Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
A Computational Grammar for Georgian
Paul Meurer
AKSIS, University of Bergen, Norway
Seventh International Tbilisi Symposiumon Language, Logic and Computation
Tbilisi, October 1 – 5, 2007
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 1 / 54
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 2 / 54
Introduction: The Georgian grammar project
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 3 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
The Georgian grammar project
is aimed at developing a full-scale computational grammar forGeorgian
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 4 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
The Georgian grammar project
is aimed at developing a full-scale computational grammar forGeorgian
uses LFG (Lexical Functional Grammar) as a syntactic formalism(output of a parse consists of c- and f-structures, no semanticrepresentation yet)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 4 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
The Georgian grammar project
is aimed at developing a full-scale computational grammar forGeorgian
uses LFG (Lexical Functional Grammar) as a syntactic formalism(output of a parse consists of c- and f-structures, no semanticrepresentation yet)
uses XLE (Xerox Linguistic Environment) as a parsing engine
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 4 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
The Georgian grammar project
is aimed at developing a full-scale computational grammar forGeorgian
uses LFG (Lexical Functional Grammar) as a syntactic formalism(output of a parse consists of c- and f-structures, no semanticrepresentation yet)
uses XLE (Xerox Linguistic Environment) as a parsing engine
uses visualization, treebanking and corpus tools developed atAksis/University of Bergen
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 4 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
The Georgian grammar project
is aimed at developing a full-scale computational grammar forGeorgian
uses LFG (Lexical Functional Grammar) as a syntactic formalism(output of a parse consists of c- and f-structures, no semanticrepresentation yet)
uses XLE (Xerox Linguistic Environment) as a parsing engine
uses visualization, treebanking and corpus tools developed atAksis/University of Bergen
is part of the international Parallel Grammar (ParGram) project,which coordinates the development of LFG grammars in a parallelmanner using XLE
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 4 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
Time frames and status quo:
started around 1995 with morphology, 2005 with syntax
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 5 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
Time frames and status quo:
started around 1995 with morphology, 2005 with syntax
most of the morphology is covered (missing: proper names;compounds)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 5 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
Time frames and status quo:
started around 1995 with morphology, 2005 with syntax
most of the morphology is covered (missing: proper names;compounds)
most basic syntactic constructions are covered (incomplete:subcategorization frames, constructions involving infinite forms,appositions, much more)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 5 / 54
Introduction: The Georgian grammar project
A computational grammar for Georgian
Time frames and status quo:
started around 1995 with morphology, 2005 with syntax
most of the morphology is covered (missing: proper names;compounds)
most basic syntactic constructions are covered (incomplete:subcategorization frames, constructions involving infinite forms,appositions, much more)
funding: none
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 5 / 54
Lexical-Functional Grammar
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 6 / 54
Lexical-Functional Grammar Introduction
Lexical-Functional Grammar (LFG)
a generative linguistic framework
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 7 / 54
Lexical-Functional Grammar Introduction
Lexical-Functional Grammar (LFG)
a generative linguistic framework
initiated by Joan Bresnan and Ronald Kaplan in the 1970s toovercome conceptual and explanatory shortcomings ofChomsky’s transformational grammar
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 7 / 54
Lexical-Functional Grammar Introduction
Lexical-Functional Grammar (LFG)
a generative linguistic framework
initiated by Joan Bresnan and Ronald Kaplan in the 1970s toovercome conceptual and explanatory shortcomings ofChomsky’s transformational grammar
constraint-based; no transformations
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 7 / 54
Lexical-Functional Grammar Introduction
Lexical-Functional Grammar (LFG)
a generative linguistic framework
initiated by Joan Bresnan and Ronald Kaplan in the 1970s toovercome conceptual and explanatory shortcomings ofChomsky’s transformational grammar
constraint-based; no transformations
rigid formalism, well-suited for implementation and efficientparsing, as well as for theoretical work
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 7 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:C(onstituent)-structure: phrase structure tree (often modeledaccording to Xbar-syntax principles)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:C(onstituent)-structure: phrase structure tree (often modeledaccording to Xbar-syntax principles)F(unctional)-structure: attribute-value matrix which recursivelycorrelates the semantic argument structure of predicates withgrammatical functions
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:C(onstituent)-structure: phrase structure tree (often modeledaccording to Xbar-syntax principles)F(unctional)-structure: attribute-value matrix which recursivelycorrelates the semantic argument structure of predicates withgrammatical functions
C- and f-structure are related by a projection relation
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:C(onstituent)-structure: phrase structure tree (often modeledaccording to Xbar-syntax principles)F(unctional)-structure: attribute-value matrix which recursivelycorrelates the semantic argument structure of predicates withgrammatical functions
C- and f-structure are related by a projection relation
Structural relations are formally described by phrase structurerules annotated with functional equations
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure
parallel description of linguistic entities by:C(onstituent)-structure: phrase structure tree (often modeledaccording to Xbar-syntax principles)F(unctional)-structure: attribute-value matrix which recursivelycorrelates the semantic argument structure of predicates withgrammatical functions
C- and f-structure are related by a projection relation
Structural relations are formally described by phrase structurerules annotated with functional equations
Lexical items (lemmas and morphological features) are annotatedwith a lexical category and functional equations (includingargument structures for verbs)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 8 / 54
Lexical-Functional Grammar Introduction
C- and F-structure: Example
bavšvebichildren
tamašoben.they-play.
‘The children are playing.’
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 9 / 54
Lexical-Functional Grammar Introduction
C- and F-structure: Example
bavšvebichildren
tamašoben.they-play.
