Examples for enhanced CONJ tags

Defaults

Conjunction often leads to ambiguous structures. In such cases, the annotation defaults (in order of priority) are:
- attach high (concerns ambiguous modifiers such as any PPs in the structure)
- avoid gapping
- minimize structure
Illustrative examples are included in the sections below.
Conjunctions that are part of numbers that might be replaced by digits are (or will eventually be) tagged as NUM.
Where an overt conjunction cooccurs with punctuation, tag the conjunction.
Where two punctuation marks cooccur, each of which might be functioning as a conjunction (say, comma preceding dash), tag the first one.

Sentence-initial conjunction

CONJ_INITIAL indicates that the scope of conjunction is the entire token.

( (IP-MAT (CONJ_INITIAL And)
          (ADVP-TMP (ADV then))
	  (NP-SBJ (D the) (N rain))
          (VBD started)))

( (IP-MAT (CONJ_INITIAL For)
          (NP-SBJ (PRO we))
	  (HVP have)
          (ADVP-TMP (ADV always))
	  (VBN known)
          (NP-ACC (PRO it))))

CONJ_INITIAL is also used for full (= ungapped) parenthetical clauses.

( (IP-MAT (CONJ_INITIAL For)
          (NP-SBJ (PRO we))
	  (HVP have)
          (ADVP-TMP (ADV always))
	  (, ,)
	  (IP-MAT-PRN (CONJ_INITIAL and)
	  	      (NP-SBJ (PRO I))
	  	      (HVP have)
		      (VBN pointed)
		      (NP-ACC (D this))
		      (RP out)
		      (ADVP (ADV repeatedly))
	  (, ,)
	  (VBN known)
          (NP-ACC (PRO it))))

Word-level conjunction

CONJ_WORD indicates that the scope of conjunction is the next word.

(XP (X ...) (CONJ_WORD and) (X ...))				← like categories

(XP (X ...) (CONJ_WORD and) (Y ...))				← unlike categories

(ADJP (VAN interested) (CONJ_WORD and) (ADJ enthusiastic))

(NP (NS sticks) (CONJ_WORD and) (NS stones))

(NP-SBJ (NPR John) (CONJ_WORD and) (PRO I))

(ADVP-DIR (ADV hither) (CONJ_WORD and) (ADV thither))

(IP-MAT (NP-SBJ (PRO I))
	(VBD (VBD came) (CONJ_WORD and) (VBD saw) (CONJ_WORD and) (VBD conquered)))

(CP-ADV (VP (VB (VB come) (CONJ_WORD and) (VB go)))
        (C though)
        (IP-SUB he may ...))

(WNP (WPRO who) (CONJ_WORD or) (WPRO what))

In the cases of BOTH ... AND, EITHER ... OR, and NEITHER ... NOR, the first member (BOTH, EITHER, NEITHER) is tagged CONJ_PRE.

(NP (CONJ_PRE both) (NS sticks) (CONJ_WORD and) (NS stones))

(ADVP-DIR (CONJ_PRE either) (ADV hither) (CONJ_WORD or) (ADV thither))

Phrasal conjunction

CONJ_ADJP, CONJ_ADVP, etc. indicates that the scope of conjunction is an entire phrase (ADJP, ADVP, INTJP, NP, PP, QP, QTP, VP, WADJP, WADVP, WNP, WPP).

In cases of unlike category conjunction, the CONJ subtag indicates the category of the first conjunct and, by convention, the category that will end up dominating the entire conjunction structure in the parse. In other words, in this case, the CONJ looks "up" the tree to the conjunct that precedes it.

(XP (XP ...)
    (CONJP (CONJ_XP and)			← like category coordination
           (XP ...)))

(XP (XP ...)
    (CONJP (CONJ_XP and)			← unlike category coordination - note CONJ_XP
           (YP ...)))

