| { | |
| "paper_id": "W11-0113", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T05:36:40.276467Z" | |
| }, | |
| "title": "An Abstract Schema for Representing Semantic Roles and Modelling the Syntax-Semantics Interface", | |
| "authors": [ | |
| { | |
| "first": "Voula", | |
| "middle": [], | |
| "last": "Gotsoulia", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Essex", | |
| "location": {} | |
| }, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "This paper presents a novel approach to semantic role annotation implementing an entailmentbased view of the concept of semantic role. I propose to represent arguments of predicates with grammatically relevant primitive properties entailed by the semantics of predicates. Such meaning components generalise over a range of semantic relations which humans tend to express systematically through language. In a preliminary study, I show that we can model linguistic knowledge at a general, principled syntax-semantics interface by incorporating a layer of skeletal, entailment-based representation of word meaning in large-scale corpus annotation.", | |
| "pdf_parse": { | |
| "paper_id": "W11-0113", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "This paper presents a novel approach to semantic role annotation implementing an entailmentbased view of the concept of semantic role. I propose to represent arguments of predicates with grammatically relevant primitive properties entailed by the semantics of predicates. Such meaning components generalise over a range of semantic relations which humans tend to express systematically through language. In a preliminary study, I show that we can model linguistic knowledge at a general, principled syntax-semantics interface by incorporating a layer of skeletal, entailment-based representation of word meaning in large-scale corpus annotation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Large-scale lexical semantic resources that provide relational information about words have recently received much focus in the field of Natural Language Processing (NLP). In particular, data-driven models for lexical semantics require the creation of broad-coverage, hand-annotated corpora with predicateargument information, i.e. rich information about words expressing a semantic relation having argument slots filled by the interpretations of their grammatical complements. Corpora combining semantic and syntactic annotations constitute the backbone for the development of probabilistic models that automatically identify the semantic relationships, or semantic roles, conveyed by sentential constituents (Gildea and Jurafsky, 2002) . That is, given an input sentence and a target predicator the system labels constituents with general roles like Agent, Patient, Theme, etc., or more specific roles, as in (1) 1 The task of automatic semantic role labelling (or shallow semantic parsing) is a first step towards text understanding and has found use in a variety of NLP applications including information extraction (Surdeanu et al., 2003) , machine translation (Boas, 2002) , question answering (Narayanan and Harabagiu, 2004) , summarisation (Melli et al., 2005) , recognition of textual entailment relations (Burchardt and Frank, 2006) , etc.", | |
| "cite_spans": [ | |
| { | |
| "start": 710, | |
| "end": 737, | |
| "text": "(Gildea and Jurafsky, 2002)", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 911, | |
| "end": 914, | |
| "text": "(1)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 915, | |
| "end": 916, | |
| "text": "1", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 1120, | |
| "end": 1143, | |
| "text": "(Surdeanu et al., 2003)", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 1166, | |
| "end": 1178, | |
| "text": "(Boas, 2002)", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 1200, | |
| "end": 1231, | |
| "text": "(Narayanan and Harabagiu, 2004)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 1248, | |
| "end": 1268, | |
| "text": "(Melli et al., 2005)", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 1315, | |
| "end": 1342, | |
| "text": "(Burchardt and Frank, 2006)", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Corpora with semantic role labels additionally lend themselves to extraction of linguistic knowledge at the syntax-semantics interface. The range of semantic and syntactic combinatorial properties (valences) of each word in each of its senses is documented in terms of annotated corpus attestations. For instance, the valence pattern for the use of admire in (1) is shown in (2). This data enables the quantitative study of various linguistic phenomena and the investigation of the relationship between the distinct linguistic layers comprised by predicate-argument analysis. Furthermore, the formulation of generalisations over predicate-specific annotations can capture how predicates relate in terms of both semantic and syntactic features. Such syntax-semantics mappings (so-called linking generalisations) encode regularities concerning the associations of semantic roles with grammatical functions and are essential for a linguistic knowledge base for NLP applications. This paper addresses the problem of generalising over the valences of individual predicators and proposes an abstract semantic basis for the representation of participant roles. The definition of semantic notions at an appropriate level of abstraction is the prerequisite for the formulation of a general, principled syntax-semantics interface. This is in accordance with a somewhat intuitive conception of semantic roles as classificatory notions encoding semantic similarities across different types of events or situations in the world. In effect, all conceptions of semantic roles as opposed to predicate-specific roles, such as admirer-admired, posit some sort of semantic classification of arguments across predicators while indicating an acknowledgment that the syntax-semantics interface (referred to with the term linking) is not completely arbitrary. Put differently, semantic roles constitute a level of representation suitable for capturing semantic generalisations which humans tend to express systematically through language.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The structure of the paper is organised as follows. Section 2 looks at conceptions of semantic roles in state-of-the-art approaches to semantic annotation indicating problems or complications related to the question of whether or how these roles can support generalisations across predicates. Section 3 calls attention to the theoretical underpinnings of the notion of semantic role and introduces an annotation schema which departs from the traditional view of semantic roles as atomic, undecomposable categories. Following the insight of Dowty's (1991) theory of Proto-Roles, I will propose analytical representations of verbal arguments based on semantically well-founded, grammatically relevant meaning components entailed by the semantics of predicates (Proto-Role entailments). Finally, section 4 presents a study in which lexical entailments are marked in a corpus in accordance with the proposed schema. General syntax-semantics mappings are extracted from the annotated data and are formalised in abstract classes which readily encode generalisations concerning linking to syntactic form.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Semantically annotated corpora currently available for English implement two distinct approaches to the prickly notion of semantic role. The Proposition Bank (PropBank) (Kingsbury et al., 2002 ) is a one million word corpus in which predicate-argument relations are hand-annotated for every occurrence of every verb in the Wall Street Journal part of the Penn Treebank (Marcus et al., 1994) . Verb senses are distinguished informally on the basis of semantic as well as syntactic criteria. The semantic arguments of a verb are numbered sequentially. PropBank uses a common set of role labels (Arg0 up to Arg5) for all predicators, but these labels are defined on a per-verb basis, i.e. they have verb-specific meanings. Example PropBank annotations:", | |
