This is part of archived UD v1 documentation. See http://universaldependencies.org/ for the current version.

home issue tracker

Regenerated 8 years ago

Aux chain

Auxiliary dependencies should not form a chain.

Search expression: _ <aux (_ <aux _)

Correct example:

edit

  

    
input
    
brat
    
info
  
Do you think that he will have left when we come ?
aux(think, Do)
aux(left, will)
aux(left, have)
{
    "text": "Do you think that he will have left when we come ? ",
    "entities": [
        [ "embedded-1-T1", "token", [ [ 0, 2 ] ] ],
        [ "embedded-1-T2", "token", [ [ 3, 6 ] ] ],
        [ "embedded-1-T3", "token", [ [ 7, 12 ] ] ],
        [ "embedded-1-T4", "token", [ [ 13, 17 ] ] ],
        [ "embedded-1-T5", "token", [ [ 18, 20 ] ] ],
        [ "embedded-1-T6", "token", [ [ 21, 25 ] ] ],
        [ "embedded-1-T7", "token", [ [ 26, 30 ] ] ],
        [ "embedded-1-T8", "token", [ [ 31, 35 ] ] ],
        [ "embedded-1-T9", "token", [ [ 36, 40 ] ] ],
        [ "embedded-1-T10", "token", [ [ 41, 43 ] ] ],
        [ "embedded-1-T11", "token", [ [ 44, 48 ] ] ],
        [ "embedded-1-T12", "token", [ [ 49, 50 ] ] ]
    ],
    "relations": [
        [ "embedded-1-R1", "aux", [ [ "arg1", "embedded-1-T3" ], [ "arg2", "embedded-1-T1" ] ] ],
        [ "embedded-1-R2", "aux", [ [ "arg1", "embedded-1-T8" ], [ "arg2", "embedded-1-T6" ] ] ],
        [ "embedded-1-R3", "aux", [ [ "arg1", "embedded-1-T8" ], [ "arg2", "embedded-1-T7" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "1" ]
}

Incorrect example:

edit

  

    
input
    
brat
    
info
  
Do you think that he will have left when we come ?
aux(think, Do)
aux(have, will)
aux(left, have)
{
    "text": "Do you think that he will have left when we come ? ",
    "entities": [
        [ "embedded-2-T1", "token", [ [ 0, 2 ] ] ],
        [ "embedded-2-T2", "token", [ [ 3, 6 ] ] ],
        [ "embedded-2-T3", "token", [ [ 7, 12 ] ] ],
        [ "embedded-2-T4", "token", [ [ 13, 17 ] ] ],
        [ "embedded-2-T5", "token", [ [ 18, 20 ] ] ],
        [ "embedded-2-T6", "token", [ [ 21, 25 ] ] ],
        [ "embedded-2-T7", "token", [ [ 26, 30 ] ] ],
        [ "embedded-2-T8", "token", [ [ 31, 35 ] ] ],
        [ "embedded-2-T9", "token", [ [ 36, 40 ] ] ],
        [ "embedded-2-T10", "token", [ [ 41, 43 ] ] ],
        [ "embedded-2-T11", "token", [ [ 44, 48 ] ] ],
        [ "embedded-2-T12", "token", [ [ 49, 50 ] ] ]
    ],
    "relations": [
        [ "embedded-2-R1", "aux", [ [ "arg1", "embedded-2-T3" ], [ "arg2", "embedded-2-T1" ] ] ],
        [ "embedded-2-R2", "aux", [ [ "arg1", "embedded-2-T7" ], [ "arg2", "embedded-2-T6" ] ] ],
        [ "embedded-2-R3", "aux", [ [ "arg1", "embedded-2-T8" ], [ "arg2", "embedded-2-T7" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "2" ]
}

Link to documentation

MWE chain

MWE dependencies should not be chained. All dependents should be attached directly to the first one.

Search expression: _ <mwe (_ <mwe _)

Link to documentation

Foreign chain

Foreign dependencies should not be chained. All dependents should be attached directly to the first one. If we wish to annotate real syntactic structure of foreign material, we must not use the foreign relation.

