Data format

Verb∋Net uses VerbNet's XML format, with a few French-specific syntactic elements. The RELAX NG schema VERBENET.rnc describes it formally, and every file in the repository validates against it.

Files

verbenet/<name>-<number>.xml

One file per top-level class, e.g. put-9.1.xml. The root element is VNCLASS, and its ID is the file name without the extension. Names and numbers come from VerbNet and Levin (1993). A few classes, like dévaler-51.9, have French names because they have no English counterpart.

verbenet/VERBENET.xml

A VERBENET root element that includes every class file with <xi:include>. Run your XInclude processor on it (xmllint --xinclude, lxml's xinclude(), …) to get the whole lexicon as a single document.

verbenet/VERBENET.rnc

The RELAX NG compact schema. It accepts either a single VNCLASS or the VERBENET aggregate.

# validate one class, or all of them (jing: https://relaxng.org/jclark/jing.html)
jing -c verbenet/VERBENET.rnc verbenet/put-9.1.xml
cd verbenet && jing -c VERBENET.rnc $(ls *.xml | grep -vx VERBENET.xml)

A class, annotated

<VNCLASS ID="accept-77" ladl="6[+N1 =: Qu Psubj]" lvf="P1h.1">   ← class id, LG table(s), LVF class(es)
  <MEMBERS>
    <MEMBER name="nier"/>                         ← one verb per MEMBER
    <MEMBER name="expliquer s'"/>                 ← pronominal verbs: clitic after the verb
  </MEMBERS>
  <THEMROLES>
    <THEMROLE type="Agent">
      <SELRESTRS logic="or">                      ← +organization or +animate
        <SELRESTR Value="+" type="organization"/>
        <SELRESTR Value="+" type="animate"/>
      </SELRESTRS>
    </THEMROLE>
    <THEMROLE type="Theme"/>
  </THEMROLES>
  <FRAMES>
    <FRAME>
      <DESCRIPTION primary="NP V Qu Psubj" syntax="Agent V Theme&lt;+Qu Psubj&gt;"/>
      <EXAMPLES><EXAMPLE>Luc a accepté que son fils devienne écrivain</EXAMPLE></EXAMPLES>
      <SYNTAX>
        <NP value="Agent"><SYNRESTRS/></NP>
        <VERB/>
        <PSUBJ value="Theme"/>
      </SYNTAX>
      <SEMANTICS>approve(during(E), Agent, Theme)</SEMANTICS>
    </FRAME>
  </FRAMES>
  <SUBCLASSES>
    <VNSUBCLASS ID="accept-77-1"> … same content model, recursively … </VNSUBCLASS>
  </SUBCLASSES>
</VNCLASS>

Elements

VNCLASS, VNSUBCLASS

Contain, in this order: MEMBERS, THEMROLES, FRAMES and an optional SUBCLASSES. Subclass IDs extend the parent ID (accept-77-1, accept-77-1-1). As in VerbNet, the members of a subclass also take the roles and frames of the classes above it, so a subclass only lists what it adds.

@ladl, @lvf

The Lexique-Grammaire tables and Les Verbes Français classes that were linked to the class when it was built. Alternatives are separated by ou. LG tables may have column constraints in brackets, e.g. 6[+N1 =: Qu Psubj]. A few top-level classes lack one of the two, and about half of the subclasses have their own.

MEMBER/@name

A verb lemma, one per element. For a pronominal verb, the clitic comes after the verb: expliquer s'.

THEMROLE/@type

A thematic role from a closed list (see below). It can have a SELRESTR restriction, or a SELRESTRS group of them combined with logic="and" (the default) or "or".

DESCRIPTION

@primary is the frame in constituent notation (NP V NP PP). @syntax names the role of each constituent, e.g. Agent V Theme<+Qu Psubj>.

EXAMPLE

A French sentence illustrating the frame. Its constituents and roles should match the SYNTAX.

SYNTAX

The ordered constituents of the frame (see below).

SEMANTICS

The semantic formula, a conjunction of predicates over the event E and the roles. It is taken from VerbNet and stored as plain text, e.g. cause(Agent, E) desire(result(E), Patient, Stimulus).

Syntactic constituents

The schema requires this order: a subject (NP with an optional LUI, or impersonal IL), then VERB, then an optional ADV/PIND. After that come any number of NPs or PREP + complement pairs, and finally an optional ADJ, P, PIND, PSUBJ, VINF or ensemble. Class pages show frames with this colour code:

NPAgent V sur | dans PSUBJTheme ADJ

NP @value @modifier?

A noun phrase with its role. It always contains an empty SYNRESTRS. @modifier is one of +plural, +Adj, +genitive(Role), +Qu Pind, +Qu Psubj or +VRole-inf.

VERB @pronominal? @restr?

The verb. pronominal="true" marks a pronominal use, and @restr gives its kind: neutre (La compagnie A s'est associée à la compagnie B), middle (Les branches de pin se plient facilement), reflexive (Luc s'est caché dans le placard) or reciproque (Luc et Marie se bataillent sur ce point).

PREP

A preposition. The possible prepositions are listed in SELRESTRS/SELRESTR/@Value, separated by ; (e.g. à;avec;dans). Value="+" type="loc" stands for any locative preposition.

PSUBJ, PIND @value @introduced_by?

A finite clause in the subjunctive or the indicative (que…). @introduced_by adds a preposition (Luc compatit à ce que Marie soit au chômage) or, for PIND, si (L'inspecteur a évalué si le bâtiment allait s'écrouler).

P @value @introduced_by

An interrogative clause introduced by si, comment or combien: L'inspecteur estime combien ça peut coûter.

VINF @value @introduced_by? @emptysubjectrole?

An infinitive clause, possibly introduced by de, à or comment. @emptysubjectrole names the role that controls its implicit subject (Luc accepte de partir: the Agent is the one who leaves).

ADJ, ADV

An adjective (attribute) or an adverb.

IL, LUI, ensemble

The impersonal subject il (Il plaît à Marie qu'on fasse cela), the dative clitic lui (Luc lui a shampouiné les cheveux) and the adverb ensemble (Max a mélangé les oeufs et la crème ensemble).

Inventories

These lists are computed from the current data. The schema lists every allowed value.