| "cite_spans": [ | |
| { | |
| "start": 169, | |
| "end": 192, | |
| "text": "(Kingsbury et al., 2002", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 369, | |
| "end": 390, | |
| "text": "(Marcus et al., 1994)", | |
| "ref_id": "BIBREF15" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Corpora with Semantic Roles and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "( As illustrated in (3), argument labels are consistent across alternate syntactic patterns of a given predicator in a given sense. However, PropBank refrains from formalising the semantics of the role labels and does not ensure their coherence across verbs. This is particularly clear with higher numbered labels, which correspond to distinct types of participants: Arg2 marks an Instrument for break (3), a Benefactive for provide (4), and a Source for buy (5) . Lower-numbered labels denote various roles as well, but they are less arbitrary across verbs: Arg0 corresponds to traditional Agents, Experiencers, certain types of Theme, etc. which surface as subjects of transitive verbs and a class of intransitives called unergatives; Arg1, on the other hand, is assigned to objects of transitive verbs and subjects of unaccusatives and is the equivalent of traditional Patients, Themes, etc. While the PropBank corpus enables empirical insight into a variety of linguistic phenomena (e.g. variations in the grammatical expression of arguments) providing useful frequency information for the uses of predicators, it does not lend itself to extraction of a principled linguistic knowledge base with semantic generalisations across predicates. Inasmuch as no consistent mapping is ensured between a label and a semantic role, the argument labels result seriously overloaded across verbs. This explains why role recognition models have particularly poor performance in assigning the labels Arg2-Arg5. In fact, an attempt is currently made to map PropBank argument labels to semantically coherent roles specified by VerbNet (Kipper et al., 2000) (i.e. a broad-coverage verb lexicon based on Levin's (1993) classification of English verbs according to shared meaning and behaviour). Even though VerbNet specifies a small list of abstract roles (23 in total) which are intended to support generalisations, these roles are not defined as global primitives, but are meaningful only within verb classes. Because mappings of labels to semantic roles with class-specific interpretations would lead to very sparse data, argument labels are subdivided into groupings of VerbNet roles. The latter are created manually on the basis of analysis of argument use. 3 The subdivided (more coherent) PropBank labels perform better for semantic role labelling (Loper et al., 2007) .", | |
| "cite_spans": [ | |
| { | |
| "start": 459, | |
| "end": 462, | |
| "text": "(5)", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 1622, | |
| "end": 1643, | |
| "text": "(Kipper et al., 2000)", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 2248, | |
| "end": 2249, | |
| "text": "3", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 2340, | |
| "end": 2360, | |
| "text": "(Loper et al., 2007)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Corpora with Semantic Roles and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "A different paradigm for semantic role annotation is put forth by FrameNet. The Berkeley FrameNet project (Baker et al., 1998) is creating an online lexical database containing semantic descriptions of words based on Fillmore's (1985) theory of frame semantics. The basic unit of analysis is the semantic frame, i.e. a schematic representation of a stereotypical scene or situation. Each frame is associated with a set of predicates (including verbs, nouns, and adjectives) and a set of semantic roles (called Frame Elements, FEs) encoding the participants and props in the designated scene. FrameNet includes manually annotated example sentences from the British National Corpus incorporating additional layers of phrase structure and grammatical function annotation. It also includes two small corpora of full-text annotation intended to facilitate statistical analysis of frame-semantic structures. Currently it contains more than 960 frames covering more than 11,600 lexical items exemplified in more than 150,000 annotated sentences. The Judgment frame evoked by admire in (1) is shown in Table 3 FrameNet avoids the difficulties of attempting to pin down a small set of general roles. Instead Frame Elements are defined locally, i.e. in terms of frames. Frames are situated in semantic space by means of directed (asymmetric) relations. Each frame-to-frame relation associates a less dependent or more general frame (Super frame) with a more dependent or less general one (Sub frame). The hierarchical organisation of frames along with FE identities or analogs across frames are intended to enable the formulation of generalisations concerning the combinatorial properties (valences) of predicates. In practice, however, the frame hierarchy turns out to be somewhat complicated. Inheritance (i.e. the strongest semantic relation and the most plausible to propagate valence information across frames) is conditioned on complex sets of semantic components underlying frame definitions, ranging from FE membership and relations to other frames to relationships among FEs and Semantic Types on frames and FEs. 4 This kind of frame dependence based on fine-grained semantic or ontological distinctions is doomed to miss argument structure commonalities in predicates evoking frames that are related at a more abstract, essentially structural semantic level. Section 4 includes a concrete example of the complications in generalising valence information across FrameNet frames.", | |
| "cite_spans": [ | |
| { | |
| "start": 106, | |
| "end": 126, | |
| "text": "(Baker et al., 1998)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 217, | |
| "end": 234, | |
| "text": "Fillmore's (1985)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 2112, | |
| "end": 2113, | |
| "text": "4", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 1094, | |
| "end": 1101, | |
| "text": "Table 3", | |
| "ref_id": "TABREF9" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Corpora with Semantic Roles and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Researchers working in the FrameNet paradigm have proposed different approaches for abstracting over the properties of individual predicators and increasing the size of training data for semantic role labelling systems. Gildea and Jurafsky (2002) attempt to generalise the behaviour of semantically related predicates experimenting with a small set of abstract semantic roles mapped to FrameNet roles. Frank (2004) discusses the potential of applying various generalisation 'filters' to corpus-induced syntaxsemantics mappings for abstraction of a general linguistic knowledge base. The generalisations proposed by Frank are intended to apply within frames but not across frames. Baldewein et al. (2004) have trained semantic role classifiers re-using training instances of roles that are similar to the target role. As similarity measures, they use the FrameNet hierarchy, peripheral roles of FrameNet and clusters of roles constructed automatically. Matsubayashi et al. (2009) also explore various machine learning features for generalising semantic roles in FrameNet, namely role hierarchy, human-understandable descriptors of Frame Elements, Semantic Types of filler phrases, and mappings of FrameNet roles to roles of VerbNet. The experimental result of the role classification using these generalisation features shows significant improvements in the system. This is due to the fact that role generalisations can form a remedy for the severe problem of sparse data which is inherent in lexical semantic corpus annotation. Data sparseness, i.e. the insufficient coverage of the range of predicate senses and constructions within sensible sizes of manually annotated data, is a bottleneck both for acquisition of linguistic knowledge for the semantic lexicon and for automated techniques for semantic role assignment.", | |