Search expression: _ <foreign (_ <foreign _)

Link to documentation

Conj is right-headed

Coordination dependencies should be left-headed, not right.

Search expression: _ <conj@R _

Correct example:

edit

  

    
input
    
brat
    
info
  
Bill is big and honest
conj(big, honest)
{
    "text": "Bill is big and honest ",
    "entities": [
        [ "embedded-3-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-3-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-3-T3", "token", [ [ 8, 11 ] ] ],
        [ "embedded-3-T4", "token", [ [ 12, 15 ] ] ],
        [ "embedded-3-T5", "token", [ [ 16, 22 ] ] ]
    ],
    "relations": [
        [ "embedded-3-R1", "conj", [ [ "arg1", "embedded-3-T3" ], [ "arg2", "embedded-3-T5" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "3" ]
}

Incorrect example:

edit

  

    
input
    
brat
    
info
  
Bill is big and honest
conj(honest, big)
{
    "text": "Bill is big and honest ",
    "entities": [
        [ "embedded-4-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-4-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-4-T3", "token", [ [ 8, 11 ] ] ],
        [ "embedded-4-T4", "token", [ [ 12, 15 ] ] ],
        [ "embedded-4-T5", "token", [ [ 16, 22 ] ] ]
    ],
    "relations": [
        [ "embedded-4-R1", "conj", [ [ "arg1", "embedded-4-T5" ], [ "arg2", "embedded-4-T3" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "4" ]
}

Link to documentation

Conj is left-headed

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. In contrast, this test shows examples where coordination is left-headed.

Search expression: _ <conj@L _

Link to documentation

Appos is right-headed

Apposition dependencies should be left-headed, not right.

Search expression: _ <appos@R _

Correct example:

edit

  

    
input
    
brat
    
info
  
Sam , my brother , arrived
appos(Sam, brother)
{
    "text": "Sam , my brother , arrived ",
    "entities": [
        [ "embedded-5-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-5-T2", "token", [ [ 4, 5 ] ] ],
        [ "embedded-5-T3", "token", [ [ 6, 8 ] ] ],
        [ "embedded-5-T4", "token", [ [ 9, 16 ] ] ],
        [ "embedded-5-T5", "token", [ [ 17, 18 ] ] ],
        [ "embedded-5-T6", "token", [ [ 19, 26 ] ] ]
    ],
    "relations": [
        [ "embedded-5-R1", "appos", [ [ "arg1", "embedded-5-T1" ], [ "arg2", "embedded-5-T4" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "5" ]
}

Incorrect example:

edit

  

    
input
    
brat
    
info
  
Sam , my brother , arrived
appos(brother, Sam)
{
    "text": "Sam , my brother , arrived ",
    "entities": [
        [ "embedded-6-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-6-T2", "token", [ [ 4, 5 ] ] ],
        [ "embedded-6-T3", "token", [ [ 6, 8 ] ] ],
        [ "embedded-6-T4", "token", [ [ 9, 16 ] ] ],
        [ "embedded-6-T5", "token", [ [ 17, 18 ] ] ],
        [ "embedded-6-T6", "token", [ [ 19, 26 ] ] ]
    ],
    "relations": [
        [ "embedded-6-R1", "appos", [ [ "arg1", "embedded-6-T4" ], [ "arg2", "embedded-6-T1" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "6" ]
}

Link to documentation

Copula is not VERB

Copulas should always be verbs. Not punctuation (dashes) and definitely not nominals.

Search expression: (!(VERB|AUX)) <cop _

Link to documentation

PRON is mark

Pronouns must not be attached using the mark relation. Relative pronouns must not be confused with subordinating conjunctions, even if the word is ambiguous.

