             (NP (D a) (ADJ grete) (N payre)
                 (PP (P of)
                     (NP (NS bootis)))))
      (PUNC ,)
      (CONJP (CONJ and)                                          ← recursive CONJP 
             (NP (NP (D a) (N boowe))
                 (CONJP (CONJ and)
                        (NP (NS arowis))))))
  (ID CMMALORY-M4,30.935))

Prices

Prices as in the following examples are always treated as phrasal conjunction structures. The conventional written order of currency symbols and numbers can be reversed to reflect the spoken order, with an editorial comment about the written order.

(NP (NP (NUM 749)
        (NS l.))
    (CONJP (NP (NUM 0)
               (NS s.)))
    (CONJP (NP (NUM 7)
               (NS d.))))

(NP (NP (NUM 749))               ← phrase-level conjunction even with one-word conjuncts
    (PUNC :)
    (CONJP (NP (NUM 0)))
    (PUNC :)
    (CONJP (NP (NUM 7))))

(NP (NP (NUM 749)
        (NS +L))
    (CODE {TEXT:+L749})
    (CONJP (NP (NUM 0)
               (NS s.)))
    (CONJP (NP (NUM 7)
               (NS d.))))

Conjunction of unlike categories

Conjuncts are generally of the same category, but not always. In the case of unlike categories, the category enclosing the entire conjunction structure is, by default, that of the first conjunct.

( (DOD (DOD did) (CONJ and) (VBD said)))

( (VBD (VBD said) (CONJ and) (DOD did)))

( (ADJP (ADJP (Q half) (ADJ dead))
	(CONJP (CONJ and)
	       (PP (Q half) (P+N alive)))))

( (PP (PP (Q half) (P+N alive))
      (CONJP (CONJ and)
	     (ADJP (Q half) (ADJ dead)))))

( (ADVP (ADVP (ADV slowly))
	(CONJP (CONJ and)
	       (PP (P with)
		   (NP (Q some) (N difficulty))))))

( (PP (PP (P with)
	  (NP (Q some) (N difficulty)))
      (CONJP (CONJ and)
	     (ADVP (ADV slowly)))))

Correlative conjunction

Word-level correlative conjunction

Single-word cases are bracketed flat, as in the ordinary case.

( (NP (CONJ both) (N body) (CONJ and) (N soul))
  (ID CMAELR3-M23,31.137))

( (ADJP (CONJ both) (ADJ big) (CONJ and) (ADV little)))

( (ADVP (CONJ neither) (ADV quickly) (CONJ nor) (ADV slowly)))

( (QP (CONJ neither) (QR more) (CONJ nor) (QR less)))

( (VB (CONJ either) (VB sing) (CONJ or) (VB dance)))

Phrase-level correlative conjunction

In phrase-level corelative conjunction, the correlative word attaches as high as possible within the conjunction structure, regardless of word order.

( (NP-SBJ (CONJ neither)
          (NP (PRO he))
          (CONJP (CONJ nor)
                 (NP (PRO$ his) (N wife)))))

( (PP (CONJ both)                                    ← BOTH attaches high in PP
      (PP (P in)
          (NP (N height)))
      (CONJP (CONJ and)
             (PP (P in)
                 (NP (N width))))))

( (PP (PP (P in)
          (NP (N height)))
      (CONJP (CONJ and)
             (PP (P in)
		 (NP (N width))))
      (CONJ both)))	                             ← BOTH still attaches high in PP

( (PP (P in)
      (NP (ADJP (CONJ both)
		(ADJ material)
                (CONJP *ICH*-1))
	  (NS respects)
	  (CONJP-1 (CONJ and) (ADJ spiritual)))))

( (PP (P in)
      (NP (ADJP	(ADJ material)
                (CONJP *ICH*-1))
	  (NS respects)
	  (CONJP-1 (CONJ and)
		   (ADJ spiritual)
		   (CONJ both)))))	             ← BOTH attaches high within CONJP

Floated correlative conjunction

Floated correlative conjunctions are attached
as high as possible, usually at the IP level, but lower if necessary. Where possible, they are coindexed with an *ICH* trace.

( (IP-MAT (CONJ-1 Either)
          (NP-SBJ (PRO you))
          (BEP are)
          (NP-PRD (CONJ *ICH*-1)
                  (NP (NPR God))
                  (CONJP (CONJ or)
			 (NP (NPR$ God's) (N disciple))))
	  (PUNC .)))