| "cite_spans": [ | |
| { | |
| "start": 220, | |
| "end": 246, | |
| "text": "Gildea and Jurafsky (2002)", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 402, | |
| "end": 414, | |
| "text": "Frank (2004)", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 680, | |
| "end": 703, | |
| "text": "Baldewein et al. (2004)", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 952, | |
| "end": 978, | |
| "text": "Matsubayashi et al. (2009)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Corpora with Semantic Roles and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "From the presentation of different annotation projects it becomes evident that semantic role annotation is a complicated task whose product is deeply influenced by its initial design philosophy and underlying criteria. 5 Among these criteria the notion of semantic role itself is central. PropBank uses general role labels that lack semantic coherence. VerbNet and FrameNet, on the other hand, specify coherent roles at a more fine-grained level (i.e. roles with class-specific or frame-specific interpretations). In this section, I consider the linguistic contours of the concept of semantic role proposing an annotation schema based upon theoretically well-founded role concepts which meet the requirements of both generality and coherence. This schema is intended at enabling the formulation of a general syntax-semantics interface suitable for modelling the relations of predicates in terms of combinatorial features.", | |
| "cite_spans": [ | |
| { | |
| "start": 219, | |
| "end": 220, | |
| "text": "5", | |
| "ref_id": "BIBREF4" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "An Abstract Semantic Basis for the Representation of Participant Roles", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Espousing and extending Dowty's (1991) Proto-Role hypothesis, I propose to associate arguments of predicates with properties entailed by the semantics of predicates. 6 Mappings of entailments to syntactic constituents can be many-to-one. That is, an argument can be marked with one or more properties necessarily entailed by the meaning of the predicator. 7 Prepositional complements are also marked with verbal entailments to which prepositions may contribute more specific content. In this paper, I will make no attempt to formalise the content added by prepositions; prepositional semantics is represented solely in terms of the common entailment basis it shares with verbal meaning.", | |
| "cite_spans": [ | |
| { | |
| "start": 166, | |
| "end": 167, | |
| "text": "6", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 356, | |
| "end": 357, | |
| "text": "7", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "An Abstract Semantic Basis for the Representation of Participant Roles", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Each Proto-Role entailment indicates a grammatically pervasive concept, i.e. a property having direct effect on the grammatical behaviour of predicates. It is defined in terms of an abstract semantic relation underlying the lexical meaning of the predicate. Five such relations are identified in terms of which entailment-based representations are specified: Notion, Causation, Motion, Possession, Conditioning. Note that contrary to mere ontological labels, entailment-based representations encode structural characterisations of the semantics of arguments. Consider, for instance, the sentence in 1 A structural representation of the meaning of this construction will explicitly encode the relationships between each of the arguments of admire, i.e. between the NP I and the NP him, between the NP him and the PP for his bravery and his cheerfulness, and between the NP I and the PP for his bravery and his cheerfulness. By contrast, the FrameNet roles shown above do not model the fact that the semantic content of an Evaluee requires a Cognizer, or that a Reason requires both a Cognizer and an Evaluee. The view that the semantic properties underlying lexical meaning are relational in nature (i.e. they are not to be conceived entirely independently of one another) has been advocated by several researchers, among others Wechsler (1995), Pinker (1989) , Jackendoff (1990) , and Davis (2001), on whose work I build.", | |
| "cite_spans": [ | |
| { | |
| "start": 1345, | |
| "end": 1358, | |
| "text": "Pinker (1989)", | |
| "ref_id": "BIBREF20" | |
| }, | |
| { | |
| "start": 1361, | |
| "end": 1378, | |
| "text": "Jackendoff (1990)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "An Abstract Semantic Basis for the Representation of Participant Roles", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "In the rest of this section, I define a set of recurring entailments which underlie the semantics of a range of verbs displaying various syntactic patterns. Note that this set can be extended on the basis of additional primitive meaning components of the sort described above, covering the semantics of broad verb classes. situation (in every model) in which the first is true, the second is also true. For linguistic predicates, in particular, an entailment (or lexical entailment) is an analytic implication following from the meaning of the predicate in question. 7 The presence of 'necessarily' in this sentence is somewhat redundant, in that its meaning is incorporated by the notion of entailment. I insist, however, on emphasising it to indicate that semantic properties that are accidentally associated with the meaning of a particular use of a verb will not be annotated. Dowty points out that entailments of the predicate must be distinguished from what follows from any one sentence as a whole (e.g. entailments that may arise from NP meanings) (Dowty, 1991 :572, footnote 16). For example, in the sentence Mary slapped John, assuming that John is a human entity, it follows from the meaning of the sentence that John will perceive something as a result of the action of slapping. But this 'entailment' is not intrinsically tied to the meaning of slap, because the sentences Mary slapped the table or Mary slapped the corpse are also felicitous. That is, sentience of the direct object is not an essential component of the semantics of slap, in the way it is for a verb like awaken. The sentences Mary awakened the table and Mary awakened the corpse are clearly anomalous. True entailments of predicators (which are the ones that will be annotated) must be detectable in every possible environment in which the predicator is used. 8 The examples used to illustrate the proposed schema are from the British National Corpus. Some of them are slightly modified for reasons of conciseness. The predicates in (7)-(13) are represented in terms of a Notion relation. That is, they involve a Conceiver who is entailed to have a notion or perception of a Conceived participant (while the reverse entailment does not necessarily go through). 