‘The children are playing.’
c-structure
IP
NP
N
bavšvebi
I
tamašoben
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 9 / 54
Lexical-Functional Grammar Introduction
C- and F-structure: Example
bavšvebichildren
tamašoben.they-play.
‘The children are playing.’
c-structure
IP
NP
N
bavšvebi
I
tamašoben
f-structure2
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘tamaši<[1:bavšvi]>’
SUBJ
1
2
6
4
PRED ‘bavšvi’CASE nomNUM pl
3
7
5
TNS-ASP
2
6
4
TENSE presASPECT imperfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 9 / 54
Lexical-Functional Grammar Introduction
C- and F-structure: Example
bavšvebichildren
tamašoben.they-play.
‘The children are playing.’
c-structure
IP
NP
N
bavšvebi
I
tamašoben
f-structure2
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘tamaši<[1:bavšvi]>’
SUBJ
1
2
6
4
PRED ‘bavšvi’CASE nomNUM pl
3
7
5
TNS-ASP
2
6
4
TENSE presASPECT imperfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 9 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysis
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
lexemes and morphological features are entries in the LFG lexicon
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
lexemes and morphological features are entries in the LFG lexiconthey are annotated with (sub)lexical categories and functionalequations
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
lexemes and morphological features are entries in the LFG lexiconthey are annotated with (sub)lexical categories and functionalequationsthese are used to initialize the parse chart
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
lexemes and morphological features are entries in the LFG lexiconthey are annotated with (sub)lexical categories and functionalequationsthese are used to initialize the parse chart
Chart parsing with phrase structure rules
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Lexical-Functional Grammar Parsing with XLE
Parsing with XLE
Parsing steps:
Tokenization
Morphological analysisLexical insertion
lexemes and morphological features are entries in the LFG lexiconthey are annotated with (sub)lexical categories and functionalequationsthese are used to initialize the parse chart
Chart parsing with phrase structure rules
Solving of functional equations
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 10 / 54
Morphology and Morphosyntax
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 11 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analyses
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
New implementation based on fst (Xerox finite state tool)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
New implementation based on fst (Xerox finite state tool)automatically derived from old implementation
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
New implementation based on fst (Xerox finite state tool)automatically derived from old implementationflag diacritics mimic feature structure unification; compiled out atthe end ⇒ pure finite state
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
New implementation based on fst (Xerox finite state tool)automatically derived from old implementationflag diacritics mimic feature structure unification; compiled out atthe end ⇒ pure finite statelexicon compiled into the transducer
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
Parsing modelFirst approach: Finite state transducer augmented with featurestructure unification
Disjunctive unification with a lexicon of existing forms to discardnonexisting verb analysesImplemented in Common Lisp, based on Parc Xerox’s old fsamodule
New implementation based on fst (Xerox finite state tool)automatically derived from old implementationflag diacritics mimic feature structure unification; compiled out atthe end ⇒ pure finite statelexicon compiled into the transducerinterfaces well with XLE
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 12 / 54
Morphology and Morphosyntax Morphology
Morphology
The lexicon
Derived from Kita Tschenkélis ‘Georgisch-Deutsches Wörterbuch’(52 000 entries, 3 823 verb entries) and other sources (74 000nouns and adjectives altogether)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 13 / 54
Morphology and Morphosyntax Morphology
Morphology
The lexicon
Derived from Kita Tschenkélis ‘Georgisch-Deutsches Wörterbuch’(52 000 entries, 3 823 verb entries) and other sources (74 000nouns and adjectives altogether)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 13 / 54
Morphology and Morphosyntax Morphology
Morphology
Typical analyses:
‘wine’gvino → gvino+N+Nom+Sg
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 14 / 54
Morphology and Morphosyntax Morphology
Morphology
Typical analyses:
‘wine’gvino → gvino+N+Nom+Sg
‘for the girls, too’gogo-eb-isa-tvis-ac → gogo+N+Anim+Full+Gen+Pl+Tvis+C
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 14 / 54
Morphology and Morphosyntax Morphology
Morphology
Typical analyses:
‘wine’gvino → gvino+N+Nom+Sg
‘for the girls, too’gogo-eb-isa-tvis-ac → gogo+N+Anim+Full+Gen+Pl+Tvis+C
‘in childhood’bavšvob-isa-s → bavšvoba+N+DGen+DSg+Dat+Sg
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 14 / 54
Morphology and Morphosyntax Morphology
Morphology
Typical analyses:
‘wine’gvino → gvino+N+Nom+Sg
‘for the girls, too’gogo-eb-isa-tvis-ac → gogo+N+Anim+Full+Gen+Pl+Tvis+C
‘in childhood’bavšvob-isa-s → bavšvoba+N+DGen+DSg+Dat+Sg
‘I apparently painted it’/‘he will paint it for me’da-mi-xat.-av-s →
{ da-xat.va-3569-5+V+Trans+Perf+Subj1Sg+Obj3| da-xat.va-3569-18+V+Trans+Perf+Subj1Sg+Obj3| da-xat.va-3569-18+V+Trans+Fut+Subj3Sg+Obj1Sg } s
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 14 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
Each verb lexeme in the LFG lexicon is associated with one or moresubcategorization frames (argument structures) and a mapping ofeach of the arguments to a grammatical function (one of SUBJ(ect),OBJ(ect), OBJben(eficiary), OBL(ique), etc.).