(ADJP (ADJP (ADJ lonely))
      (CONJP (CONJ_ADJP and)			← CONJ_ADJP even though first conjunct is a single word
             (ADJP (ADJ far)
	   	   (PP (P from)
		       (NP (N home))))))

(ADJP (ADJP (ADJ far)
	    (PP (P from)
		(NP (N home))))
      (CONJP (CONJ_ADJP and)
	    ← CONJ_ADJP even though second conjunct is a single word
             (ADJP (ADJ lonely))))

(INTJP (INTJP (INTJ Jiminy) (INTJ crickets))
       (CONJP (CONJ_INTJP and)
              (INTJP (INTJ holy) (INTJ shmoly))))

( (IP-MAT (QTP (QTP (NP (NEG Not) (D that)))
               (CONJP (CONJ_QTP or)
                      (QTP (ADJ hot) (D that))))  ← not a normal constituent - hence QTP
          (NP-SBJ (PRO He))
  	  (VBD muttered)))

(NP (NP (D the) (ADJ big) (N boy))
    (CONJP (CONJ_NP and)
           (NP (D the) (ADJ little) (N girl))))

(PP (PP (RP out)
        (P of)
        (NP (D the) (N frying) (N pan)))
    (CONJP (CONJ_PP and)
           (PP (P into)
	       (NP (D the) (N fire)))))

(NP (NP (D a) (N Republican))
    (CONJP (CONJ_NP and)			← unlike category coordination - note CONJ_NP
           (ADJP (ADJ proud)
	   	 (PP (P of)
		     (NP (PRO it))))))

(CP-ADV (VP (VP (VB arrive)			← rare because VP is rare
	    	(ADVP-TMP (ADV late)))
            (CONJP (CONJ_VP but)
                   (VP (VB leave)
		       (ADVP-TMP (ADV early)))))
        (C though)
        (IP-SUB he may ...))

(WADVP (WADVP (WADV how) (WADV often))
       (CONJP (CONJ_WADVP and)			← unlike category coordination - note CONJ_WADVP
       	      (WPP (P under)
	      	   (WNP (WD what) (NS circumstances)))))

(WNP (WNP (WD which) (N person))
     (CONJP (CONJ_WNP or)
            (WNP (WD which) (N organization))))

(WPP (WPP (P under)
	  (WNP (WD what) (NS circumstances)))
     (CONJP (CONJ_WPP and)			← unlike category coordination - note CONJ_WPP
     	    (WADVP (WADV how) (WADV often))))

Intermediate-level conjunction

CONJ_ADJX, CONJ_ADVX, CONJ_NX indicates that the scope of conjunction is an intermediate projection level (ADJX, ADVX, NX). The CONJ label looks "down" the tree to the conjunct that follows it.

There shouldn't be any cases of unlike category conjunction with intermediate-level conjuncts.

(NP (D the) (ADJ big) (N boy)
    (CONJP (CONJ_NX and)
           (NX (ADJ little) (N girl))))

(NP (D the) (ADJ big) (N boy)			← CONJ_NX, not CONJ_WORD because of the "premodifiers"
    (CONJP (CONJ_NX and)
           (NX (N girl))))

(WNP (WD which) (N person)
     (CONJP (CONJ_NX or)
            (NX (N organization))))

Clause-level conjunction

CONJ_CP, CP_IP, CONJ_RRC indicates conjunction of full clauses.

( (IP-MAT (NP-SBJ (PRO He))
          (VBD said)
	  (CP-THT (CP-THT (C that / 0 )		          ← same whether C in first conjunct is overt or not
	  	          (IP-SUB (NP-SBJ (PRO he))	
		      	          (BED was)
			          (VAG coming)))
                  (CONJP (CONJ_CP but)
		  	 (CP-THT (C that)
			         (IP-SUB (NP-SBJ (PRO he))	
		      	                 (HVD had)
			                 (VBN eaten))))))))

In cases of ambiguity between conjoining two IPs or conjoining two CPs containing silent complementizers, the default is the two-IP structure, in accordance with the default of "minimize structure".