9 In situation types in which a Conceiver is entailed to have a notion of more than one participant, Conceived arguments are distinguished on the basis of their salience in the overall semantics of the predicate. For instance, test (8) intuitively lexicalises a dyadic relation between a Conceiver (tester) and a Conceived (tested) entity. A sought entity denoted by a for-PP is represented as part of a secondary Notion relation situated at the background of the primary (testing) relation. Conceived entities that are peripheral to the essential relation lexicalised by the predicate are associated with a more specific property termed Conceived background state of affairs (Conceived bsoa). These arguments receive less focus in the meaning of the predicate, in a sense that they are not absolutely necessary to understand the predicate's meaning. The representation of test (8), stereotype (10), and find out (11) in terms of two Notion relations, one of which is treated as more salient, reifies the concept of relative significance of Proto-Role properties in the verbal semantics. This concept is related to the weighting of entailments in the overall semantics of a verb, which plays a critical role in determining the syntactic patterns in which the verb appears (i.e. the grammatical realisations of its arguments). 10 The verbs in (8) and 9involve an additional entailment of Intentionality. This is used to mark entities characterised by conscious choice, decision, or control over the course of inherently intentional actions. Intentional participants necessarily have a notion/perception of some event participant(s). The annotations in (12) and (13) include the Entity and Property tags which are intended to distinguish Conceived arguments in terms of a predicative relation assigned in the Conceiver's mental model. The Property label corresponds to a representation of the form P(x) denoting a property P which is predicated of some object x.", | |
| "cite_spans": [ | |
| { | |
| "start": 567, | |
| "end": 568, | |
| "text": "7", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 1056, | |
| "end": 1068, | |
| "text": "(Dowty, 1991", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 1842, | |
| "end": 1843, | |
| "text": "8", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 2243, | |
| "end": 2244, | |
| "text": "9", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 3156, | |
| "end": 3160, | |
| "text": "(11)", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 3569, | |
| "end": 3571, | |
| "text": "10", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 3585, | |
| "end": 3588, | |
| "text": "(8)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 3894, | |
| "end": 3898, | |
| "text": "(12)", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 3903, | |
| "end": 3907, | |
| "text": "(13)", | |
| "ref_id": "BIBREF12" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "An Abstract Semantic Basis for the Representation of Participant Roles", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "The entailments of Notion are not applicable in the semantics of the predicates in (14)-(15) below. These verbs refer to situations with affected participants and are described in terms of an abstract relation of Causation. In the denoted events, a Causer is entailed to affect some entity (the Causee) either physically or mentally. Causally affected participants sometimes undergo radical changes in their (physical or mental) state, which are identified in terms of a readily observable transition from a source to a final (result) state, as shown in (15) . Verbs of caused Motion (19) or caused Possession (20) are represented in terms of both Causation and Motion/Possession, i.e. as meaning 'cause to move' (set to motion) or 'cause to possess'. This analysis posits a main (causal) event and a caused sub-event. The entailments associated with the latter are marked in square brackets. (19) Proto-Role entailments are defined in terms of inherently asymmetric semantic relations involving fixed role positions. Each of these relations (with the exception of Motion) can be thought of as instance of a more general relation entailing that properties of an entity \u03b2 are dependent on an entity \u03b1. For example, a conceived entity in a Notion relation depends on the existence of a conceiver (it is taken to be within the scope of the conceiver's beliefs). An affected or possessed object in a causation or possession relation depends on the existence of some causer or possessor, respectively. I refer to this relation as Conditioning relation and associate it with appropriate Proto-Role properties capturing the semantics of a broad range of verbs for which none of the entailments specified so far seems to hold. These verbs conform to the basic transitivity pattern that motivated Dowty's Proto-Role hypothesis. Below are some characteristic examples: A Conditioning relation encodes the asymmetries in such predicators in terms of the underlying entailment that the properties of a participant \u03b1 impose a condition on properties of a participant \u03b2. In each of the sentences above we can conclude something about the object participant (e.g. that it is necessary, illegal, or linguistically expressed) on the basis of the subject referent (i.e. the characteristics of the game, the regulations specified by the code, the usage of the adjective 'beautiful'). By contrast, no property of the subject referent is necessarily conditioned on the object: the semantics of ban, for example, does not allow us to characterise code 1425 as fair/unfair, severe/lax, complete/incomplete, new/old, etc. on the basis of the object NP 'large trucks in tunnels'; similarly, we cannot infer the precise meaning of the word 'beautiful' or whether it is a verb or a noun or an adjective on the basis of the content of the NP 'a quality which can be found in many different objects'. A more precise definition of the Conditioning relation could state that the intrinsic (i.e. invariable) properties of a participant \u03b1 determine or condition some non-intrinsic (i.e. variable or event-dependent) property of a participant \u03b2 while the converse entailment does not go through. In (24), for example, the sociolinguistics domain is associated with a property of being diverse whereas the intrinsic properties of the domain have no significance for the definition of 'diversity' or what this notion may characterise.", | |
| "cite_spans": [ | |
| { | |
| "start": 554, | |
| "end": 558, | |
| "text": "(15)", | |
| "ref_id": "BIBREF14" | |
| }, | |
| { | |
| "start": 584, | |
| "end": 588, | |
| "text": "(19)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 610, | |
| "end": 614, | |
| "text": "(20)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 893, | |
| "end": 897, | |
| "text": "(19)", | |