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 15 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
Each verb lexeme in the LFG lexicon is associated with one or moresubcategorization frames (argument structures) and a mapping ofeach of the arguments to a grammatical function (one of SUBJ(ect),OBJ(ect), OBJben(eficiary), OBL(ique), etc.).
ga-v-u-k. et-eb ‘I will do it for him/her’:
ga-k.eteba<agent, benefic, theme>
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 15 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
Each verb lexeme in the LFG lexicon is associated with one or moresubcategorization frames (argument structures) and a mapping ofeach of the arguments to a grammatical function (one of SUBJ(ect),OBJ(ect), OBJben(eficiary), OBL(ique), etc.).
ga-v-u-k. et-eb ‘I will do it for him/her’:
ga-k.eteba<agent, benefic, theme>
ga-k.eteba<SUBJ, OBJben, OBJ>
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 15 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
Each verb lexeme in the LFG lexicon is associated with one or moresubcategorization frames (argument structures) and a mapping ofeach of the arguments to a grammatical function (one of SUBJ(ect),OBJ(ect), OBJben(eficiary), OBL(ique), etc.).
ga-v-u-k. et-eb ‘I will do it for him/her’:
ga-k.eteba<agent, benefic, theme>
ga-k.eteba<SUBJ, OBJben, OBJ>
The verb classification in Tschenkéli’s Georgisch–deutschesWörterbuch could be used directly to automatically derive apreliminary version of the Georgian LFG verb lexicon
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 15 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
Each verb lexeme in the LFG lexicon is associated with one or moresubcategorization frames (argument structures) and a mapping ofeach of the arguments to a grammatical function (one of SUBJ(ect),OBJ(ect), OBJben(eficiary), OBL(ique), etc.).
ga-v-u-k. et-eb ‘I will do it for him/her’:
ga-k.eteba<agent, benefic, theme>
ga-k.eteba<SUBJ, OBJben, OBJ>
The verb classification in Tschenkéli’s Georgisch–deutschesWörterbuch could be used directly to automatically derive apreliminary version of the Georgian LFG verb lexicon
Example: Tschenkéli’s class T3 maps to the argument structure
P<SUBJ, OBJben, OBJ>
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 15 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
In many cases the correct frames are not (easily) deducible fromTschenkéli´s classification:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 16 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
In many cases the correct frames are not (easily) deducible fromTschenkéli´s classification:
verbs taking oblique or genitive arguments:
ca-tvla<SUBJ, OBJ, OBLadv > ‘consider sb. to be sth.’še-šineba<SUBJ, OBJgen> ‘be afraid of sb./sth.’
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 16 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
In many cases the correct frames are not (easily) deducible fromTschenkéli´s classification:
verbs taking oblique or genitive arguments:
ca-tvla<SUBJ, OBJ, OBLadv > ‘consider sb. to be sth.’še-šineba<SUBJ, OBJgen> ‘be afraid of sb./sth.’
Class III verbs: can be transitive and intransitive
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 16 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
In many cases the correct frames are not (easily) deducible fromTschenkéli´s classification:
verbs taking oblique or genitive arguments:
ca-tvla<SUBJ, OBJ, OBLadv > ‘consider sb. to be sth.’še-šineba<SUBJ, OBJgen> ‘be afraid of sb./sth.’
Class III verbs: can be transitive and intransitive
verbs taking clausal arguments
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 16 / 54
Morphology and Morphosyntax Morphosyntax
From lexeme to grammatical function
In many cases the correct frames are not (easily) deducible fromTschenkéli´s classification:
verbs taking oblique or genitive arguments:
ca-tvla<SUBJ, OBJ, OBLadv > ‘consider sb. to be sth.’še-šineba<SUBJ, OBJgen> ‘be afraid of sb./sth.’
Class III verbs: can be transitive and intransitive
verbs taking clausal arguments
morphological passives: can be passives or unaccusatives
ga-k. et-deba (mtavrobis mier): ‘it will be done (by the government)’ga-k.eteba<OBL-AG, SUBJ>
da-brun-deba (*dedis mier): ‘(s)he will return’da-bruneba<SUBJ>
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 16 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment
st.udent.-mastudent.ERG
c.eril-iletter.NOM
mo-m-c.er-a.he-wrote-it-to_me.
The student wrote me a letter.
mi-c.era< SUBJ, OBJben, OBJ >
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 17 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment
st.udent.-mastudent.ERG
c.eril-iletter.NOM
mo-m-c.er-a.he-wrote-it-to_me.
The student wrote me a letter.
mi-c.era< SUBJ, OBJben, OBJ >
st.udent.ma.ERG c.erili.NOM m-
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 17 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment
st.udent.-mastudent.ERG
c.eril-iletter.NOM
mo-m-c.er-a.he-wrote-it-to_me.
The student wrote me a letter.
mi-c.era< SUBJ, OBJben, OBJ >
st.udent.ma.ERG c.erili.NOM m-
?
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 17 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment
st.udent.-mastudent.ERG
c.eril-iletter.NOM
mo-m-c.er-a.he-wrote-it-to_me.
The student wrote me a letter.
mi-c.era< SUBJ, OBJben, OBJ >
st.udent.ma.ERG c.erili.NOM m-
?
