25

Revisiting Border Conflicts between FrameNet and ... · FrameNet and Construction Grammar: annotation policies for the Brazilian Portuguese Constructicon Tiago Torrent, Ludmila Lage,

  • Upload
    buidiep

  • View
    225

  • Download
    0

Embed Size (px)

Citation preview

Revisiting Border Conflicts between FrameNet and Construction

Grammar: annotation policies for the Brazilian Portuguese Constructicon

Tiago Torrent, Ludmila Lage, Thais Sampaio, Tatiane Tavares and Ely Matos,

Universidade Federal de Juiz de Fora

Introduction •  FrameNet Brasil is creating a lexical resource

based on Frame Semantics and supported by corpus evidence in cooperation with FrameNet

•  Following the steps of FN, FrameNet Brasil has also been building a repertoire of Brazilian P o r t u g u e s e c o n s t r u c t i o n s : t h e B P Constructicon

•  So far, the Para Infinitive family of constructions has been fully described in the BP Constructicon

Research Questions •  Provided that the BP Constructicon is

conceived as a complementary resource to FN-Br… •  Which kinds of linguistic structure should count

as constructions and which could be analyzed as valence patterns of a LU?

•  How should Construction Elements be defined?

•  How should frames and constructions interact in the FN-Br database?

Objectives •  Define annotation policies for the BP

Constructicon that take into consideration the continuity between grammar and the lexicon

•  Discuss the interconnection between the Constructicon database and the Frame database

•  Contribute to the development of other FrameNet inspired Constructicons

Annotation Policy #1 •  Motivation:

•  FrameNet lexicographic annotation is constructionally inspired

•  Construction Grammar claims the continuity between grammar and the lexicon

•  Uncontrolled redundancy must be avoided in databases

Annotation Policy #1 •  Question: Given a piece of language, should

it be annotated in the Constructicon or should it be annotated only in the lexicographic database?

•  Pilot study: the Para Infinitive Family of Constructions

Annotation Policy #1 •  The case of the Manipulative Cxn

•  The Manipulative Cxn [NP VASKING para NP VINF] was described in Torrent (2009, 2011, 2012) as an argument structure verb-class specific construction

Ele pediu para o senhor contar um caso

he.NOM ask.PAST.3SG for the sir tell.INF a case

“He asked you to tell another story.”

Annotation Policy #1

Annotation Policy #1

Annotation Policy #1 •  The case of the Manipulative Cxn

•  Regardless of the valence pattern in which the verbs filling the finite V slot of the Manipulative Cxn occur, the annotated sentences exemplify the Asking frame

•  Hence, it is possible to analyze those sentences as different valence patterns of verbs evoking the Asking frame

Annotation Policy #1 •  The case of the Dative with Infinitive Cxn

•  The Dative with Infinitive Cxn [NP V NP para NP VINF] was described in Torrent (2009, 2011, 2012) as a schematic construction evoking the Ability/Enablement frame

Ela deu mil reais para mim fazer o serviço

she.NOM give.PAST.3SG thousand reais for me.DAT do.INF the job

“She gave me a thousand reais for me to do the job.”

Annotation Policy #1

Annotation Policy #1

Annotation Policy #1 •  The case of the Dative with Infinitive Cxn •  Several different verb classes evoking several

different frames occur in the DWI Cxn

•  Even if we propose Purpose_of_theme / goods / knowledge / etc Extra-Thematic FEs for each frame evoked by the verbs in this construction, such an analysis would not surpass the constructional analysis in generality

•  Also, some constructions in the Para Infinitive family are very idiomatic, and do not evoke the same frames evoked by the verbs in them

Annotation Policy #1 Being X a lexically specified material, is there any X in the construction candidate?

• NO = Construction candidate

• YES = Move to next criterion

Being F a frame and X a lexically specified material, does X evoke F?

• NO = Construction candidate

• YES = Move to next criterion

Being F a frame and X a lexically specified material, does X evoke F in a different syntactic configuration?

• NO = Construction candidate

• YES = Annotate as lexical entry only

Annotation Policy #2 •  Question: How should CEs be defined?

•  CEs in Berkeley FN’s Beyond the Core project are sometimes similar to FEs and sometimes defined as formal entities

•  Ex.: Theme in Way Cxn and Head in the Infinitival Relative Cxn

Annotation Policy #2 •  The case of the Dative with Infinitive Cxn •  The Head CE in this construction is a Resource

that enables the achievment of a Purpose

•  The VP_para_inf CE is the Purpose

•  Instead of defining CEs as form-meaning pairings, FN-Br has decided to define them as formal elements only and, in case the construction evokes a frame, map each CE to the FE it encodes

Annotation Policy #3 •  Question: Given that constructions may

evoke frames and that the verbs in them may also evoke frames, how should we account for the interaction between Cxn-evoked frames and verb-evoked frames?

•  The Non-Redundancy Principle: the same information must not be annotated twice in regards to a same set of data

Annotation Policy #3

Annotation Policy #3 •  The case of the Dative with Infinitive Cxn •  The Ability / Enablement frame is evoked by

the DWI Cxn

•  The Commerce_buy frame is evoked by the verb comprar

•  Instead of creating Extra-Thematic FEs such as Purpose_of_goods, sentences in which the verb comprar fills the finite V slot of a Dative with Infinitive Cxn shoud be annotated also in the Constructicon, which will lead to a gain in analytical generality and economy

Contributions •  Define criteria for deciding whether a given

piece of language should be annotated as a construction

•  Define criteria for the definition of CEs

•  Define guidelines for the integration of the C o n s t r u c t i c o n d a t a b a s e w i t h t h e Lexicographic database

•  Elucidate some of the methodological foundations and theoretical insights of FrameNet to non-experts

Questions?