| "ref_id": "BIBREF18" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "An Abstract Semantic Basis for the Representation of Participant Roles", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "A preliminary study has been carried out mapping state-of-the-art semantic role annotations to lexical entailment representations. In particular, a portion of the FrameNet corpora has been annotated with Proto-Role properties by a single annotator. The study focuses on a set of English verbs selected from 250 random FrameNet frames. For each verb in these frames, collections of example annotated sentences as well as sentences from the FrameNet full-text annotation corpora (where available) were extracted. More than 900 lexical units were considered in \u223c20K sentences. Proto-Role entailments were annotated on top of FrameNet's syntactic annotations in accordance with the schema sketched out above. The annotations were produced semi-automatically following a three-stage procedure: (i) mapping Frame Elements (FEs) to entailments at a frame level (ii) automatically adding this information to the data in a new annotation layer, (iii) manually correcting the novel annotations by examining the argument structures of individual predicators for finer semantic distinctions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Formulation of a General Syntax-Semantics Interface", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "From the newly annotated data mappings of entailments to grammatical categories were acquired. The syntactic realisations of Proto-Role properties were found to readily generalise over combinatorial features of verbs pertaining to various FrameNet frames. Valence information can be formally rendered in entailment-based classes called Lexicalisation Types (L-Types) abstracting away from the semantics of predicators. L-Types are defined on the basis of grammatically relevant meaning components and encode linking generalisations cutting across FrameNet frames.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Formulation of a General Syntax-Semantics Interface", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "For instance, predicates such as believe and desire (evoking the frames Religious Belief and Desiring, respectively) involve arguments that are equivalent in terms of entailments, as illustrated in (25)-(26) below. Hence they are categorised in the Notion L-Type shown in Table 2 . In a similar fashion, operate, research, and ratify can be grouped together in a L-Type based on the underlying property of Intentionality. Examples (27)-(28) show that these verbs share common valence patterns despite the differences in the definition of the frames they evoke (Using, Research and Ratification): Role and Purpose are core Frame Elements in the Using frame, while Purpose is peripheral in Research and Ratification. Research and Ratification have no Role FE (but this kind of argument is clearly present in the constructions exemplified in (28b-c). In the same Intentionality L-Type we also categorise verbs such as carry out and visit evoking the frames Intentionally act and Visiting. It is important to note that despite the argument structure similarities of these predicators, it is not possible to establish an identity link between the Act FE of the Intentionally act frame and the Entity FE of Visiting in terms of the frame hierarchy, because the FEs are associated with different Semantic Types in the corresponding frame definitions, i.e. Act is of type State of affairs whereas Entity is of type Physical object. The examples (29)-(30) illustrate the common use of these verbs in the transitive construction. The (a) sentences show the FE annotation while the (b) sentences show the annotated entailments. Predicates grouped together in L-Types have some but not necessarily all their grammatical properties in common. This is in accordance with the fact that L-Types are essentially semantically-driven modelling recurring, abstract features in the semantics of predicators while disregarding ephemeral properties as well as lexical idiosyncrasies. 11 In addition to the set of entailments discussed in the previous section, L-Types may also incorporate more fine-grained properties that are clearly relevant to linking. For instance, verbs lexicalising a Desiring situation were found with prepositional complements introduced by for, after, to, towards, of, or over (e.g. long for, hanker after, aspire to, pine over, etc.), but not on, upon, at, or about (like other Notion verbs, such as ponder, muse, think, etc.). Inasmuch as a Desiring relation is identified as a recurring concept systematically associated with a particular grammatical relation (e.g. a for-PP), it can be represented in a separate L-Type inheriting from the Notion L-Type presented previously. 12 An initial classification like the one exemplified above captures general conditions which determine possible associations between the semantics of predicators and grammatical relations realising their arguments (e.g. the fact that a conceived entity can only surface in subject position in a passive sentence). It can be extended and refined on the basis of more specific semantic relations. Moreover, L-Types can be organised in hierarchical structures. They can form the upper portion of a principled hierarchy of classes encoding successively broader levels of generalisations concerning argument linking.", | |
| "cite_spans": [ | |
| { | |
| "start": 1961, | |
| "end": 1963, | |
| "text": "11", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 2682, | |
| "end": 2684, | |
| "text": "12", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 272, | |
| "end": 279, | |
| "text": "Table 2", | |
| "ref_id": "TABREF7" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Formulation of a General Syntax-Semantics Interface", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "This study indicated that a small number of Lexicalisation Types abstracts over a wide range of FrameNet frames. 13 More precisely, in the annotated dataset 48 L-Types were identified based on various combinations of entailments: 9 Notion Types, 7 Intentionality Types, 10 Causation Types, 7 Communication (Caused Notion) Types, 7 Motion (including Caused Motion) Types, 7 Possession (including Caused Possession) Types, and 1 Conditioning Type. These Types readily abstract over associations of semantic properties and grammatical functions attested in over 200 FrameNet frames. 14 In the FrameNet paradigm, L-Types can be modelled as non-lexicalised frames specifying syntactic mapping constraints.", | |
| "cite_spans": [ | |
| { | |
| "start": 113, | |
| "end": 115, | |
| "text": "13", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 580, | |
| "end": 582, | |
| "text": "14", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Formulation of a General Syntax-Semantics Interface", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Mappings between FrameNet frames and L-Types can be stated by means of a separate relation in addition to the frame relations currently specified by FrameNet. A relation generalising the combinatorial properties of lexical items across frames would simplify the picture of the frame hierarchy, in that it would essentially decouple purely lexical semantic information (encoded by existing frame-to-frame relations) from information pertaining exactly to the interface of syntax and semantics. In future work, our intention is to test whether the proposed semantic role schema and the attested L-Types can be useful for dealing with the sparse data problem and increasing the performance of semantic role labelling systems.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Formulation of a General Syntax-Semantics Interface", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "This annotated example is from the FrameNet lexicon (discussed in the next section). In all examples throughout the paper, predicators are marked in italics.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "ArgM-TMP indicates a temporal adjunct modifier.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "This endeavour is part of the SemLink project which aims at developing computationally explicit connections between lexical semantic resources (PropBank, VerbNet, FrameNet, WordNet). The idea is to combine the advantages of these resources and overcome their limitations by bridging the complementary lexical information they offer. In a related vein, the LIRICS (i.e. Linguistic Infrastructure for Interoperable Resources and Systems) project has recently evaluated several approaches for semantic role annotation (PropBank, VerbNet, FrameNet, among others) aiming to propose ISO ratified standards for semantic representation that will enable the exchange and reuse of (multilingual) language resources(Petukhova and Bunt, 2008).