Georgian is head- and dependent-marking: verbal affixes and nominalcase code grammatical functions.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 17 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment: Facts
Case alignmentpatterns
SUBJ OBJ OBJbenA NOM DAT DAT
B ERG NOM DAT
C DAT NOM -tvis
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 18 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment: Facts
Case alignmentpatterns
SUBJ OBJ OBJbenA NOM DAT DAT
B ERG NOM DAT
C DAT NOM -tvis
Alignmentdepending onverb class andtense group
I II III IV
trans. unacc. unerg. indir.present A A A Caorist B A B Cperfect C A C C
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 18 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment: Facts
Case alignmentpatterns
SUBJ OBJ OBJbenA NOM DAT DAT
B ERG NOM DAT
C DAT NOM -tvis
Alignmentdepending onverb class andtense group
I II III IV
trans. unacc. unerg. indir.present A A A Caorist B A B Cperfect C A C C
Person/numberaffix alignmentpatterns
SUBJ OBJ OBJbenA = B v- (FSUBJ) m- (FOBJ) h- (FOBJ)
C h- (FOBJ) v- (FSUBJ) -
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 18 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment: Implementation
Affixes: Alignment coded in the morphology
Example: 1st person plural (morphological) subject marker
gv- → +FObj1Pl → +Subj1Pl (perfect)→ +Obj1Pl (otherwise)
ga-gv-i-k. et-eb-i-a ‘we apparently did it’
→ ga-k.eteba-904-3+V+Perf+FSubj3+FObj1Pl+Trans
→ ga-k.eteba-904-3+V+Perf+Subj1Pl+Obj3+Trans
Functional equations are attached to morphology features:
+Subj1Pl: (↑ SUBJ PERS) = 1(↑ SUBJ NUM) = pl.
+Obj1Pl: (↑ _MORPH-SYNT _AGR _OBJ PERS) = 1(↑ _MORPH-SYNT _AGR _OBJ NUM) = pl.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 19 / 54
Morphology and Morphosyntax Morphosyntax
Case and affix alignment: Implementation
Case: Alignment coded in the syntax
Equations attached to verb lexicon entry
Example: transitive/unergative subject
{ (↑ SUBJ PRED) = ‘pro’| @(ifelse (↑ _TENSEGROUP) =c pres
[ (↑ SUBJ CASE) = nom ][ @(ifelse (↑ _TENSEGROUP) =c aor
[ (↑ SUBJ CASE) = erg ][ (↑ SUBJ CASE) = dat ] ) ] ) }.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 20 / 54
Some aspects of Georgian syntax
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 21 / 54
Some aspects of Georgian syntax Free word order, Information structure
Free word order
‘Free word order’ at the phrase level
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 22 / 54
Some aspects of Georgian syntax Free word order, Information structure
Free word order
‘Free word order’ at the phrase level
Subject and complements cannot be distinguishedconfigurationally: no VP
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 22 / 54
Some aspects of Georgian syntax Free word order, Information structure
Free word order
‘Free word order’ at the phrase level
Subject and complements cannot be distinguishedconfigurationally: no VP
Finite verb and other constituents can occur in almost arbitraryorder; or: an arbitrary permutation of the toplevel constituents of agrammatical sentence results in a grammatical sentence with thesame propositional truth value
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 22 / 54
Some aspects of Georgian syntax Free word order, Information structure
Free word order
‘Free word order’ at the phrase level
Subject and complements cannot be distinguishedconfigurationally: no VP
Finite verb and other constituents can occur in almost arbitraryorder; or: an arbitrary permutation of the toplevel constituents of agrammatical sentence results in a grammatical sentence with thesame propositional truth value
This is to be expected: since grammatical functions are codedmorphologically, there is no need to repeat the codingconfigurationally
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 22 / 54
Some aspects of Georgian syntax Free word order, Information structure
Free word order
‘Free word order’ at the phrase level
Subject and complements cannot be distinguishedconfigurationally: no VP
Finite verb and other constituents can occur in almost arbitraryorder; or: an arbitrary permutation of the toplevel constituents of agrammatical sentence results in a grammatical sentence with thesame propositional truth value
This is to be expected: since grammatical functions are codedmorphologically, there is no need to repeat the codingconfigurationally
⇒ First approximation:
S → V, XP*
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 22 / 54
Some aspects of Georgian syntax Free word order, Information structure
Information structure and Discurse functions
Position is significant for Information structure: it is used to code thediscurse functions FOCUS and TOPIC.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 23 / 54
Some aspects of Georgian syntax Free word order, Information structure
Information structure and Discurse functions
Position is significant for Information structure: it is used to code thediscurse functions FOCUS and TOPIC.
FOCUS: immediately in front of inflected verb or in last position
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 23 / 54
Some aspects of Georgian syntax Free word order, Information structure
Information structure and Discurse functions
Position is significant for Information structure: it is used to code thediscurse functions FOCUS and TOPIC.
FOCUS: immediately in front of inflected verb or in last position
TOPIC: initial position(s) to the left of FOCUS and verb
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 23 / 54
Some aspects of Georgian syntax Free word order, Information structure
Information structure and Discurse functions
Position is significant for Information structure: it is used to code thediscurse functions FOCUS and TOPIC.
FOCUS: immediately in front of inflected verb or in last position
TOPIC: initial position(s) to the left of FOCUS and verb
⇒ Revision of phrase structure rules (compliant with Xbar theory,Bresnan 2001):
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 23 / 54
Some aspects of Georgian syntax Free word order, Information structure
Information structure and Discurse functions
Position is significant for Information structure: it is used to code thediscurse functions FOCUS and TOPIC.
FOCUS: immediately in front of inflected verb or in last position
TOPIC: initial position(s) to the left of FOCUS and verb
⇒ Revision of phrase structure rules (compliant with Xbar theory,Bresnan 2001):
I → finite verb
I’ → I (S)
S → XP+
IP → (XP) I’
IP → XP IP
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 23 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
I is the category of the finite (inflected) verb:
I → finite verb
IP
I
dac. era
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 24 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
The nonprojective category S is the Complement of I:
I’ → I (S)
IP
I’
I
dac. era
S
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 25 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
S contains all material to the right of the verb:
S → XP+
IP
I’
I
dac. era
S
NP
N
c. erili
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 26 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
S contains all material to the right of the verb:
S → XP+
IP
I’
I
dac. era
S
NP
N
st.udent.ma
NP
N
c. erili
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 27 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
The Specifier of I is often TOPIC or FOCUS position:
IP → (XP) I’
IP
ADVP
ADVtmp
gušin
TOPIC?FOCUS?