( (IP-SUB (NP-SBJ (D the) (N king))
          (MD should)
          (CONJ-1 neither)
          (VB carry)
          (NP-OB1 (CONJ *ICH*-1) (N sword) (CONJ nor) (N knife))))

( (IP-SUB (NP-SBJ (PRO$ his) (N body))
          (BED was)
	  (ADJP-PRD (ADJ full)
                    (PP (P of)
                        (NP (ADJ great) (NS wounds))))
	  (PUNC ,)
	  (CONJ-1 both)
	  (PP (P on)
              (NP (CONJ *ICH*-1)
		  (D the) (NS shoulders)		← no NP because of subsequent NX 
		  (PUNC ,)
		  (CONJP (NX (NS arms)))
		  (PUNC ,)
		  (CONJP (CONJ and)
			 (NX (N face)))))))

( (IP-MAT (CONJ and)
          (NP-SBJ (ADJ merciful) (NPR Christ))
          (VBD granted)
          (NP-OB2 (PRO him))
          (NP-OB1 (N health)
                  (CONJ-1 both)		                ← CONJ floated as high as possible
                  (PP (P of)
                      (NP (CONJ *ICH*-1) (N body) (CONJ and) (N soule))))
          (PUNC .)))

( (IP-MAT (CONJ for)
          (PP (P in)
              (NP (Q all) (NS tournaments)		← no NP because of subsequent NX 
                  (PUNC ,)
                  (CONJP (NX (NS jousts)))
                  (PUNC ,)
                  (CONJP (CONJ and)
                         (NX (NS deeds)
                             (PP (P of)
                                 (NP (NS arms)))))
                  (PUNC ,)
                  (CONJ-1 both)		                ← CONJ floated as high as possible
                  (PP (P for)
                      (NP (CONJ *-1) (N life) (CONJ and) (N death)))))
          (PUNC ,)
          (NP-SBJ (PRO he))
          (VBD surpassed)
          (NP-OB1 (Q all) (OTHER other) (NS knights))
	  (PUNC ,)))

Conjunction and negation

In both of the following constructions, NOT is attached low because it forms a constituent with the phrase that it negates. In other words, NOT is never treated as a correlative conjunction or as part of one.

X (AND/BUT) NOT Y

(XP (XP (X first-conjunct))
    (CONJP (CONJ conjunction)
           (YP (NEG not) (Y second-conjunct))))		← X AND/BUT NOT Y

(XP (XP (X first-conjunct))
    (CONJP (YP (NEG not) (Y second-conjunct))))		← X NOT Y

( (NP (NP (D the) (N man))
      (CONJP (CONJ and)
             (NP (NEG not) (D the) (N horse)))))

( (ADVP-LOC (ADVP (ADV inside))
            (PUNC ,)
            (CONJP (CONJ but)
                   (ADVP (NEG not) (ADV outside)))))

( (PP (PP (P to)
          (NP (PRO us)))
      (CONJP (CONJ and)
             (PP (NEG not)
                 (P to)
                 (NP (PRO them))))))

( (PP (PP (P in)
          (NP (N hope)
              (PP (P of)
                  (NP (NPR$ God's) (N help)))))
      (PUNC .)
      (CONJP (CONJ &)                                       ← recursive CONJP
             (PP (P in)
                 (NP (N trust)
                     (PP (PP (P of)
                             (NP (PRO$ his) (N grace)))
                         (PUNC .)
                         (CONJP (PP (NEG not)
                                    (P on)
                                    (NP (PRO$ their) (N strength))))))))))

NOT (ONLY) X BUT (ALSO) Y

Unlike NOT (ONLY), ALSO attaches high - as a sister to the conjunction.

(XP (XP (NEG not) (FP only) (X first-conjunct) ...)
    (CONJP (CONJ conjunction)
           (ALSO also)
           (YP (Y second-conjunct) ...)))

( (NP (NP (NEG not)
          (NP (ADJ harsh) (NS curses)))
      (CONJP (CONJ but)
             (NP (ADJ gentle) (NS entreaties)))))

( (PP (PP (NEG not)
	  (P with)
	  (NP (ADJ harsh) (NS curses)))
      (CONJP (CONJ but)
	     (PP (P with)
		 (NP (ADJ gentle) (NS entreaties))))))

( (NP (NP (NEG not) (FP only)			← low attachment of NOT ONLY
          (D the) (NS faculty))
      (CONJP (CONJ but)
	     (ALSO also)			← high attachment of ALSO
             (NP (D the) (NS students)))))

Extraposed CONJP

Cases where an extraposed CONJP is construed with a subject are treated by default as gapping.