Search expression: PRON <mark _

Correct example:

edit

  

    
input
    
brat
    
info
  
This is a fact that we cannot ignore . You know that we cannot ignore it .
dobj(ignore-8, that-5)
mark(ignore-15, that-12)
{
    "text": "This is a fact that we cannot ignore . You know that we cannot ignore it . ",
    "entities": [
        [ "embedded-7-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-7-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-7-T3", "token", [ [ 8, 9 ] ] ],
        [ "embedded-7-T4", "token", [ [ 10, 14 ] ] ],
        [ "embedded-7-T5", "token", [ [ 15, 19 ] ] ],
        [ "embedded-7-T6", "token", [ [ 20, 22 ] ] ],
        [ "embedded-7-T7", "token", [ [ 23, 29 ] ] ],
        [ "embedded-7-T8", "token", [ [ 30, 36 ] ] ],
        [ "embedded-7-T9", "token", [ [ 37, 38 ] ] ],
        [ "embedded-7-T10", "token", [ [ 39, 42 ] ] ],
        [ "embedded-7-T11", "token", [ [ 43, 47 ] ] ],
        [ "embedded-7-T12", "token", [ [ 48, 52 ] ] ],
        [ "embedded-7-T13", "token", [ [ 53, 55 ] ] ],
        [ "embedded-7-T14", "token", [ [ 56, 62 ] ] ],
        [ "embedded-7-T15", "token", [ [ 63, 69 ] ] ],
        [ "embedded-7-T16", "token", [ [ 70, 72 ] ] ],
        [ "embedded-7-T17", "token", [ [ 73, 74 ] ] ]
    ],
    "relations": [
        [ "embedded-7-R1", "dobj", [ [ "arg1", "embedded-7-T8" ], [ "arg2", "embedded-7-T5" ] ] ],
        [ "embedded-7-R2", "mark", [ [ "arg1", "embedded-7-T15" ], [ "arg2", "embedded-7-T12" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "7" ]
}

Incorrect example:

edit

  

    
input
    
brat
    
info
  
This is a fact that we cannot ignore .
mark(ignore-8, that-5)
{
    "text": "This is a fact that we cannot ignore . ",
    "entities": [
        [ "embedded-8-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-8-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-8-T3", "token", [ [ 8, 9 ] ] ],
        [ "embedded-8-T4", "token", [ [ 10, 14 ] ] ],
        [ "embedded-8-T5", "token", [ [ 15, 19 ] ] ],
        [ "embedded-8-T6", "token", [ [ 20, 22 ] ] ],
        [ "embedded-8-T7", "token", [ [ 23, 29 ] ] ],
        [ "embedded-8-T8", "token", [ [ 30, 36 ] ] ],
        [ "embedded-8-T9", "token", [ [ 37, 38 ] ] ]
    ],
    "relations": [
        [ "embedded-8-R1", "mark", [ [ "arg1", "embedded-8-T8" ], [ "arg2", "embedded-8-T5" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "8" ]
}

Link to documentation

Relation det is used for node that is neither DET nor PRON

The det relation is primarily intended for determiners, i.e. words tagged DET. Pronouns are tolerated at least until the borderline between the two classes is better investigated and defined.

Search expression: (!DET&!PRON) <det _

Link to documentation

Relation punct is used for node that is not PUNCT

Only nodes tagged PUNCT can be attached using the punct relation.

Search expression: !PUNCT <punct _

Link to documentation

PUNCT is attached as neither punct nor root

Nodes tagged PUNCT can only be attached using the punct relation (and exceptionally, if it is the only node in the sentence, also root).

Search expression: PUNCT (<!punct&<!root) _

Link to documentation

PRON or DET lacks the PronType feature

All pronouns and determiners should be further categorized using the PronType feature. Other POS may or may not have this feature.

Search expression: (PRON|DET)&!PronType

Link to documentation

NUM lacks the NumType feature

All numerals should be further categorized using the NumType feature. Other POS may or may not have this feature.

Search expression: NUM&!NumType

Link to documentation

VERB or AUX lacks the VerbForm feature

All verbs should be further categorized using the VerbForm feature. Other POS may or may not have this feature.

Search expression: (VERB|AUX)&!VerbForm

Link to documentation

Finite verb is not a verb

Only non-finite VerbForms are expected to appear with non-verb parts of speech (NOUN, ADJ, ADV).

Search expression: (VerbForm=Fin)&!(VERB|AUX)

Link to documentation

Finite verb lacks the Mood feature

All finite verb forms should be further categorized using the Mood feature.