I’
I
dac. era
S
NP
N
c. erili
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 28 / 54
Some aspects of Georgian syntax Free word order, Information structure
Phrase structure rules
Material adjoint to IP is TOPIC:
IP → XP IP
IP
NP
N
st.udent.ma
TOPIC
IP
NP
N
c. erili
TOPIC
IP
ADVP
ADVtmp
gušinFOCUS?
I
dac. era
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 29 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
c.erililetter.NOM
dac.era.he-wrote-it.AOR
‘He wrote a letter.’
IP
NP
N
c. erili
I
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 30 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
NP
N
c. erili
I
dac. era+Trans
+Subj3Sg
+Obj3
+Aor
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 31 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
NP
N
c. erili+Nom
+Sg
I
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 32 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
(↑ GF)=↓
NP
N
c. erili
I
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 33 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
(↑ GF)=↓
NP
N
c. erili
I
{ (↑ SUBJ PRED)=’pro’| (↑ SUBJ CASE)=erg }{ (↑ OBJ PRED)=’pro’| (↑ OBJ CASE)=nom }
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 34 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
(↑ GF)=↓
NP
N
c. erili
I
{ (↑ SUBJ PRED)=’pro’| (↑ SUBJ CASE)=erg }{ (↑ OBJ PRED)=’pro’| (↑ OBJ CASE)=nom }
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 35 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop and case syntax: Example
‘He wrote a letter.’
IP
(↑ GF)=↓
NP
N
c. erili
I
{ (↑ SUBJ PRED)=’pro’| (↑ SUBJ CASE)=erg }{ (↑ OBJ PRED)=’pro’| (↑ OBJ CASE)=nom }
dac. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘da-c.era<[1:pro],[2:c.erili]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ2
"
PRED ‘c.erili’PERS 3, NUM sg, CASE nom
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 36 / 54
Some aspects of Georgian syntax Pro-drop and case syntax
Pro-drop: Example
‘She wrote it to me.’
IP
I
momc. era
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘mi-c.era<[1:pro],[2:pro],[3:pro]>’
SUBJ1
"
PRED ‘pro’PERS 3, NUM sg
#
OBJ3
"
PRED ‘pro’PERS 3
#
OBJth2
"
PRED ‘pro’PERS 1, NUM sg
#
TNS-ASP
2
6
4
TENSE aorASPECT perfMOOD indicative
3
7
5
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 37 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination
Predicate coordination: two verbs coordinate at V level (as opposed tosentence coordination with ellipsis), and share their arguments.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 38 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination
Predicate coordination: two verbs coordinate at V level (as opposed tosentence coordination with ellipsis), and share their arguments.
1 Crosslinguistically common restriction: Both verbs should assignthe same semantic roles and grammatical functions to theirshared arguments.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 38 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination
Predicate coordination: two verbs coordinate at V level (as opposed tosentence coordination with ellipsis), and share their arguments.
1 Crosslinguistically common restriction: Both verbs should assignthe same semantic roles and grammatical functions to theirshared arguments.
2 Frequent additional restriction: The case of a common argumentshould be licensed by both verbs, or, when there is casesyncretism, be compatible with both verbs’ requirements.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 38 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination
Predicate coordination: two verbs coordinate at V level (as opposed tosentence coordination with ellipsis), and share their arguments.
1 Crosslinguistically common restriction: Both verbs should assignthe same semantic roles and grammatical functions to theirshared arguments.
2 Frequent additional restriction: The case of a common argumentshould be licensed by both verbs, or, when there is casesyncretism, be compatible with both verbs’ requirements.
In Georgian: less restrictive conditions:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 38 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination
Predicate coordination: two verbs coordinate at V level (as opposed tosentence coordination with ellipsis), and share their arguments.
1 Crosslinguistically common restriction: Both verbs should assignthe same semantic roles and grammatical functions to theirshared arguments.
2 Frequent additional restriction: The case of a common argumentshould be licensed by both verbs, or, when there is casesyncretism, be compatible with both verbs’ requirements.
In Georgian: less restrictive conditions:1 The case of an argument needs only to be licensed by the verb
nearest to it.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 38 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
mashe.DAT
[uqvarsloves
daand
apasebs]esteems
tavishis-own
meugle-s.wife.DAT
‘He loves and esteems his own wife.’
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 39 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
mashe.DAT
[uqvarsloves
daand
apasebs]esteems
tavishis-own
meugle-s.wife.DAT
‘He loves and esteems his own wife.’
uqvars (IV) apasebs (I)‘he loves her’ ‘he esteems her’
thematic roles < exp, theme > < exp, theme >functions SUBJ, OBJ SUBJ, OBJ
case marking DAT, NOM NOM, DAT
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 39 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
*meI
[miqvarsI-love-her
daand
mapasebs]she-esteems-me
cemimy
meugle.wife.NOM
‘I [love, and am esteemed by,] my own wife.’