( (IP-MAT (NP-SBJ (PRO He))
          (VBD said)
	  (CP-THT (C that / 0)			          ← same whether C is overt or not
	  	  (IP-SUB (IP-SUB (NP-SBJ (PRO he))	
		                  (BED was)
			          (VAG coming)))
                          (CONJP (CONJ_IP but)			← CONJ_IP, not CONJ_CP !
			         (IP-SUB (NP-SBJ *con*)
		      	                 (HVD had)
			                 (VBN eaten))))))))

( (IP-MAT (NP-SBJ (NPR John))
          (, ,)
	  (IP-PPL (IP-PPL (VAG walking)
    	                  (PP (P along)
	                      (NP (D the) (N beach))))
                  (CONJP (CONJ_IP and)
                         (IP-PPL (VAG noticing)
		                 (NP-ACC (D a) (N glint)))))
          (, ,)
          (VBD found)
	  (NP-ACC (D the) (N treasure))))

(NP (Q all) (NS students)
    (RRC (RRC (VAG majoring)
    	      (PP (P in)
	          (NP (N linguistics))))
         (CONJP (CONJ_RRC and)
                (RRC (VAG graduating)
		     (NP-TMP (D this) (N year))))))

Conjunction in connection with gapping

CONJ_GAPPING indicates conjunction of gapped material.

( (IP-MAT (IP-MAT-1 (NP-SBJ (PRO I))
                    (HVP have)
	            (VBN stood)
	            (PP (P on)
	                (NP (D the) (N beach))))
	  (CONJP (CONJ_GAPPING and)
	  	 (IP-MAT=1 (VBN heard)
		 	   (NP-ACC (D the) (N ocean))))))

(IP-INF-1 (TO to)
          (VB pick)
	  (RP up)
	  (IP-INF=1 (CONJ_GAPPING and)
	  	    (VB put)
		    (RP down))
	  (NP-ACC (D the) (N book)))

Let's discuss cases of backwards gapping as they come up.

Various structures are systematically ambiguous between conjunction + gapping versus conjunction of (possibly extraposed) ungapped constituents. The default is to avoid gapping.

Assume a silent subject in the second conjunct

    Let (IP (IP them eat cake)
        (CONJP (CONJ_IP and)
               (IP *con* drink champagne)))

rather than:

    Let (IP (IP-1 them eat cake)
            (CONJP (CONJ_GAPPING and)
                   (IP=1 drink champagne)))

Assume a silent TO in the second conjunct

    (IP (IP to eat meat)
        (CONJP (CONJ_IP and)
               (IP * drink mead)))

rather than:

     Let (IP (IP-1 to eat meat) 
         (CONJP (CONJ_GAPPING and)
                (IP=1 drink mead)))

Assume extraposition of the second conjunct

    (IP We 
        saw
        (NP (NP Fanny)
            (CONJP-1 *T*))
        yesterday, 
        (CONJP-1 (CONJ_NP and)
	         little Edward))

rather than:

    (IP (IP-1 We saw (NP Fanny) yesterday, )
        (CONJP (CONJ_GAPPING and)
               (IP=1 little Edward)))

The default covers difficult cases like "Fanny, of course, belonging to, and little Edward, for that day", where it is unclear whether the scope of "for that day" extends to Fanny.

In some cases, however, gapping is the only analysis. This is particularly common in connection with expressions like "especially", "in particular", "rather", and so on. We take these to attach high (following our default for high attachment).

Note, however, that negation in the X NOT Y, X AND NOT Y, X BUT NOT Y construction and focus particles attach low.

(IP (IP-1 to eat beans)
    (CONJP (CONJ_GAPPING or)
           (IP=1 (ADVP rather, )
                 (NP peas))))

rather than:

(IP to
    eat 
    (NP (NP beans)
        (CONJP (CONJ_NP or)
               (NP rather, peas))))

Uncertainty

If you're unsure of which CONJ subtype is appropriate, use CONJ_X or add a comment in the tagged file.

Special cases

Etc. - Comma preceding etc. and its spelling variants is treated as punctuation, not as conjunction.

To be dealt with

FRAG
FS = conjunction at the beginning of a false start