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Semantic Types encode information that is not representable in terms of frames and FE hierarchies, e.g. basic typing of fillers of FEs referring to some (external) ontological classification, descriptions of aspects of semantic variation between lexical units such as the Positive and Negative types in the Judgment frame above, etc.5 This point is discussed in detail by Ellsworth et al., 2004.6 The term entailment is used in the standard logical sense according to which one formula entails another if in every possible", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The Notion relation, as defined byWechsler (1995), essentially reconstructs the entailment of sentience, which was proposed byDowty (1991).10 Arguments identified as conceived bsoas have many of the syntactic properties of so-called semantic adjuncts. However, I refrain from invoking an argument versus adjunct division, in that it is known to involve serious theoretical pitfalls. Instead I classify conceived participants on the basis of the concept of importance of entailments, which lies exactly at the syntaxsemantics interface. This concept is defined in terms of the lexicalised event rather than the real-world event that traditional analyses of adjuncthood appeal to.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "L-Types crucially differ from verb classes in VerbNet, which are based on a rigorous commitment to syntax. This commitment yields fine-grained distinctions that very often split semantically coherent classes. In fact, L-Types abstract over VerbNet classes encoding broader levels of linking generalisations.12 For-PPs are indeed associated with a desiderative sense with a wide range of verbs in various argument positions: 'He desperately hunted for a new job '. 'They searched the ground for traces'. 'John ran for cover when it started to rain'.13 Note that inasmuch as L-Types abstract over both VerbNet classes and FrameNet frames, they can also be useful for combining the two resources.14 About 30 frames contained predicates for which none of our entailments seemed to hold. Most of these verbs (e.g. resemble, adjoin, concern, fit, suit, etc.) involve what Dowty (1991) called perspective-dependent semantic roles traditionally described with labels such asFigure andGround. The lexicalisation patterns of these verbs have been shown to depend on pragmatic or discourse factors rather than intrinsic semantic properties. Such predicates display great variability in their argument realisation options and are outside the scope of this study.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "The Berkeley FrameNet project", | |
| "authors": [ | |
| { | |
| "first": "Collin", | |
| "middle": [ | |
| "F" | |
| ], | |
| "last": "Baker", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Charles", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "John", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Fillmore", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Lowe", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Proceedings of the COLING-ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Baker, Collin F., Charles J. Fillmore, and John B. Lowe. 1998. The Berkeley FrameNet project. In Proceedings of the COLING-ACL. Montreal, Canada.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Semantic Role Labeling with Similarity-Based Generalisation Using EM-Based Clustering", | |
| "authors": [ | |
| { | |
| "first": "Ulrike", | |
| "middle": [], | |
| "last": "Baldewein", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Katrin", | |
| "middle": [], | |
| "last": "Erk", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sebastian", | |
| "middle": [], | |
| "last": "Pad\u00f3", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Detlef", | |
| "middle": [], | |
| "last": "Prescher", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of Senseval-3", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "64--68", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Baldewein, Ulrike, Katrin Erk, Sebastian Pad\u00f3, and Detlef Prescher. 2004. Semantic Role Labeling with Similarity-Based Generalisation Using EM-Based Clustering. In Proceedings of Senseval-3, pp. 64-68. Barcelona, Spain.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Bilingual FrameNet Dictionaries for Machine Translation", | |
| "authors": [ | |
| { | |
| "first": "Hans", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Boas", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the Third International Conference on Language Resources and Evaluation", | |
| "volume": "IV", | |
| "issue": "", | |
| "pages": "1364--1371", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Boas, Hans C. 2002. Bilingual FrameNet Dictionaries for Machine Translation. In Proceedings of the Third International Conference on Language Resources and Evaluation. Las Palmas, Spain. Vol. IV: pp. 1364-1371.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Approximating Textual Entailment with LFG and FrameNet Frames", | |
| "authors": [ | |
| { | |
| "first": "Aljoscha", | |
| "middle": [], | |
| "last": "Burchardt", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Anette", | |
| "middle": [], | |
| "last": "Frank", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the second PASCAL Recognizing Textual Entailment Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "92--97", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Burchardt, Aljoscha and Anette Frank. 2006. Approximating Textual Entailment with LFG and FrameNet Frames. In Proceedings of the second PASCAL Recognizing Textual Entailment Workshop. Venice, Italy, pp. 92-97.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Linking by types in the hierarchical lexicon", | |
| "authors": [ | |
| { | |
| "first": "Anthony", | |
| "middle": [], | |
| "last": "Davis", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Davis, Anthony. 2001. Linking by types in the hierarchical lexicon. CSLI Publications.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Thematic Proto-Roles and Argument Selection", | |
| "authors": [ | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Dowty", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1991, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "547--619", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dowty, David. 1991. Thematic Proto-Roles and Argument Selection. Language 67.3, pp. 547-619.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "PropBank, SALSA, and FrameNet: How Design Determines Product", | |
| "authors": [ | |
| { | |
| "first": "Ellsworth", | |
| "middle": [], | |
| "last": "Michael", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Katrin", | |
| "middle": [], | |
| "last": "Erk", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Kingsbury", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sebastian", | |
| "middle": [], | |
| "last": "Pad\u00f3", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the LREC 2004 Workshop on Building Lexical Resources from Semantically Annotated Corpora", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ellsworth Michael, Katrin Erk, Paul Kingsbury, and Sebastian Pad\u00f3. 2004. PropBank, SALSA, and FrameNet: How Design Determines Product. In Proceedings of the LREC 2004 Workshop on Building Lexical Resources from Semantically Annotated Corpora. Lisbon.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Frames and the semantics of understanding", | |