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 40 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
*meI
[miqvarsI-love-her
daand
mapasebs]she-esteems-me
cemimy
meugle.wife.NOM
‘I [love, and am esteemed by,] my own wife.’
uqvars (IV) apasebs (I)‘he loves her’ ‘he esteems her’
thematic roles < exp, theme > < exp, theme >functions SUBJ, OBJ SUBJ, OBJ
case marking DAT, NOM NOM, DAT
*
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 40 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
??misihis
mogvac.eobapublic-activity
mosc.onsit-likes-it
daand
pasdebait-is-esteemed
xalxisthe-people
mier.by
‘His public activity is liked and esteemed by the people.’
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 41 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Examples
??misihis
mogvac.eobapublic-activity
mosc.onsit-likes-it
daand
pasdebait-is-esteemed
xalxisthe-people
mier.by
‘His public activity is liked and esteemed by the people.’
mosc.ons (IV) pasdeba (II)‘he likes it’ ‘it is esteemed’
case marking DAT, NOM mier, NOM
functions SUBJ, OBJ OBL, SUBJ
thematic roles < exp, theme > < exp, theme >
*
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 41 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
Simple case: both verbs license the same cases
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 42 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
Simple case: both verbs license the same cases
giamGia
iqidabought
daand
c.aik. itxaread
esthis
c.igni.book.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 42 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
Simple case: both verbs license the same cases
giamGia
iqidabought
daand
c.aik. itxaread
esthis
c.igni.book.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 42 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
Simple case: both verbs license the same cases
giamGia
iqidabought
daand
c.aik. itxaread
esthis
c.igni.book.
IP
(↑ GF)=↓NP
giam
I’
Icoord
↓∈↑I
iqida
CONJ
da
↓∈↑I
c. aik. itxa
S
(↑ GF)=↓NP
es c. igni
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 42 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
Simple case: both verbs license the same cases
giamGia
iqidabought
daand
c.aik. itxaread
esthis
c.igni.book.
IP
(↑ GF)=↓NP
giam
I’
Icoord
↓∈↑I
iqida
CONJ
da
↓∈↑I
c. aik. itxa
S
(↑ GF)=↓NP
es c. igni
2
6
6
6
6
6
6
6
6
6
6
6
6
6
4
8
>
>
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
>
>
:
2
6
6
6
6
4
PRED ‘qidva<[2:gia],[3:c. igni]>’
SUBJ2
h
PRED ‘gia’, CASE ergi
OBJ3
h
PRED ‘c. igni’, CASE nomi
3
7
7
7
7
5
,
2
6
4
PRED ‘c.a-k. itxva<[2:gia],[3:c. igni]>’SUBJ [2]OBJ [3]
3
7
5
9
>
>
>
>
>
>
>
>
>
>
>
>
=
>
>
>
>
>
>
>
>
>
>
>
>
;
3
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 42 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
General case: the verbs license different cases
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 43 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
General case: the verbs license different cases
IP
(↑ LEFT GF)=↓PRONP
mas
I’
Icoord
↓∈↑(↑ LEFT)=↓
I
uqvars
CONJ
da
↓∈↑(↑ RIGHT)=↓
I
apasebs
S
(↑ RIGHT GF)=↓NP
kartvelebs
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 43 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
General case: the verbs license different cases
IP
(↑ LEFT GF)=↓PRONP
mas
I’
Icoord
↓∈↑(↑ LEFT)=↓
I
uqvars
CONJ
da
↓∈↑(↑ RIGHT)=↓
I
apasebs
S
(↑ RIGHT GF)=↓NP
kartvelebs
2
6
6
6
6
6
6
6
6
4
8
>
>
>
>
>
<
>
>
>
>
>
: 1
2
6
6
6
6
4
PRED ‘qvareba<[2:is],[3:kartveli]>’
SUBJ2
h
PRED ‘is’, CASE dati
OBJ3
h
PRED ‘kartveli’, CASE nomi
3
7
7
7
7
5
,
4
2
6
6
6
6
4
PRED ‘še-paseba<[5:is],[6:kartveli]>’
SUBJ5
h
PRED ‘is’, CASE nomi
OBJ6
h
PRED ‘kartveli’, CASE dati
3
7
7
7
7
5
9
>
>
>
>
>
=
>
>
>
>
>
;
LEFT [1], RIGHT [4]
3
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 43 / 54
Some aspects of Georgian syntax Predicate coordination
Predicate coordination: Implementation
IP
(↑ LEFT GF)=↓PRONP
mas
I’
Icoord
↓∈↑(↑ LEFT)=↓
I
uqvars
(↑ LEFT SUBJ)\CASE
=(↑ RIGHT SUBJ)\CASE
(↑ LEFT OBJ)\CASE
=(↑ RIGHT OBJ)\CASE
CONJ
da
↓∈↑(↑ RIGHT)=↓
I
apasebs
S
(↑ RIGHT GF)=↓NP
kartvelebs
2
6
6
6
6
6
6
6
6
4
8
>
>
>
>
>
<
>
>
>
>
>
: 1
2
6
6
6
6
4
PRED ‘qvareba<[2:is],[3:kartveli]>’
SUBJ2
h
PRED ‘is’, CASE dati
OBJ3
h
PRED ‘kartveli’, CASE nomi
3
7
7
7
7
5
,
4
2
6
6
6
6
4
PRED ‘šepaseba<[5:is],[6:kartveli]>’
SUBJ5
h
PRED ‘is’, CASE nomi
OBJ6
h
PRED ‘kartveli’, CASE dati
3
7
7
7
7
5
9
>
>
>
>
>
=
>
>
>
>
>
;
LEFT [1], RIGHT [4]
3
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 44 / 54
Some aspects of Georgian syntax Discontinuous constituents
Discontinuous constituents
cem-i[[my.NOM].POSS
gvar-ilast-name.NOM].NP
arnot
v-u-txar-iI.to-him.told.it.