When a non-first conjunct of a conjunction structure is extraposed, it is coindexed with an *ICH* trace.

( (NP (ADJP (ADJP (ADJ young))
            (CONJP *ICH*-1))
      (N people)
      (CONJP-1 (CONJ and)
               (ADJP (ADV musically) (ADJ talented)))))

This treatment extends to cases where the non-first conjunct is a single word. Although word-level conjunction does not ordinarily contain CONJP, when extraposition is involved, a CONJP is added (motivated by the fact of extraposition itself). But no additional phrasal nodes are added. As in the phrasal case just discussed, the extraposed CONJP is coindexed with an *ICH* trace.

( (NP (ADJP (ADJP (ADJ young))
            (CONJP *ICH*-1))
      (N people)
      (CONJP-1 (CONJ and) (ADJ talented))))

( (NP (PRO$ hir)
      (ADJP (ADJ good)
            (CONJP *ICH*-1))
      (N conuersacyon)
      (PUNC ,)
      (CONJP-1 (CONJ and) (ADJ honest)))
  (ID CMAELR4-M4,3.73))

( (IP-MAT (NP-SBJ (PRO$ Hare) (NS ouerherren))
          (VBP (VBP witi+d)
               (CONJP *ICH*-1))
          (NP-OB1 (PRO ham))
          (CONJP-1 (CONJ ant)
                   (VBP wrea+de+d))
          (PUNC .))
  (ID CMHALI-M1,150.309))

( (IP-MAT (CONJ for)
          (NP-SBJ (PRO ich))
          (MD chulle)
          (VB (VB lowse)
              (CONJP *ICH*-1))
          (NP-OB1 (PRO +te))
          (CONJP-1 (CONJ &)
                   (VB leten))
          (PP (P hwen)
              (CP-ADV (C 0)
                      (IP-SUB (NP-SBJ *exp*)
                              (NP-OB2 (PRO me))
                              (VBP +tunche+d))))
          (PUNC .))
  (ID CMJULIA-M1,109.223))

Clausal conjunction

Conjoined matrix IPs

Main clauses are divided into separate sentence tokens even when overtly conjoined. An exception is made for pseudo-imperatives in order to enable their retrieval. An empty subject (NP-SBJ *con*) is included in any non-first conjunct.

( (IP-MAT (ADVP-TMP (ADV Than))
          (NP-SBJ (NPR Arthure))
          (VBD toke)
          (NP-OB1 (D the) (N swerde))
          (PP (P by)
              (NP (D the) (N sheethe)
                  (CONJP (CONJ and)
                         (NX (N gurdil))))))
  (ID CMMALORY-M4,46.1496))

( (IP-MAT (CONJ and)
          (NP-SBJ *con*)
          (VBD pulled)
          (PP (P at)
              (NP (PRO hit)))
          (ADVP (ADV egirly))
          (PUNC ,))
  (ID CMMALORY-M4,46.1497))


( (IP-MAT (NP-SBJ (D Thys) (N damesell)) (ADVP-TMP (ADV than)) (VBD behelde) (NP-OB1 (D thys) (ADJ poure) (N knyght))) (ID CMMALORY-M4,46.1526)) ( (IP-MAT (CONJ and) (NP-SBJ *con*) (VBD saw) (CP-THT (C 0) (IP-SUB (NP-SBJ (PRO he)) (BED was) (NP-PRD (D a) (ADJ lyckly) (N man)))) (PUNC ;)) (ID CMMALORY-M4,46.1427))

Conjoined embedded IPs

The absence of a VP level in our annotation system forces us to treat both VP and IP conjunction in the same way - namely, as IP conjunction. An
empty subject (NP-SBJ *con*) is included in any non-first conjunct. Any traces of movement are annotated across the board (that is, in each conjunct). Cases where more than the subject is missing from a non-first conjunct are annotated as gapping rather than as simple conjunction.

