Abstract
PreMOn is a freely available linguistic resource for exposing predicate models (PropBank, NomBank, VerbNet, and FrameNet) and mappings between them (e.g., SemLink and the predicate matrix) as linguistic linked open data (LOD). It consists of two components: (1) the PreMOn Ontology, that builds on the OntoLex-Lemon model by the W3C ontology-Lexica community group to enable an homogeneous representation of data from various predicate models and their linking to ontological resources; and, (2) the PreMOn Dataset, a LOD dataset integrating various versions of the aforementioned predicate models and mappings, linked to other LOD ontologies and resources (e.g., FrameBase, ESO, WordNet RDF). PreMOn is accessible online in different ways (e.g., SPARQL endpoint), and extensively documented.
This is a preview of subscription content, log in via an institution to check access.
Notes
We provide here a concise description of PreMOn. Further details are available on the PreMOn website, the ontology LODE documentation or Corcoglioniti et al. (2016b).
pm: http://premon.fbk.eu/resource/ is the namespace of PreMOn Dataset; the fragments nb10, pb215, and so on identify the resource and its version (e.g., NB 1.0, PB 2.1.5).
In case of a \(\langle \)pmo:SemanticClass, ontolex:LexicalEntry\(\rangle \) pair mapped to multiple ontology entities (e.g., in FrameBase and ESO), using ontolex:LexicalSense instead of pmo:Conceptualization would require all those entities to be owl:sameAs one another, a strong inference generally not anticipated nor intended by the authors of mappings (especially if defined independently).
PIKES is already using PreMOn identifiers for SRL annotations. (c.f. http://pikes.fbk.eu/).
We observed the presence of lexical entries with different VN conceptualizations but the same WordNet synset, which implies that the synset alone is insufficient to disambiguate among VN classes/conceptualizations and thus unambiguously relate them across VN versions.
Try it at http://bit.ly/2w8W744.
E.g., frames/verb classes and lexical units/class members both modelled as lemon LexicalSenses and owl:sameAs links between syntactic and semantic arguments (thus implying they are the same).
C.f. the group’s mailing list thread: https://lists.w3.org/Archives/Public/public-ontolex/2015Jul/0045.html.
C.f. the group’s mailing list thread: https://lists.w3.org/Archives/Public/public-ontolex/2014Jul/0021.html.
While this class does not necessarily specialize pmo:SemanticRole with additional properties or restrictions, we add it to ease the retrieval of PB-specific semantic roles, something handy when the same repository contains roles from several predicate models.
Further details on the formalization of selectional/syntactic restrictions are reported on PreMOn website.
We relied on the standard first/next/item pattern for lists.
References
Baker, C. F., Fillmore, C. J., & Lowe, J. B. (1998). The Berkeley FrameNet project. In Proceeding of COLING-ACL’98.
Burns, G. A. P. C., Hermjakob, U., & Ambite, J. L. (2016). Abstract meaning representations as linked data. In Proceeding of ISWC (pp. 12–20).
Chiarcos, C., McCrae, J., Cimiano, P., & Fellbaum, C. (2013). Towards open data for linguistics: Linguistic linked data. In A. Oltramari et al. (Eds.), New trends of research in ontologies and lexical resources (pp. 7–25). Berlin: Springer.
Corcoglioniti, F., Rospocher, M., & Aprosio, A. P. (2016a). Frame-based ontology population with PIKES. IEEE Transactions on Knowledge and Data Engineering, 28(12), 3261–3275.
Corcoglioniti, F., Rospocher, M., Aprosio, A. P., & Tonelli, S. (2016b). Premon: A lemon extension for exposing predicate models as linked data. In Proceeding of LREC.
Corcoglioniti, F., Rospocher, M., Mostarda, M., & Amadori, M. (2015). Processing billions of RDF triples on a single machine using streaming and sorting. In Proceeding of SAC (pp. 368–375).
Eckle-Kohler, J., McCrae, J. P., & Chiarcos, C. (2015). lemonUby—A large, interlinked, syntactically-rich lexical resource for ontologies. Semantic Web Journal, 6(4), 371–378.
Fellbaum, C. (1998). WordNet: An electronic lexical database. New York: Bradford Books.
Gangemi, A. (2010). What’s in a schema? In C. Huang, N. Calzolari, A. Gangemi, A. Lenci, A. Oltramari, L. Prevot (Eds.), Ontology and the lexicon (pp. 144–182). Cambridge: Cambridge University Press.
Gangemi, A., Alam, M, Asprino, L., Presutti, V., & Recupero, D. R. (2016). Framester: A wide coverage linguistic linked data hub. In Proceeding of EKAW (pp. 239–254).
Hellmann, S., Lehmann, J., Auer, S., & Brümmer, M. (2013). Integrating NLP using linked data. In Proceeding of ISWC 2013 (pp. 98–113).
Hovy, E., Marcus, M., Palmer, M., Ramshaw, L., & Weischedel, R. (2006). OntoNotes: The 90% solution. In Proceeding of HLT-NAACL (short papers) (pp. 57–60).