| "authors": [ | |
| { | |
| "first": "Charles", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Fillmore", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1985, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "222--254", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fillmore, Charles J. 1985. Frames and the semantics of understanding. Quaderni di Semantica 6.2, pp. 222-254.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Generalizations over corpus-induced frame assignment rules", | |
| "authors": [ | |
| { | |
| "first": "Anette", | |
| "middle": [], | |
| "last": "Frank", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the LREC 2004 Workshop on Building Lexical Resources from Semantically Annotated Corpora", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "31--38", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Frank, Anette. 2004. Generalizations over corpus-induced frame assignment rules. In Charles Fillmore, Manfred Pinkal, Collin Baker and Katrin Erk (eds.): Proceedings of the LREC 2004 Workshop on Building Lexical Resources from Semantically Annotated Corpora. Lisbon, Portugal, pp. 31-38.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Automatic labeling of semantic roles", | |
| "authors": [ | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Gildea", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Jurafsky", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Computational Linguistics", | |
| "volume": "28", | |
| "issue": "3", | |
| "pages": "245--288", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gildea, Daniel and Daniel Jurafsky. 2002. Automatic labeling of semantic roles. Computational Linguistics 28 (3), pp. 245-288.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Semantic Structures. Cambridge, MA", | |
| "authors": [ | |
| { | |
| "first": "Ray", | |
| "middle": [], | |
| "last": "Jackendoff", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA, MIT Press.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "From Treebank to PropBank", | |
| "authors": [ | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Kingsbury", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the LREC", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kingsbury, Paul and Martha Palmer. 2002. From Treebank to PropBank. In Proceedings of the LREC, Las Palmas, Canary Islands, Spain.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Class-based construction of a verb lexicon", | |
| "authors": [ | |
| { | |
| "first": "Karin", | |
| "middle": [], | |
| "last": "Kipper", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hoa", | |
| "middle": [ | |
| "Trang" | |
| ], | |
| "last": "Dang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Proceedings of the Seventh National Conference on Artificial Intelligence (AAAI-2000)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kipper, Karin, Hoa Trang Dang, and Martha Palmer. 2000. Class-based construction of a verb lexicon. In Proceedings of the Seventh National Conference on Artificial Intelligence (AAAI-2000), Austin,TX, July-August.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "English verb classes and alternations: A preliminary investigation", | |
| "authors": [ | |
| { | |
| "first": "Beth", | |
| "middle": [], | |
| "last": "Levin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Levin, Beth. 1993. English verb classes and alternations: A preliminary investigation. University of Chicago Press.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Combining Lexical Resources: Mapping Between PropBank and VerbNet. Proceedings of the 7th International Workshop on Computational Semantics", | |
| "authors": [ | |
| { | |
| "first": "Edward", | |
| "middle": [], | |
| "last": "Loper", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Szu-Ting", | |
| "middle": [], | |
| "last": "Yi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Loper, Edward, Szu-ting Yi and Martha Palmer. 2007. Combining Lexical Resources: Mapping Between PropBank and VerbNet. Pro- ceedings of the 7th International Workshop on Computational Semantics. Tilburg, the Netherlands.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "The Penn Treebank: Annotating predicate argument structure", | |
| "authors": [ | |
| { | |
| "first": "Mitchell", | |
| "middle": [], | |
| "last": "Marcus", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Grace", | |
| "middle": [], | |
| "last": "Kim", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mary", | |
| "middle": [ | |
| "Ann" | |
| ], | |
| "last": "Marcinkiewicz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Robert", | |
| "middle": [], | |
| "last": "Macintyre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ann", | |
| "middle": [], | |
| "last": "Bies", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Ferguson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Karen", | |
| "middle": [], | |
| "last": "Katz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Britta", | |
| "middle": [], | |
| "last": "Schasberger", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Proceedings AR-PAHLT Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marcus, Mitchell, Grace Kim, Mary Ann Marcinkiewicz, Robert MacIntyre, Ann Bies, Mark Ferguson, Karen Katz, and Britta Schas- berger. 1994. The Penn Treebank: Annotating predicate argument structure. In Proceedings AR-PAHLT Workshop.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "A Comparative Study on Generalization of Semantic Roles in FrameNet", | |
| "authors": [ | |
| { | |
| "first": "Yuichiroh", | |
| "middle": [], | |
| "last": "Matsubayashi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Naoaki", | |
| "middle": [], | |
| "last": "Okazaki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "19--27", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Matsubayashi, Yuichiroh, Naoaki Okazaki, and Jun'ichi Tsujii. 2009. A Comparative Study on Generalization of Semantic Roles in FrameNet. In Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pp.19-27. Suntec, Singapore.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Description of SQUASH, the SFU question answering summary handler for the DUC-2005 Summarization Task", | |
| "authors": [ | |
| { | |
| "first": "Gabor", | |
| "middle": [], | |
| "last": "Melli", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yang", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yudong", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Mehdi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhongmin", | |
| "middle": [], | |
| "last": "Kashani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Baohua", | |
| "middle": [], | |
| "last": "Shi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Gu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Proceedings of the HLT/EMNLP Document Understanding Workshop (DUC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Melli, Gabor, Yang Wang, Yudong Liu, Mehdi M. Kashani, Zhongmin Shi, Baohua Gu, Anoop Sarkar, and Fred Popowich. 2005. Description of SQUASH, the SFU question answering summary handler for the DUC-2005 Summariza- tion Task. In Proceedings of the HLT/EMNLP Document Understanding Workshop (DUC). Vancouver, Canada, available at http://duc.nist.gov/pubs/2005papers/simonfraseru.sarkar.pdf.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Question Answering Based on Semantic Structures", | |
| "authors": [ | |
| { | |
| "first": "Srini", | |
| "middle": [], | |
| "last": "Narayanan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sanda", | |
| "middle": [], | |