‘I did not tell him my last name.’
gvar-i[last-name.NOM].NP
arnot
v-u-txar-iI.to-him.told.it
cem-i[my.NOM].POSS.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 45 / 54
Some aspects of Georgian syntax Discontinuous constituents
Discontinuous constituents
IP
(↑ OBJ)=↓
POSSP
(↑ SPEC POSS)=↓
POSS
cemi
NP
N
gvari
I’
I
ar vutxari
S
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 46 / 54
Some aspects of Georgian syntax Discontinuous constituents
Discontinuous constituents
IP
(↑ OBJ)=↓
POSSP
(↑ SPEC POSS)=↓
POSS
cemi
NP
N
gvari
I’
I
ar vutxari
S
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘ubnoba<[1:pro],[2:gvari],[3:pro]>’
SUBJ1
"
PRED ‘pro’PERS 1 NUM sg
#
OBJ
2
2
6
6
6
6
6
6
6
4
PRED ‘gvari’CASE nom
SPEC
2
6
6
4
POSS
2
6
4
PRED ‘pro’PERS 1 NUM sgCASE nom
3
7
5
3
7
7
5
3
7
7
7
7
7
7
7
5
OBJben3
"
PRED ‘pro’PERS 3
#
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 46 / 54
Some aspects of Georgian syntax Discontinuous constituents
Discontinuous constituents
IP
(↑ OBJ)=↓
POSSP
(↑ SPEC POSS)=↓
POSS
cemi
NP
N
gvari
I’
I
ar vutxari
S
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
PRED ‘ubnoba<[1:pro],[2:gvari],[3:pro]>’
SUBJ1
"
PRED ‘pro’PERS 1 NUM sg
#
OBJ
2
2
6
6
6
6
6
6
6
4
PRED ‘gvari’CASE nom
SPEC
2
6
6
4
POSS
2
6
4
PRED ‘pro’PERS 1 NUM sgCASE nom
3
7
5
3
7
7
5
3
7
7
7
7
7
7
7
5
OBJben3
"
PRED ‘pro’PERS 3
#
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
IP
(↑ OBJ)=↓
NP
N
gvari
I’
I
ar vutxari
S
(↑ GF SPEC POSS)=↓
POSS
cemi
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 46 / 54
Tools for LFG grammar development
Outline
1 Introduction: The Georgian grammar project
2 Lexical-Functional Grammar
3 Morphology and Morphosyntax
4 Some aspects of Georgian syntax
5 Tools for LFG grammar development
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 47 / 54
Tools for LFG grammar development
XLE and fst
XLE (Xerox Linguistic Environment)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 48 / 54
Tools for LFG grammar development
XLE and fst
XLE (Xerox Linguistic Environment)
is a sophisticated development platform for LFG grammarsdeveloped by the Palo Alto Research Center with activeparticipation of some of the inventors of LFG.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 48 / 54
Tools for LFG grammar development
XLE and fst
XLE (Xerox Linguistic Environment)
is a sophisticated development platform for LFG grammarsdeveloped by the Palo Alto Research Center with activeparticipation of some of the inventors of LFG.
consists of a parser, a generator and a transfer module
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 48 / 54
Tools for LFG grammar development
XLE and fst
XLE (Xerox Linguistic Environment)
is a sophisticated development platform for LFG grammarsdeveloped by the Palo Alto Research Center with activeparticipation of some of the inventors of LFG.
consists of a parser, a generator and a transfer module
can be used both from Emacs via a Tcl/Tk interface that providespowerful viewing and debugging facilities, and as a shared library,which opens up for integrating XLE into custom software
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 48 / 54
Tools for LFG grammar development
XLE and fst
XLE (Xerox Linguistic Environment)
is a sophisticated development platform for LFG grammarsdeveloped by the Palo Alto Research Center with activeparticipation of some of the inventors of LFG.
consists of a parser, a generator and a transfer module
can be used both from Emacs via a Tcl/Tk interface that providespowerful viewing and debugging facilities, and as a shared library,which opens up for integrating XLE into custom software
Tokenization and morphological analysis is normally done with theXerox finite state tool, fst
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 48 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
easy-to-use pedagogical Web interface to XLE for parsingsentences on the fly
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
easy-to-use pedagogical Web interface to XLE for parsingsentences on the fly
in use for several of the ParGram grammars (among othersNorwegian, English, German, Welsh and Malagassy)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
easy-to-use pedagogical Web interface to XLE for parsingsentences on the fly
in use for several of the ParGram grammars (among othersNorwegian, English, German, Welsh and Malagassy)
display of c- and f-structures of LFG analyses
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
easy-to-use pedagogical Web interface to XLE for parsingsentences on the fly
in use for several of the ParGram grammars (among othersNorwegian, English, German, Welsh and Malagassy)
display of c- and f-structures of LFG analyses
visualization of the mapping from c- to f-structure
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development XLE-Web: A Web interface for XLE
XLE-Web
XLE-Web
easy-to-use pedagogical Web interface to XLE for parsingsentences on the fly
in use for several of the ParGram grammars (among othersNorwegian, English, German, Welsh and Malagassy)
display of c- and f-structures of LFG analyses
visualization of the mapping from c- to f-structure
display of compact packed representations of c- and f-structuresthat combine the c- resp. f-structures of all analyses of a givenparse into one c- resp. f-structure graph
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 49 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
Tasks when developing a large grammar:In order to monitor progress, to assess coverage and to compareanalyses across different grammar versions:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 50 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
Tasks when developing a large grammar:In order to monitor progress, to assess coverage and to compareanalyses across different grammar versions:
run the grammar on a set of sample sentences
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 50 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
Tasks when developing a large grammar:In order to monitor progress, to assess coverage and to compareanalyses across different grammar versions:
run the grammar on a set of sample sentences
store the parse results
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 50 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
Tasks when developing a large grammar:In order to monitor progress, to assess coverage and to compareanalyses across different grammar versions:
run the grammar on a set of sample sentences
store the parse results
rerun successive versions of the grammar on the same sentencesand compare the results
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 50 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