( (CP-THT (C that)
          (IP-SUB (IP-SUB (NP-SBJ (NUM two) (NS ladyes))
                          (VBD mette)
                          (PP (P with)
                              (NP (PRO hym))))
                  (PUNC ,)
                  (CONJP (CONJ and)
                         (IP-SUB (IP-SUB (NP-SBJ (D that) (ONE one))
                                         (VBD sate)
                                         (PP (P uppon)
                                             (NP (D a) (N lyon))))
                                 (PUNC ,)
                                 (CONJP (CONJ and)
                                        (IP-SUB (NP-SBJ (D that) (OTHER other))
                                                (VBD sate)
                                                (PP (P uppon)
                                                    (NP (D a) (N serpente)))))))))
  (ID CMMALORY-M4,665.4767))

( (CP-ADV (C 0)
          (IP-SUB (IP-SUB (NP-SBJ (PRO he))
                          (BED was)
                          (NP-PRD (N kynge)
                                  (PP (P of)
                                      (NP (Q all) (NPR Englond)))))
                  (CONJP (CONJ and)
                         (IP-SUB (NP-SBJ *con*)
                                 (ADVP (ADV so))
                                 (VBD regned)))))
  (ID CMMALORY-M4,2.7))

( (CP-REL (WPP-1 (WADV+P wherof))
          (C 0)
          (IP-SUB (IP-SUB (PP *T*-1)                               ← across-the-board extraction 
                          (NP-SBJ (D the) (NS barons))
                          (BED were)
                          (ADJP-PRD (ADV sore) (ADJ agreved)))
                  (CONJP (CONJ and)
			 (IP-SUB (PP *T*-1)                        ← across-the-board extraction 
                                 (NP-SBJ *con*)
                                 (VBD put)
                                 (NP-OB1 (PRO it))
                                 (RP of)
                                 (PP (P in)
                                     (NP (N delay)))
                                 (PP (P till)
                                     (NP (D the) (ADJ hyghe) (N feste)
                                         (PP (P of)
                                             (NP (NPR Eester)))))))))
  (ID CMMALORY-M4,10.289))

Conjoined direct speech

After a verb of saying, whether in a matrix or subordinate clause, all full conjoined clauses containing direct speech after the first one are split off as separate tokens.

( (IP-MAT (CONJ and)
          (NP-SBJ (PRO she))
          (VBD seyde)
          (PUNC ,)
          (PUNC ')
          (IP-MAT-SPE (NP-VOC (NPR sir) (NPR Percivale))
                      (PUNC ,)
                      (NP-SBJ (PRO I))
                      (VBP playne)
                      (PP (P unto)
                          (NP (PRO you)))
                      (PP (P of)
                          (NP (CP-FRL-SPE (WNP-1 0)
                                          (C that)
                                          (IP-SUB-SPE (NP-OB1 *T*-1)
                                                      (NP-SBJ (PRO ye))
                                                      (HVP have)
                                                      (DON done)
                                                      (PP (P unto)
                                                          (NP (PRO me))))))))
          (PUNC ,))
  (ID CMMALORY-M4,665.4774))

( (IP-MAT-SPE (CONJ and)			← separate sentence token
              (NP-SBJ (PRO I))
              (HVP have)
              (NEG nat)
              (VBN offended)
              (PP (P unto)
                  (NP (PRO you)))
              (PUNC .)
              (PUNC '))
  (ID CMMALORY-M4,665.4775))

Pseudo-imperative

In the so-called pseudo-imperative, an imperative and a following matrix declarative correspond pragmatically to the protasis and apodosis of a conditional construction. In order to represent the tight link between the two clauses and to enable their retrieval, pseudo-imperatives are annotated as single tokens. The constituent clauses are annotated according to form, not function, and the root node is IP-IMP in accordance with the guidelines for
conjoining unlike categories. In doubtful cases, the default annotation is as a single token, in order to enable retrieval.

(IP-IMP (IP-IMP (VBI Cross)
                (NP-OB1 (D the) (N line)))
        (PUNC ,)
        (CONJP (CONJ and)
               (IP-MAT (NP-SBJ (PRO I))
                       (VBP shoot)))
        (PUNC .))

(IP-IMP (IP-IMP (VBI Stay)
                (PP (P behind)
                    (NP (D the) (N line))))
        (PUNC ,)
        (CONJP (CONJ or)
               (IP-MAT (NP-SBJ (PRO I))
                       (VBP shoot)))
        (PUNC .))

Right node raising

Right node raising is treated differently depending on whether the raised constituent is smaller than NP or not. In the first case, that constituent is immediately dominated by NP; in the second (general) case, by some type of IP. In the examples below, the raised constituent is underlined.

an enlarging and a restraining power		← right node raised constituent = N
She asked for, but she did not get, a raise.	← right node raised constituent = NP
The NP-internal type is treated like ordinary conjunction, except that the head of the first conjunct is not expressed.