| "last": "Harabagiu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the 20th International Conference on Computational Linguistics (COLING)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "693--701", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Narayanan, Srini and Sanda Harabagiu. 2004. Question Answering Based on Semantic Structures. In Proceedings of the 20th International Conference on Computational Linguistics (COLING), pp. 693 -701. Geneva, Switzerland.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "LIRICS semantic role annotation: design and evaluation of a set of data categories", | |
| "authors": [ | |
| { | |
| "first": "Volha", | |
| "middle": [], | |
| "last": "Petukhova", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Harry", | |
| "middle": [], | |
| "last": "Bunt", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC 2008)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Petukhova, Volha and Harry Bunt. 2008. 'LIRICS semantic role annotation: design and evaluation of a set of data categories.' In Pro- ceedings of the Sixth International Conference on Language Resources and Evaluation (LREC 2008), Marrakech, Morocco, May 28-30.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "Learnability and Cognition", | |
| "authors": [ | |
| { | |
| "first": "Steven", | |
| "middle": [], | |
| "last": "Pinker", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pinker, Steven. 1989. Learnability and Cognition. Cambridge, MA, MIT Press.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Using Predicate Arguments Structures for Information Extraction", | |
| "authors": [ | |
| { | |
| "first": "Mihai", | |
| "middle": [], | |
| "last": "Surdeanu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sanda", | |
| "middle": [], | |
| "last": "Harabagiu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "John", | |
| "middle": [], | |
| "last": "Williams", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Aarseth", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Surdeanu, Mihai, Sanda Harabagiu, John Williams, and Paul Aarseth. 2003. Using Predicate Arguments Structures for Information Extraction. In Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, pp. 815. Sapporo, Japan.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "The semantic basis of argument structure", | |
| "authors": [ | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Wechsler", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wechsler, Stephen. 1995. The semantic basis of argument structure. Stanford, CA. CSLI Publications.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "Cognizer: Noun Phrase (NP), Subject Evaluee: Noun Phrase (NP), Object Degree: Adverbial Dependent Reason: Prepositional Dependent", | |
| "num": null | |
| }, | |
| "FIGREF1": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "[ Conceiver Opposition members] accuse [ Conceived,Entity the council] [ Conceived,P roperty of acting purely ideologically].", | |
| "num": null | |
| }, | |
| "FIGREF2": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "[ Causer Diet] influences [ Causee disease]. (15) [ Causer The sun] has changed [ Causee,Change of state her hair color] [ Source state from red] [ End state to blue]. Verbs as in (16)-(17) are represented in terms of a Motion relation involving a Moving entity (i.e. an object entailed to change location) and Stationary reference frame. Locations at the start, end, or intermediate points of the stationary frame are tagged with the labels Path source, Path goal, and Path, respectively. (16) [ M oving The car] passed [ Stationary the railway station].(17) [ M oving The river] flowed silently [ P ath through the forest]. Finally, verbs such as own, possess, acquire, lack, etc. are treated in terms of a Possession relation involving a Possessor and an entity entailed to be Possessed (18). (18) [ P ossessor This house] lacks [ P ossessed a guest room].", | |
| "num": null | |
| }, | |
| "FIGREF3": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "[ Condition This game] demands [ Conditioned great skill]. (22) [ Condition Code 1425] bans [ Conditioned large trucks in tunnels]. (23) [ Condition The adjective 'beautiful'] denotes [ Conditioned a quality which can be found in many different objects]. (24) [ Condition Diversity] characterises [ Conditioned the sociolinguistics domain].", | |
| "num": null | |
| }, | |
| "TABREF0": { | |
| "content": "<table><tr><td>.</td></tr><tr><td>(1) [</td></tr></table>", | |
| "text": "Cognizer I] admired [ Evaluee him] [ Degree greatly] [ Reason for his bravery and his cheerfulness].", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF1": { | |
| "content": "<table><tr><td>(5) In addition, [ Arg0 the bank] has an option to buy [ Arg1 a 30% stake in BIP] [ Arg2 from Societe</td></tr><tr><td>Generale] [ ArgM \u2212T M P after Jan.1, 1990] [ Arg3 at 1,015 francs a share]. 2</td></tr></table>", | |
| "text": "3) a. [ Arg0 John] broke [ Arg1 the window] [ Arg2 with a rock]. Arg2 to the industrial average].", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF3": { | |
| "content": "<table/>", | |
| "text": "", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF4": { | |
| "content": "<table><tr><td>, repeated here as</td></tr><tr><td>(6):</td></tr><tr><td>(6) [</td></tr></table>", | |
| "text": "Cognizer I] admired [ Evaluee him] [ Reason for his bravery and his cheerfulness].", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF5": { | |
| "content": "<table/>", | |
| "text": "[ Conceiver The other two] pondered [ Conceived over this morsel] as they tramped along behind him.", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF6": { | |
| "content": "<table/>", | |
| "text": "Causer Lucie] threw [ Causee,[M oving] him] [ [P ath source] from the parapet of a bridge] [ [P ath goal] into deep water]. (20) [ Causer He] handed [ [P ossessed] the letter] [ [P ossessor] to Weir], who nodded.", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF7": { | |
| "content": "<table><tr><td>Notion L-Type</td><td colspan=\"2\">Religious belief Desiring</td></tr><tr><td>Conceiver</td><td>Believer</td><td>Experiencer</td></tr><tr><td>Conceived, (Entity)</td><td>Element</td><td>Focal participant</td></tr><tr><td colspan=\"2\">Conceived bsoa, Property Role</td><td>Role of focal participant</td></tr></table>", | |
| "text": "includes the correspondences between combinations of entailments and FrameNet Frame Elements.", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF8": { | |
| "content": "<table/>", | |
| "text": "", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF9": { | |
| "content": "<table/>", | |
| "text": "Mappings between Intentionality L-Type and FrameNet frames (27) a. [ Conceiver,Intentional We] operate [ Conceived a menu] [ Conceived bsoa,Intention to get the best out of rations]. b. [ Conceiver,Intentional We] research [ Conceived this fungus] [ Conceived bsoa,Intention to fight aliments in tobacco and tomato fields]. c. [ Conceiver,Intentional They] had to ratify [ Conceived the amendments] [ Conceived bsoa,Intention to be readmitted to the Union]. Conceived,Entity Thin films] are being researched [ Conceived bsoa,P roperty as a potential medium for integrated optical circuits]. c. [ Conceived,Entity Such agreements] may be ratified [ Conceived bsoa,P roperty as being in the public interest].", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| }, | |
| "TABREF10": { | |
| "content": "<table/>", | |
| "text": "(29) a. [ Agent They] had carried out [ Act 113 uranium conversion experiments]. b. [ Conceiver,Intentional They] had carried out [ Conceived 113 uranium conversion experiments]. (30) a. [ Agent You] have to visit [ Entity your parents] every once in a while.", | |
| "html": null, | |
| "type_str": "table", | |
| "num": null | |
| } | |
| } | |
| } | |
| } |