Search expression: (VerbForm=Fin)&!Mood

Link to documentation

Degree feature used with a word that is neither adjective nor adverb

The Degree feature is normally associated with (a subset of) adjectives and adverbs. Other tags than ADJ and ADV are probably wrong.

Search expression: Degree&!ADJ&!ADV

Link to documentation

Maximum one subject

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. No predicate can have more than one subject. Note that subordinate clauses do not head copulas for this reason.

Search expression: _ >nsubj|>csubj|>nsubjpass|>csubjpass _ >nsubj|>csubj|>nsubjpass|>csubjpass _

Link to documentation

Maximum one direct object

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. No predicate can have more than one direct object. Ccomp counts as direct object. (To certain extent xcomp does too, but dobj can co-occur with xcomp in cases of secondary predication, thus this test does not look at xcomp.)

Search expression: _ >dobj|>ccomp _ >dobj|>ccomp _

Link to documentation

Case not dependent on nmod

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. If a word is marked as having a case dependency, its head should usually be marked as being a nominal modifier. But note that several legitimate exceptions occur, in particular, through coordination (where the head is labeled conj instead of nmod) and in nominal clauses (where the head is the main predicate of a clause). Also, in some languages case can modify also objects and other nominals.

Search expression: _ <case (_ (<!nmod&<!nmod:agent&<!conj) _)

Link to documentation

NOUN and case

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. A word marked as a NOUN should not be a case dependency.

Search expression: NOUN <case _

Link to documentation

ADP is not leaf

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. By default adposition is a leaf node attached to a nominal as case. Exceptions where adpositions are not leaves involve technical relations such as mwe and conj.

Search expression: ADP >!conj&>!cc&>!punct&>!mwe&>!foreign _

Link to documentation

Appos chain

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Apposition dependencies should not be chained. Multiple appositions should all be attached to the head, or they should be coordinated (with the first apposition as the head). Legitimate exceptions do occur, however, in the rare case when an apposition has an apposition of its own, which does not apply to head of the first apposition.

Search expression: _ <appos (_ <appos _)

Correct example:

edit

  

    
input
    
brat
    
info
  
Sam , my brother , John 's cousin , arrived
appos(Sam, brother)
appos(Sam, cousin)
{
    "text": "Sam , my brother , John 's cousin , arrived ",
    "entities": [
        [ "embedded-9-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-9-T2", "token", [ [ 4, 5 ] ] ],
        [ "embedded-9-T3", "token", [ [ 6, 8 ] ] ],
        [ "embedded-9-T4", "token", [ [ 9, 16 ] ] ],
        [ "embedded-9-T5", "token", [ [ 17, 18 ] ] ],
        [ "embedded-9-T6", "token", [ [ 19, 23 ] ] ],
        [ "embedded-9-T7", "token", [ [ 24, 26 ] ] ],
        [ "embedded-9-T8", "token", [ [ 27, 33 ] ] ],
        [ "embedded-9-T9", "token", [ [ 34, 35 ] ] ],
        [ "embedded-9-T10", "token", [ [ 36, 43 ] ] ]
    ],
    "relations": [
        [ "embedded-9-R1", "appos", [ [ "arg1", "embedded-9-T1" ], [ "arg2", "embedded-9-T4" ] ] ],
        [ "embedded-9-R2", "appos", [ [ "arg1", "embedded-9-T1" ], [ "arg2", "embedded-9-T8" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "9" ]
}