(NP (NP (D an) (ADJ enlarging)
    (CONJP (CONJ and)
           (NP (D a) (ADJ restraining) (N power))))
In the clausal type, the incomplete IP is coindexed with the IP from which its missing parts can be reconstructed. The incomplete IP shares a dash tag with the complete IP and is further annotated as a parenthetical, yielding IP-MAT-PRN, IP-SUB-PRN, and so on. The incomplete IP includes no information beyond part-of-speech and minimal phrasal labeling (possibly including the annotation of grammatical functions). PPs in such clauses need not include a prepositional object.

( (IP-MAT-1 (NP-SBJ (PRO She))			  ← complete clause (includes raised constituent)
            (VBD asked)
            (PP (P for)
                (IP-MAT-PRN=1 (CONJ but) 	  ← incomplete clause (excludes raised constituent)
                              (NP-SBJ (PRO she))
                              (DOD did)
                              (NEG not)
                              (VB get))
                (NP (D a) (N raise)))))

( (IP-MAT-1 (NP-SBJ (PRO She))
            (VBD received)
            (IP-MAT-PRN=1 (PP (P though)
                              (CP-ADV (C 0)
			              (IP-SUB (NP-SBJ (PRO she))
                                              (DOD did)
                                              (NEG not)
                                              (VB know)
					      (CP-QUE-SUB (WNP-2 (WPRO what))
                                                          (IP-INF (NP-OB1 *T*-2)
							          (TO to)
                                                                  (DO do)
                                                                  (PP (P with))))))))	← missing object
            (NP-OB1 (D a) (N raise))))

( (IP-INF-1 (TO to)
            (VB se)
            (IP-INF-PRN=1 (CONJ and)
                          (TO to)
                          (VB beholde))
            (NP-OB1 (D the) (N multitude)
                    (PP (P of)
                        (NP (N peple)
                            (CP-REL (WNP-2 0)
                                    (C that)
                                    (IP-SUB (NP-SBJ *T*-2)
                                            (VBD fledde)))))))
  (ID CMMALORY-M4,25.787))

( (IP-SUB-1 (PP (P by)
                (NP (N grace)))
            (NP-SBJ (PRO he))
            (VBP sese)
            (PUNC ,)
            (IP-SUB-PRN=1 (CONJ or)
                          (NP-SBJ (PRO he))
                          (VBP heres))
            (PUNC ,)
            (IP-SUB-PRN=1 (CONJ or)
                          (NP-SBJ (PRO he))
                          (VBP felys))
            (NP-OB1 (FP anely) (NPR Godd))
            (PP (P in)
                (NP (Q all) (NS creaturs))))
  (ID CMROLLTR-M24,16.384))

( (IP-MAT-1 (CONJ And)
            (PP (P+D for-+ti))
            (NP-SBJ (PRO he))
            (MD may)
            (ADVP-TMP (ADV neuer))
            (VB fele)
            (IP-MAT-PRN=1 (CONJ ne)
                          (VB come)
                          (PP (P to)))		← missing object
            (NP-OB1 (D +te) (NS joyes)
                    (PP (P of)
                        (NP (NPR Heuene))))
            (PUNC ,)))

Gapping

The treatment of gapping resembles that of right node raising. As with right node raising, the gapped clause is coindexed with the complete clause from which it can be reconstructed.

If the gapped clause is clause-internal, it is labeled as a parenthetical. Gapping of this type is often used even without strict equivalence between the complete and the gapped clause. If the relationship is extremely loose, however, the second clause may be dominated by FRAG.

( (IP-MAT-1 (CONJ And)
            (NP-SBJ (NS men))
            (VBP comen)
            (PP (P fro)
                (ADJP (ADJ fer)))
            (PP (P by)
                (NP (N watre)))
            (PP (P in)
                (NP (NS schippes)))
            (IP-MAT-PRN=1 (CONJ &)
                          (PP (P be)
                              (NP (N londe)))
                          (PP (P with)
                              (NP (NS cartes))))
            (IP-INF-PRP (FOR for) (TO to)
                        (VB fetten)
                        (PP (P of)
                            (NP (D +tat) (N grauell)))))
  (ID CMMANDEV-M3,19.464))

( (IP-MAT-1 (CONJ and)
            (ADVP-LOC (ADV there))
            (BED was)
            (NP-SBJ (NPR Gryfflette))
            (VAN put)
            (PP (P to)
                (NP (D the) (N erth)))
            (NP-ADV (N horse) (CONJ and) (N man))
            (PUNC ,)
            (IP-MAT-PRN=1 (CONJ and)
                          (NP-SBJ (NPR Lucas)
                                  (NP-PRN (D the) (N Butler)))
                          (NP-ADV (N horse) (CONJ and) (N man)))
            (PUNC ,)
            (PP (P $by)
                (NP (NP (NPR kynge) (NPR Brandegoris))
                    (CONJP (CONJ and)
                           (NP (NPR kynge) (NPR Idres)))
                    (CONJP (CONJ and)
                           (NP (NPR kynge) (NPR Angwyshaunce)))))
            (PUNC .))
  (ID CMMALORY-M4,21.648))

If the gapped clause is sentence-final (which in right node raising is impossible by definition), it does not receive a -PRN tag.