develop a treebank in the sense of a linguistic resource
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
develop a treebank in the sense of a linguistic resource
Problems:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
develop a treebank in the sense of a linguistic resource
Problems:
sentences of only moderate complexity often are highlyambiguous
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
develop a treebank in the sense of a linguistic resource
Problems:
sentences of only moderate complexity often are highlyambiguous
the desired or correct reading is only one of the analyses offeredby the grammar
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Grammar development
When the grammar has reached acceptable coverage, one wants to:
run the grammar on a larger set of sentences (perhaps chosenfrom running text)
develop a treebank in the sense of a linguistic resource
Problems:
sentences of only moderate complexity often are highlyambiguous
the desired or correct reading is only one of the analyses offeredby the grammar
⇒ Need for manual disambiguation of the parses in an efficient way
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 51 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Supports a process flow involving
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Supports a process flow involving
automatic parsing with XLE
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Supports a process flow involving
automatic parsing with XLE
viewing with XLE-Web
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Supports a process flow involving
automatic parsing with XLE
viewing with XLE-Web
structural c- and f-structure queries based on the TIGERSearchtreebank search tool
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
LFG Parsebanker
LFG Parsebanker: Web-based toolkit for building and manualdisambiguation of an LFG treebank
developed in the Trepil project (together with Victoria Rosén andKoenraad de Smedt, Bergen)
originally for Norwegian, but language independent
Supports a process flow involving
automatic parsing with XLE
viewing with XLE-Web
structural c- and f-structure queries based on the TIGERSearchtreebank search tool
efficient manual disambiguation by means of discriminants
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 52 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammars
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammarsfour major types: lexical, morphological, c-structure and f-structurediscriminants
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammarsfour major types: lexical, morphological, c-structure and f-structurediscriminantsA lexical discriminant is a word form together with its part ofspeech
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammarsfour major types: lexical, morphological, c-structure and f-structurediscriminantsA lexical discriminant is a word form together with its part ofspeechA morphological discriminant is a base form with the tags itreceives from morphological preprocessing
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammarsfour major types: lexical, morphological, c-structure and f-structurediscriminantsA lexical discriminant is a word form together with its part ofspeechA morphological discriminant is a base form with the tags itreceives from morphological preprocessingC-structure discriminants are based on minimal subtrees, aminimal subtree being defined as a mother node and herdaughters
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development LFG Parsebanker: Grammar development and treebanks
Discriminants
Discriminant: ‘Any elementary linguistic property of an analysis that isnot shared by all analyses’ (David Carter).Our discriminants:
specifically designed for LFG grammarsfour major types: lexical, morphological, c-structure and f-structurediscriminantsA lexical discriminant is a word form together with its part ofspeechA morphological discriminant is a base form with the tags itreceives from morphological preprocessingC-structure discriminants are based on minimal subtrees, aminimal subtree being defined as a mother node and herdaughtersF-structure discriminants are based on partial paths throughf-structuresMeurer (AKSIS, UiB) A Computational Grammar for Georgian 53 / 54
Tools for LFG grammar development A Georgian corpus of non-fictional and fictional texts
A Georgian corpus of non-fictional and fictional texts
An indispensable resource for research in Georgian syntax is asearchable text corpus of decent size.Available text collections on the Internet:
the electronic newspaper archive Opentext (> 75 million words)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 54 / 54
Tools for LFG grammar development A Georgian corpus of non-fictional and fictional texts
A Georgian corpus of non-fictional and fictional texts
An indispensable resource for research in Georgian syntax is asearchable text corpus of decent size.Available text collections on the Internet:
the electronic newspaper archive Opentext (> 75 million words)
the text archive of Radio tavisupleba (the Georgian service ofRadio Free Europe/Radio Liberty) (around eight million words)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 54 / 54
Tools for LFG grammar development A Georgian corpus of non-fictional and fictional texts
A Georgian corpus of non-fictional and fictional texts
An indispensable resource for research in Georgian syntax is asearchable text corpus of decent size.Available text collections on the Internet:
the electronic newspaper archive Opentext (> 75 million words)
the text archive of Radio tavisupleba (the Georgian service ofRadio Free Europe/Radio Liberty) (around eight million words)
fictional texts (both prose and poetry): the UNESCO project Digitalcollection of Georgian classical literature (three million words)
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 54 / 54
Tools for LFG grammar development A Georgian corpus of non-fictional and fictional texts
A Georgian corpus of non-fictional and fictional texts
An indispensable resource for research in Georgian syntax is asearchable text corpus of decent size.Available text collections on the Internet:
the electronic newspaper archive Opentext (> 75 million words)
the text archive of Radio tavisupleba (the Georgian service ofRadio Free Europe/Radio Liberty) (around eight million words)
fictional texts (both prose and poetry): the UNESCO project Digitalcollection of Georgian classical literature (three million words)
I have harvested these text collections and imported them into corpusquery software based on Corpus Workbench (IMS Stuttgart) which isbeing developed at Aksis.
Meurer (AKSIS, UiB) A Computational Grammar for Georgian 54 / 54