Lacalle, M. L. D., Laparra, E., & Rigau, G. (2014). Predicate matrix: Extending SemLink through WordNet mappings. In Proceeding of LREC.
Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press.
McCrae, J. P., Aguado-De-Cea, G., Buitelaar, P., Cimiano, P., Declerck, T., Gómez-Pérez, A., et al. (2012). Interchanging lexical resources on the semantic web. Language Resources and Evaluation, 46(4), 701–719.
McCrae, J. P., Bosque-Gil, J., Gracia, J., Buitelaar, P., & Cimiano, P. (2017). The ontolex-lemon model: Development and applications. In Proceedings of eLex 2017 conference.
Meyers, A., Reeves, R., Macleod, C., Szekely, R., Zielinska, V., Young, B., & Grishman. R. (2004). The NomBank project: An interim report. In HLT-NAACL 2004 workshop: Frontiers in corpus annotation.
Ovchinnikova, E., Vieu, L., Oltramari, A., Borgo, S., & Alexandrov, T. (2010). Data-driven and ontological analysis of framenet for natural language reasoning. In N.C.C. Chair, K. Choukri, B. Maegaard, J. Mariani, J. Odijk, S. Piperidis, M. Rosner, D. Tapias (Eds.), Proceedings of the 7th international conference on language resources and evaluation (LREC’10). Valletta: European Language Resources Association (ELRA).
Palmer, M. (2009). SemLink: Linking PropBank, VerbNet and FrameNet. In Proceeding of GenLex 2009.
Palmer, M., Gildea, D., & Kingsbury, P. (2005). The proposition bank: An annotated corpus of semantic roles. Computational Linguistics, 31(1), 71–106.
Parsons, T. (1990). Events in the semantics of English: A study in subatomic semantics. Cambridge: MIT Press.
Rospocher, M., van Erp, M., Vossen, P., Fokkens, A., Aldabe, I., Rigau, G., et al. (2016). Building event-centric knowledge graphs from news. Web Semant, 37(C), 132–151.
Rouces, J., de Melo, G., & Hose, K. (2015). FrameBase: Representing N-Ary relations using semantic frames. In Proceeding of ESWC 2015.
Schuler, K. K. (2005). Verbnet: A broad-coverage, comprehensive verb lexicon. Ph.D. thesis.
Segers, R., Vossen, P., Rospocher, M., Serafini, L., Laparra, E., & Rigau, G. (2015). ESO: A frame based ontology for events and implied situations. In Proceeding of MAPLEX workshop.
Acknowledgements
The authors would like to acknowledge the contributions of Sara Tonelli and Alessio Scussel in the development of PreMOn.
Author information
Authors and Affiliations
Corresponding author
Appendix: PreMOn Ontology resource specific modules
Appendix: PreMOn Ontology resource specific modules
1.1 PreMOn Ontology: PropBank
We define classes pmopb:Roleset and pmopb:SemanticRoleFootnote 29 as subclasses of pmo:SemanticClass and pmo:SemanticRole, respectively. Each pmopb:SemanticRole instance is related (via property pmopb:argument) to exactly one pmopb:Argument, which is defined as the disjoint union of three subclasses: pmopb:NumberedArgument, containing the individuals corresponding to numbered arguments (e.g., Arg0, Arg1); pmopb:Modifier, containing the individuals corresponding to modifiers (e.g., ArgM-LOC, ArgM-TMP); and, pmopb:SecondaryAgent, containing the single individual annotating secondary agents (ArgA). While PB annotation guidelines define a single modifier (ArgM) with multiple function tags (e.g., LOC, TMP), we opt to specialize the modifier for each function tag, similarly to the way these arguments are actually annotated by state-of-the-art SRL tools. Property pmopb:tag enables associating possible tags, either a pmopb:Modifier or some additional tag defined in class pmopb:Tag, to pmopb:SemanticRoles, or nif:Annotations of semantic roles in examples. Additional classes (and related properties) are defined to represent inflectional information about examples: pmopb:Inflection, pmopb:Person, pmopb:Tense, pmopb:Aspect, pmopb:Voice, and pmopb:Form (Fig. 6).
PreMOn Ontology–PropBank module (prefix pmopb, namespace http://premon.fbk.eu/ontology/pb#)
1.2 PreMOn Ontology: NomBank
Similarly to PB, we define pmonb:Roleset and pmonb:SemanticRole as subclasses of pmo:SemanticClass and pmo:SemanticRole, respectively. Each pmonb:SemanticRole instance is related (via property pmonb:argument) to exactly one pmonb:Argument, which is defined as the disjoint union of two subclasses: pmonb:NumberedArgument, containing the individuals corresponding to numbered argument (e.g., Arg0, Arg1), and pmonb:Modifier, containing the individuals corresponding to modifiers (e.g., ArgM-LOC, ArgM-TMP). We also define class pmonb:Tag to capture (via property pmonb:tag) some specific annotations of markables (e.g., PRD, REF, SUPPORT) in the examples (Fig. 7).