Incorrect example:

edit

  

    
input
    
brat
    
info
  
Sam , my brother , John 's cousin , arrived
appos(Sam, brother)
appos(brother, cousin)
{
    "text": "Sam , my brother , John 's cousin , arrived ",
    "entities": [
        [ "embedded-10-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-10-T2", "token", [ [ 4, 5 ] ] ],
        [ "embedded-10-T3", "token", [ [ 6, 8 ] ] ],
        [ "embedded-10-T4", "token", [ [ 9, 16 ] ] ],
        [ "embedded-10-T5", "token", [ [ 17, 18 ] ] ],
        [ "embedded-10-T6", "token", [ [ 19, 23 ] ] ],
        [ "embedded-10-T7", "token", [ [ 24, 26 ] ] ],
        [ "embedded-10-T8", "token", [ [ 27, 33 ] ] ],
        [ "embedded-10-T9", "token", [ [ 34, 35 ] ] ],
        [ "embedded-10-T10", "token", [ [ 36, 43 ] ] ]
    ],
    "relations": [
        [ "embedded-10-R1", "appos", [ [ "arg1", "embedded-10-T1" ], [ "arg2", "embedded-10-T4" ] ] ],
        [ "embedded-10-R2", "appos", [ [ "arg1", "embedded-10-T4" ], [ "arg2", "embedded-10-T8" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "10" ]
}

Link to documentation

Relation advmod used for node that is not ADV

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Advmod is intended only for adverbs. Modifiers that are called adverbial in some traditional grammars, but are in fact prepositional or noun phrases, should be attached as nmod.

Search expression: (!ADV) <advmod _

Link to documentation

Heads of an advmod are nominal

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. The heads of an adverbial modifier should normally not be nominal (noun, proper noun, numeral, or pronoun). Exceptions occur in nominal clauses, where a nominal is the main predicate and can therefore take clause adverbials. Example: This is probably an exception.

Search expression: _ <advmod (NOUN|PROPN|NUM|PRON)

Link to documentation

Acl not dependent on NOUN/PROPN/PRON

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Clausal modifiers of nouns should depend on NOUN/PROPN only; those in the following table depend on other parts of speech.

Search expression: !PRON&!NOUN&!PROPN >acl _

Link to documentation

Marked as NUM but not nummod, nmod or compound

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. If a word is marked as a numeral (POS), then it should be marked as being a nummod, compound or nmod dependency. Exceptions occur when the numeral is promoted to a higher function through ellipis. Example: Take five!

Search expression: NUM (<!nummod&<!nmod&<!compound) _

Link to documentation

Marked as nummod but not NUM

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. If a word is marked as a numeric modifier, it should be marked as a numeral (POS).

Search expression: !NUM <nummod _

Link to documentation

Marked as AUX but not aux or auxpass

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. If a word is marked as the auxiliary POS, it should be marked as either aux or auxpass dependency.

Search expression: AUX (<!aux&<!auxpass) _

Link to documentation

Marked as aux or auxpass but not AUX

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. If a word is marked as a (passive) auxiliary dependency, it should be marked as the auxiliary POS.

Search expression: !AUX (<aux|<auxpass) _

Link to documentation

Marked as cc but not CONJ, SYM or ADV

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. The relation cc is for coordinating conjunctions, hence the most expected POS tag is CONJ. Symbols such as + may occasionally replace conjunctions, and in some languages some adverbs may take this syntactic function too. However, most other tags are suspicious at least.

Search expression: !CONJ&!SYM&!ADV <cc _

Link to documentation

Cc is not leaf

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Coordinating conjunction should be attached to the first conjunct and it should not have its own dependents.

Search expression: _ < (_ <cc _)

Link to documentation

Marked as neg but not PART or ADV

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. The neg relation is used to attach negating particles (sometimes also tagged as adverbs; but we probably want to choose just one tag), such as English not, to the predicate or phrase they negate. It is not used for negative determiners such as English no in there is no chance, which should be attached as det.

Search expression: !PART&!ADV <neg _

Link to documentation

Parts of speech of expl

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. The parts of speech of words marked as being expletive dependencies.

Search expression: _ <expl _

Link to documentation

Dependents of expl

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. The dependents of a word marked as being an expletive dependency. This should be empty.

Search expression: _ < (_ <expl _)

Link to documentation

Heads of expl

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Parts of speech of words which have an expletive dependency.

Search expression: _ >expl _

Link to documentation

Heads of xcomp

DEBUGGING TEST. NONZERO HITS DOES NOT MEAN THE DATA IS INVALID. Parts of speech of words which have an open clausal complement dependency.

Search expression: _ >xcomp _

Link to documentation