( (CP-ADV (C 0)
          (IP-SUB (IP-SUB-1 (NP-SBJ (NPR kynge) (NPR Uther))	        ← complete clause
                            (MD wille)
                            (ADVP (ADV wel))
                            (VB rewarde)
                            (NP-OB1 (PRO me)))
                  (CONJP (CONJ and)
                         (IP-SUB=1 (BE be)				← gapped clause
                                   (VAN sworne)
                                   (PP (P unto)
                                       (NP (PRO me)))
                                   (IP-INF (TO to)
                                           (VB fulfille)
                                           (NP-OB1 (PRO$ my) (N desyre)))))))
  (ID CMMALORY-M4,3.57))

( (IP-MAT (IP-MAT-1 (CONJ And)
                    (ADVP (ADV so))
                    (NP-SBJ (NPR sir) (NPR Launcelot))
                    (VBD rode)
                    (PP (P on)
                        (NP (D the) (ONE one) (N syde))))
          (CONJP (CONJ and)
                 (IP-MAT=1 (NP-SBJ (PRO she))
                           (PP (P on)
                               (NP (D the) (OTHER other) (N syde)))))
          (PUNC .))
  (ID CMMALORY-M4,207.3402))

( (IP-MAT (IP-MAT-1 (ADVP-TMP (ADV Than))
                    (NP-SBJ (NPR sir) (NPR Percivale))
                    (VBD smote)
                    (NP-OB1 (D the) (ADJ firste))
                    (PP (P to)
                        (NP (D the) (N erth))))
          (CONJP (CONJ and)
                 (IP-MAT=1 (NP-OB1 (PRO$ hys) (N horse))
                           (PP (P uppon)
                               (NP (PRO hym)))))
          (PUNC ,))
  (ID CMMALORY-M4,661.4640))

Gapping differs from ordinary IP conjunction in that more than the subject is missing from any non-first conjuncts. A common case is for the subject and a modal to be missing. Also, as a rule, gapped clauses dominate more than one constituent (not counting conjunctions); otherwise, the default annotation is ordinary conjunction (possibly together with extraposition).

Exception: Cases where a CONJP is extraposed from a subject are treated by default as gapping rather than conjunction.

( (IP-MAT (IP-MAT-1 (NP-SBJ (D The) (N queen))                         ← like this
		    (BEP was)
		    (VAG making)
		    (NP-OB1 (ADJ great) (N sorrow)))
	  (PUNC ,)
          (CONJP (CONJ and)
		 (IP-MAT=1 (NP-SBJ (Q all) (D the) (NS barons))))
	  (PUNC .)))

( (IP-MAT (NP-SBJ (NP (D The) (N queen))   		               ← not like this
		  (CONJP *ICH*-1))
	  (BEP was)
	  (VAG making)
	  (NP-OB1 (ADJ great) (N sorrow))
	  (CONJP-1 (CONJ and)
		   (NP (Q all) (D the) (NS barons)))
	  (PUNC .)))

The above exception to the default preference for ordinary conjunction maximizes consistency with the historical French corpora and is motivated for both English and French by the subject-verb agreement facts illustrated immediately below. In French, the facts are more salient than in English, where the loss of agreement morphology bleeds conclusive evidence for the gapping analysis.

Singular agreement consistent with gapping

      The queen is coming,
      and the barons are coming.

Singular agreement inconsistent with conjunction + extraposition

      The queen [CONJP and the barons] { *is, ✓ are } coming.

      The queen [CONJP-1 and the barons] { ✓is, *are } coming,
      [CONJP-1 and the barons ].

Backwards gapping

Backwards gapping is necessary for cases where missing material precludes a right node raising analysis. It is also used for WHETHER OR NO clauses.

In backwards gapping, the constituent that counts as gapped is the first conjunct (rather than a non-first conjunct, as with ordinary gapping). The conjunction is included under CONJP in either type.

