Difference between revisions of "NorVal resources"

Revision as of 13:47, 22 September 2021

--Lars Hellan (talk) 13:31, 7 September 2021 (UTC)

This page describes resources related to the Norwegian Valence Catalogue NorVal.

An exhaustive description is given in:

Hellan, Lars, 2021. A valence catalogue for Norwegian. (In Loukanova (ed) Natural Language Processing in Artificial Intelligence - NLPinAI 2021. Springer. (To appear)

with exemplifying files in:

Supplementary Data for 'A valence catalogue for Norwegian': https://doi.org/10.18710/8U3L2U.

Overview

NorVal represents the valence frames of about 6300 Norwegian verbs. The number of frame types used in their classification is about 300, representing valence profiles like intransitive, transitive, ditransitive, copular, and more; grammatical functions such as ‘subject’, ‘object’, etc.; for an argument to be noun-headed vs. being an embedded clause (declarative, interrogative, or infinitival, in canonical or ‘extraposed’ position); for an argument to relate syntactically and semantically to the same predicate or not; occurrence of particles; all basic structures of ‘Logical Form’ as far as argument structure is reflected.

Not included in the lexical descriptions are senses, in the sense of definitions, paraphrases or synonyms, hence it is not a dictionary in the normal sense. For this reason, and due to its stripped-down formalism, we rather call it a valence catalogue.

Given that a verb can have more than one valence frame, a valence resource needs two kinds of entries, viz. entries for verb lemmas, and entries for a verb with a given frame; the latter we refer to as a ‘lexically instantiated valence frame’, for short '‘lexval'’. The former we refer to as a valpod.

The notion of 'lexval'

The format of a lexval entry is illustrated by one of the frame environments for the verb lemma huske ‘remember’ in (2), instantiating the general coding pattern in (1):

(1)  Schema of a lexval:   
      Lemma – selectedItem (if any) __ FrameType

(2)   Example of a lexval:   
      huske-på__intrObl-oblDECL 

(2) reads as an entry with the lemma huske, the selected item på (‘on’), and the frame type ‘'intransitive with oblique'’, where the oblique argument consists of the preposition på and a declarative clause, as in (3):

(3) 	Han husket på at det var søndag  ‘He remembered that it was Sunday’

The notion of 'valpod'

The format of a multivalent verb lemma entry is illustrated in (4), where each constituting lexval is represented with ‘V’ as placeholder for the lemma (‘'EqSuInf’' stands for ‘infinitive equi-controlled by subject’). The structure of a valpod is essentially a set, although represented with ordering conventions among its members (e.g., intransitives before transitives).

(4)   huske:V__intr & V-på__intrObl-oblDECL & V-på__intrObl-oblEqSuInf & V-på__intrObl-oblINTERR & V-på__intrObl-oblN & V__tr & V__tr-obDECL & V__tr-obEqSuInf & V__tr-obINTERR

There are as many valpods as there are lemmas, i.e., 6300; the number of multi-membered valpods (like (4)) is about 3360.

The notion of 'frame type'

The notation for frame types uses the system Construction Labeling (‘CL’) (cf. Hellan and Dakubu 2010, Dakubu and Hellan 2017, and Hellan 2019), which characterizes verb-headed constructions and verb valence frames through strings of symbols built up in the following way:

   Head-POS – GlobalLabel – ArgumentLabel1- ArgumentLabel2 - …

An ArgumentLabel describes a constituent of the construction or frame, while a GlobalLabel represents a categorization of the construction or frame as a whole. Thus, for the description of the object of a clause or frame we use an ArgumentLabel, while in characterizing the clause or frame as a whole as transitive we use a GlobalLabel.

ArgumentLabels are composed of a prefix indicating the grammatical function (GF) of the constituent, followed by one or more parts indicating inherent properties of the constituent. For instance, obDECL is an ArgumentLabel with ob as GF-indicating prefix and DECL indicating that the constituent is a declarative clause.

GlobalLabels consist minimally of a symbol for overall valence, such as tr for ‘transitive’, in many cases with additional symbols indicating further structure. Thus, the frame representation intrObl-oblDECL in (2) has intrObl as GlobalLabel.

The number of ArgumentLabels and GlobalLabels is fairly large, and it is essential that the meaning they carry is in each case concise and transparent. (Moreover, to be better navigable and understandable, entries using the notations ought to be accompanied by illustrations, such as short exemplifying sentences.)

The files constituting NorVal

The resource concretely consists of five files, as follows:

'lexvals-pure': a list of currently 15,753 lexval entries in the style of (2) above;

'lexvals-exemplified': the same entries as in ‘lexvals-pure’, exemplified with a short sentence;

'multivals': a list of all multi-membered valpods (defined relative to 3,360 lemmas);

'univals': a list of all uni-membered valpods (defined relative to 2,970 lemmas);

'valtypes_exemplified': a list of the 300 frame types, each with a lexval entry, and an example;

They are all text-files, the last one also in xl format. They are as yet not accessible through a web interface.

Frame types defined in NorVal

The grammatical basis for the choices of frame type labels as applied to Norwegian are described in Hellan 2021a. General descriptions of the labeling system are given in Hellan and Dakubu 2010 and Dakubu and Hellan 2017.

The system is related to an architecture of Typed Feature Structures, and a complete set of attribute-value matrices (AVMs) following this architecture can be downloaded at

  Frame types as feature structures 

in a tabular view synchronized with the table below. (These AVMs are sustained by two computational grammars based on Typed Feature Structures. One, called ‘TypeGram’ (‘tg’), is technically fairly simple, and includes specifications of valence frame types for a wider array of languages; for description and download, see TypeGram; the design of the grammar is described in Hellan (2019). The other grammar, called ‘NorSource’, is a large scale grammar for Norwegian (cf. Hellan and Bruland 2015). It includes the ‘TypeGram’ analytic design alongside the more complex architectures of standard HPSG-based computational grammars (cf. Copestake 2002) , and also produces ‘Logical Forms’ of the type represented in ‘Minimal Recursion Semantics’ (MRS; cf. Copestake et al. 2005). A web application where, e.g., the example sentences in the table below can be parsed and displayed for their MRS, is found on http://regdili.hf.ntnu.no:8081/linguisticAce/parse. )

The following table lists the frame types currently defined in NorVal, with some additions to the file valtypes_exemplified mentioned above; it will be updated from time to time. Along with each type is entered an example, an English translation of the example, a lexval instantiating the type, and the number of instances of the type in the catalogue.

Note that most types are instantiated by many more verbs than the one offered in the example sentence. Also notice that more than half of the verbs have other frames than the frame in which they occur in the example.

In the translations, a symbol 'R', 'P', 'L', 'T', 'I', or 'NS' added to a verb or a preposition means that the verb, or preposition, differs in valence from the Norwegian counterpart with respect to, respectively, reflexive, preposition, particle, finite complementizer, infinitival marker, non-split predicate.

Table

Copula constructions

Ditransitive constructions

Intransitive constructions

Transitive constructions