PreMOn Ontology–NomBank module (prefix pmonb, namespace http://premon.fbk.eu/ontology/nb#)
1.3 PreMOn Ontology: VerbNet
We define classes pmovn:VerbClass and pmovn:SemanticRole as subclasses of pmo:SemanticClass and pmo:SemanticRole, respectively. VN class members are modeled as instances of ontolex:LexicalEntry, connected to their class via property ontolex:evokes. The VN class hierarchy is modeled via the pmovn:subclassOf property (subproperty of skos:broader), that relates a verb class (e.g., 13.1-1) with its parent class (e.g., 13.1)—see Fig. 2 for an instantiation of this property. Given the propagation of semantic roles along the class hierarchy, we introduce property pmovn:definesSemRole to differentiate the pmovn:SemanticRole instances defined on a class from the ones inherited from its ancestor classes. Each pmovn:SemanticRole instance is related (via property pmovn:thematicRole) to exactly one pmovn:ThematicRole, which contains all the thematic roles defined in VN. These thematic roles are organized in a hierarchy, which is formalized via the skos:broader property. For instance,
states that pmovn:agent is more specific than pmovn:actor. VN selectional restrictions on pmovn:SemanticRoles (e.g., restricting “theme” to something not animate) are formalized using property pmovn:restriction and class pmovn:Restriction.Footnote 30 A verb class may have one or more pmovn:VerbNetFrames (via property pmovn:frame, or its subproperty pmovn:definesFrame, to distinguish frames defined on the class or inherited from ancestors), which have one or more orderedFootnote 31pmo:SynItems, modeling a syntactic construction (e.g., “Agent V Theme [-sentential]”) shared by all members of the class, and one or more ordered semantic pmo:Preds, modeling the meaning of the event, and its participants, expressed by the verb class for that syntactic construction (e.g., “approve(during(E), Agent, Theme)”). pmo:SynItems are specialized according to their syntactical function (e.g., pmovn:NpSynItem for noun phrases). A pmovn:NpSynItem can point (via pmo:valueObj) to a pmovn:SemanticRole, and define, via pmovn:restriction, (1) a selectional restriction holding for the pmovn:SemanticRole in that frame (e.g., “animate”), or (2) some other syntactic restriction (e.g., “np_to_inf”). Similarly, selectional restrictions can be modelled on pmovn:PrepSynItem (e.g., “spatial”). Predicates in pmovn:Pred have a type (pmovn:PredType, e.g., “approve”) and can be further decomposed in pmovn:PredArg (e.g., “during(E)”) of various types (e.g., pmovn:EventPredArg). Negation of a predicate is expressed by typing the corresponding instance as pmovn:NegPred, while implicit pmovn:PredArgs are typed as pmovn:ImplicitArg (Fig. 8).
PreMOn Ontology–VerbNet module (prefix pmovn, namespace http://premon.fbk.eu/ontology/vn#)
1.4 PreMOn Ontology: FrameNet
We define classes pmofn:Frame and pmofn:FrameElement as subclasses of pmo:SemanticClass and pmo:SemanticRole, respectively. pmofn:FrameElement is further specialized in four subclasses, denoting the four typologies of FN frame elements (e.g., pmofn:CoreFrameElement). Being pmo:SemanticRoles, in PreMOn Ontology frame elements are always specific to the frame where they are defined, even for extra thematic frame elements that are typically shared across frames in FN (e.g., the “Circumstances” extra thematic frame element corresponds to multiple individuals of type pmofn:ExtraThematicFrameElement, one for each frame where it is defined). Frame element core sets of a pmofn:Frame are represented as reified objects of type pmofn:FECoreSet, having as members some pmofn:FrameElements. Relations between pmofn:Frames are modeled using the subproperties of pmofn:frameRelation (e.g., pmofn:inheritsFrom). Similarly, mappings between pmofn:FrameElements of pmofn:Frames related via some pmofn:frameRelation are represented using frame relation-specific subproperties of pmofn:frameElementRelation (e.g., pmofn:inheritsFromFER). Within a frame, a frame element may exclude/require the presence of another frame element (pmofn:excludesFrameElement/pmofn:requiresFrameElement). pmofn:LexicalUnit, associating a lexical entry with a frame, is defined as subclass of pmo:Conceptualization. A pmofn:LexicalUnit may have a development status (pmofn:LUStatus) and can incorporate a pmofn:FrameElement (e.g., “microvawe.v’, besides evoking frame “Apply heat”, also incorporates the frame element “Heating instrument”). Finally, pmofn:Frames, pmofn:FrameElements and pmofn:LexicalUnits can be constrained according to some semantic types, defined in pmofn:SemType, and organized in a hierarchy according to pmofn:subTypeOf relations between them (Fig. 9).
PreMOn Ontology–FrameNet module (prefix pmofn, namespace http://premon.fbk.eu/ontology/fn#)
Rights and permissions
About this article
Cite this article
Rospocher, M., Corcoglioniti, F. & Palmero Aprosio, A. PreMOn: LODifing linguistic predicate models. Lang Resources & Evaluation 53, 499–524 (2019). https://doi.org/10.1007/s10579-018-9437-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10579-018-9437-8