( (IP-MAT (IP-MAT=1 (NP-SBJ (PRO I))
                    (HVP have))                               ← missing "withstood"
          (CONJP (CONJ and)
                 (IP-MAT-1 (NP-SBJ *con*)
		           (MD will)
		           (VB withstand)
		           (NP-OB1 (PRO him))))))

( (PP (P+N because)
      (CP-ADV (C 0)
              (IP-SUB (IP-SUB=1 (NP-SBJ (PRO I))
                                (HVP haue)
		    		(NEG not))                    ← missing "thought"
   		      (CONJP (CONJ nor)
			     (IP-SUB-1 (NP-SBJ *con*)
                                       (MD can)
				       (NEG not)
				       (VB thinke)
				       (IP-SMC (NP-SBJ=2 (PRO it))
					       (ADJP-PRD (ADJ reasonable))
 					       (IP-INF-2 (TO to)
						    	 (VB yeld)
							 (PP (P to)
							     (NP (ADJ sondry)
                                                                 (PRO$ his)
                                                                 (NS requestes)))))))))))

Conjoined CPs

( (CP-THT (CP-THT (C that)
                  (IP-SUB (NP-SBJ (PRO we))
			  (VBP come)))
          (CONJP (CONJ or)
		 (CP-THT (C that)
			 (IP-SUB (NP-SBJ (PRO you))
				 (VBP go))))))

( (CP-QUE-SUB (CP-QUE-SUB (WNP-1 (WPRO what))
			  (IP-SUB (NP-OB1 *T*-1)
				  (NP-SBJ (PRO we))
				  (MD should)
				  (DO do)))
              (CONJP (CONJ and)
		     (CP-QUE-SUB (WADVP-2 (WADV why))
				 (IP-SUB (ADVP *T*-2)
					 (NP-SBJ (PRO we))
					 (MD should)
					 (DO do)
					 (NP-OB1 (PRO it)))))))

( (CP-QUE-SUB (CP-QUE-SUB (WNP-1 (WPRO what))
			  (IP-SUB (NP-OB1 *T*-1)
				  (NP-SBJ (PRO we))
				  (MD should)
				  (VB take)))
              (CONJP (CONJ and)
		     (CP-QUE-SUB (WNP-2 (WADV what))
				 (IP-SUB (NP-OB1 *T*-2)
					 (NP-SBJ (PRO we))
					 (MD should)
					 (VB leave))))))

Conjunction of embedded clauses is treated as CP conjunction (rather than as IP conjunction) only if the non-first conjunct contains overt CP material (that is, a complementizer or wh- phrase).

( (CP-THT (C that)
          (IP-SUB (IP-SUB (NP-SBJ (PRO we))
			  (VBP come))
		  (CONJP (CONJ or)
			 (IP-SUB (NP-SBJ (PRO you))      ← like this
				 (VBP go))))))

( (CP-THT (CP-THT (C that)
                  (IP-SUB (NP-SBJ (PRO we)) 
			  (VBP come)))
          (CONJP (CONJ or)
		 (CP-THT (C 0)                           ← not like this
			 (IP-SUB (NP-SBJ (PRO you))
				 (VBP go))))))

Shared modifier

Shared premodifier

When the first conjunct includes pre-head modifiers, it is possible in principle for conjunction to apply at the word level or at the phrase level. In order to indicate the (potential) attachment ambiguity, the non-first conjuncts are labeled with special tags ending in X - (W)ADJX, (W)ADVX, (W)NX, (W)QX, and NUMX (the last without a wh- variant). Since non-first conjuncts in these cases are never taken to be full phrases, the first conjunct is not surrounded by phrasal brackets of its own.

The shared modifier labels are used even when the attachment ambiguity is resolved by the context.

( (NP (ADJ great) (N joy)
      (CONJP (CONJ or)
             (NX (N sorrow)))))

( (NP (Q all) (D the) (NS lords)
      (PUNC ,)
      (CONJP (NX (NS knights)))
      (PUNC ,)
      (CONJP (CONJ and)
             (NX (PRO$ their) (NS ladies)))))

( (ADJP (QR more) (ADJ cursed)
        (CONJP (CONJ and)
               (ADJX (ADJ abominable)))))

( (ADVP (ADV very) (ADJ slowly)
        (CONJP (CONJ and)
               (ADVX (ADV deliberately)))))

This convention does not extend to PP.

( (PP (PP (ADV along)
	  (P with)
	  (NP (PRO me)))
      (PUNC ,)
      (CONJP (CONJ and)
	     (PP (P with)
		 (NP (D the) (N rest))))))

Shared postmodifier

In keeping with our general principles, shared postmodifiers are attached as high as possible.

( (NP (NP (D the) (ADJ regional) (NS dialects))
      (CONJP (CONJ and)
             (NP (D the) (OTHER other) (NS varieties)))
      (PP (P of)
	  (NP (NPR English)))))

( (NP (Q all) (D the) (NS officers)
      (CONJP (CONJ and)
             (NX (NS soldiers)))
      (CP-REL (WNP-1 0)
              (C that)
              (IP-SUB (NP-SBJ *T*-1)
                      (BED were)
                      (ADVP-LOC (ADV there))))))

( (NP (D the)
      (N story)
      (CP-REL (WNP-1 0)
              (C that)
              (IP-SUB (NP-OB1 *T*-1)
                      (NP-SBJ (PRO they))
                      (VBP tell)))
      (PP (P about)					        ← daughter of NP, not IP-SUB
          (NP (PRO the)
              (N king)))))

( (PP (PP (P in)
	  (NP (D the) (ADJ regional) (NS dialects)))
      (CONJP (CONJ and)
	     (PP (P in)
		 (NP (D the) (OTHER other) (NS varieties))))
      (PP (P of)                                                ← counterintuitive attachment because of second "in"
	  (NP (NPR English)))))

Clausal conjunction and shared modifier

Because of the way that we divide sentence tokens, shared adjuncts (potentially) interpreted as belonging to two tokens must be annotated as belonging to one or the other. As a result, sentence-initial adjuncts (most commonly, clause-initial WHEN and IF clauses) appear as constituents of the first token only, while later adjuncts appear as constituents of some subsequent token only.

( (IP-MAT (ADVP-TMP (ADV Tomorrow))				← potential shared adjunct
          (PUNC ,)
	  (NP-SBJ (PRO we))
	  (MD will)
	  (VB go)
	  (PP (P tp)
	      (NP (NPR Stony) (NPR Brook)))
	  (PUNC ,)))

( (IP-MAT (CONJ and)
	  (NP-SBJ (PRO we))
	  (MD will)
	  (VB visit)
	  (NP-OB1 (PRO$ our) (NS friends))
	  (PUNC .)))


( (IP-MAT (NP-SBJ (PRO we)) (MD will) (VB go) (PP (P to) (NP (NPR Stony) (NPR Brook))) (PUNC ,))) ( (IP-MAT (CONJ and) (NP-SBJ (PRO we)) (MD will) (VB visit) (NP-OB1 (PRO$ our) (NS friends)) (ADVP-TMP (ADV tomorrow)) ← potential shared adjunct (PUNC .)))

For uniformity, the guideline extends to embedded clauses.

( (CP-THT (C that)
	  (IP-SUB (IP-SUB (ADVP-TMP (ADV tomorrow))		        ← potential shared adjunct
			  (PUNC ,)
			  (NP-SBJ (PRO we))
			  (MD will)
			  (VB go)
			  (PP (P to)
			      (NP (NPR Stony) (NPR Brook))))
		  (PUNC ,)
		  (CONJP (CONJ and)
			 (IP-SUB (NP-SBJ (PRO we))
				 (MD will)
				 (VB visit)
				 (NP-OB1 (PRO$ our) (NS friends)))))))


( (CP-THT (C that) (IP-SUB (IP-SUB (NP-SBJ (PRO we)) (MD will) (VB go) (PP (P to) (NP (NPR Stony) (NPR Brook)))) (PUNC ,) (CONJP (CONJ and) (IP-SUB (NP-SBJ (PRO we)) (MD will) (VB visit) (NP-OB1 (PRO$ our) (NS friends)) (ADVP-TMP (ADV tomorrow))))))) ← potential shared adjunct

Corresponding cases involving shared arguments are treated differently - namely, as right node raising or gapping. Certain cases might require backwards gapping as a last resort.

( (IP-MAT-1 (NP-SBJ (PRO we))
	    (MD will)
	    (VB leave)
	    (PUNC ,)
	    (IP-MAT-PRN=1 (CONJ and)
			  (NP-SBJ (PRO you))
			  (MD will)
			  (VB take))
	    (NP-OB1 (D the) (ADJR heavier) (N baggage))
	    (PUNC .)))

( (IP-MAT (IP-MAT-1 (NP-OB1 (D the) (ADJR heavier) (N baggage))
		    (PUNC ,)
		    (NP-SBJ (PRO we))
		    (MD will)
		    (VB leave))
	  (PUNC ,)
	  (CONJP (CONJ and)
		 (IP-MAT=1 (NP-SBJ (PRO you))
			   (MD will)
			   (VB take)))
	  (PUNC .)))