The relationship between PROIEL dependency annotation … · The relationship between PROIEL...
Transcript of The relationship between PROIEL dependency annotation … · The relationship between PROIEL...
Background DG and PSG The conversion Some applications
The relationship between PROIEL dependencyannotation and LFG
Dag Haug
University of Oslo
January 8
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 1 / 29
Background DG and PSG The conversion Some applications
Dependency Grammar
Almost a standard in computational linguistics
computationally simplegood parsers available
A natural choice for linguistic databases
fairly close to traditional school grammar → easy to annotaterelatively ‘theory-neutral’taking traditional grammatical relations as primitives
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 2 / 29
Background DG and PSG The conversion Some applications
Dependency Grammar
Almost a standard in computational linguistics
computationally simplegood parsers available
A natural choice for linguistic databases
fairly close to traditional school grammar → easy to annotaterelatively ‘theory-neutral’taking traditional grammatical relations as primitives
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 2 / 29
Background DG and PSG The conversion Some applications
From linguistic database to theory
On the downside, few people use DG as a linguistic theory
There is no particular theory of word order
In particular, long distance dependencies are intractable
So while DG is a reasonable format to create and store linguistic datain, we might want to (implicitly or explicitly) convert the data for ourstudies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 3 / 29
Background DG and PSG The conversion Some applications
From linguistic database to theory
On the downside, few people use DG as a linguistic theory
There is no particular theory of word order
In particular, long distance dependencies are intractable
So while DG is a reasonable format to create and store linguistic datain, we might want to (implicitly or explicitly) convert the data for ourstudies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 3 / 29
Background DG and PSG The conversion Some applications
From linguistic database to theory
On the downside, few people use DG as a linguistic theory
There is no particular theory of word order
In particular, long distance dependencies are intractable
So while DG is a reasonable format to create and store linguistic datain, we might want to (implicitly or explicitly) convert the data for ourstudies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 3 / 29
Background DG and PSG The conversion Some applications
Lexical-Functional Grammar
Parallel architecture
Syntax is represented at two-levels: feature structure and constituentstructure
Much syntactic information (such as subcategorization frames, andeverything morphological) is lexicalised
But c-structure can also contribute information to f-structure, notablyin rigid word order languages
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 4 / 29
Background DG and PSG The conversion Some applications
Lexical-Functional Grammar
Parallel architecture
Syntax is represented at two-levels: feature structure and constituentstructure
Much syntactic information (such as subcategorization frames, andeverything morphological) is lexicalised
But c-structure can also contribute information to f-structure, notablyin rigid word order languages
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 4 / 29
Background DG and PSG The conversion Some applications
Lexical-Functional Grammar
Parallel architecture
Syntax is represented at two-levels: feature structure and constituentstructure
Much syntactic information (such as subcategorization frames, andeverything morphological) is lexicalised
But c-structure can also contribute information to f-structure, notablyin rigid word order languages
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 4 / 29
Background DG and PSG The conversion Some applications
Lexical-Functional Grammar
Parallel architecture
Syntax is represented at two-levels: feature structure and constituentstructure
Much syntactic information (such as subcategorization frames, andeverything morphological) is lexicalised
But c-structure can also contribute information to f-structure, notablyin rigid word order languages
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 4 / 29
Background DG and PSG The conversion Some applications
Lexical-Functional Grammar
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 5 / 29
Background DG and PSG The conversion Some applications
What we are trying to do
From the dependency annotation in the PROIEL corpus, we want togenerate LFG representation, f-structures as well as c-structures
F-structure conversion is actually quite simple, given the similaritiesbetween DG and f-structures
Grammatical relations as primitivesNo representation of word orderPROIEL DG is already ‘fixed up’ to be closer to f-structures
C-structure conversion is more complex, but more interesting forother phrase structure based approaches
Also, having a more principled representation of word order than purelinearization is interesting for word order studies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 6 / 29
Background DG and PSG The conversion Some applications
What we are trying to do
From the dependency annotation in the PROIEL corpus, we want togenerate LFG representation, f-structures as well as c-structures
F-structure conversion is actually quite simple, given the similaritiesbetween DG and f-structures
Grammatical relations as primitivesNo representation of word orderPROIEL DG is already ‘fixed up’ to be closer to f-structures
C-structure conversion is more complex, but more interesting forother phrase structure based approaches
Also, having a more principled representation of word order than purelinearization is interesting for word order studies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 6 / 29
Background DG and PSG The conversion Some applications
What we are trying to do
From the dependency annotation in the PROIEL corpus, we want togenerate LFG representation, f-structures as well as c-structures
F-structure conversion is actually quite simple, given the similaritiesbetween DG and f-structures
Grammatical relations as primitivesNo representation of word orderPROIEL DG is already ‘fixed up’ to be closer to f-structures
C-structure conversion is more complex, but more interesting forother phrase structure based approaches
Also, having a more principled representation of word order than purelinearization is interesting for word order studies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 6 / 29
Background DG and PSG The conversion Some applications
What we are trying to do
From the dependency annotation in the PROIEL corpus, we want togenerate LFG representation, f-structures as well as c-structures
F-structure conversion is actually quite simple, given the similaritiesbetween DG and f-structures
Grammatical relations as primitivesNo representation of word orderPROIEL DG is already ‘fixed up’ to be closer to f-structures
C-structure conversion is more complex, but more interesting forother phrase structure based approaches
Also, having a more principled representation of word order than purelinearization is interesting for word order studies
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 6 / 29
Background DG and PSG The conversion Some applications
What do DGs represent
At the most basic level, DGs represent assymetric (head-dependent)relations between words
Typically these dependencies are labeled with grammatical relations
The analysis shows which words ‘belong together’, whether theystand together or not
The notion of syntactic category is not explicitly represented, thoughit can possibly be retrieved from morphological annotation
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 7 / 29
Background DG and PSG The conversion Some applications
What do DGs represent
At the most basic level, DGs represent assymetric (head-dependent)relations between words
Typically these dependencies are labeled with grammatical relations
The analysis shows which words ‘belong together’, whether theystand together or not
The notion of syntactic category is not explicitly represented, thoughit can possibly be retrieved from morphological annotation
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 7 / 29
Background DG and PSG The conversion Some applications
What do DGs represent
At the most basic level, DGs represent assymetric (head-dependent)relations between words
Typically these dependencies are labeled with grammatical relations
The analysis shows which words ‘belong together’, whether theystand together or not
The notion of syntactic category is not explicitly represented, thoughit can possibly be retrieved from morphological annotation
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 7 / 29
Background DG and PSG The conversion Some applications
What do DGs represent
At the most basic level, DGs represent assymetric (head-dependent)relations between words
Typically these dependencies are labeled with grammatical relations
The analysis shows which words ‘belong together’, whether theystand together or not
The notion of syntactic category is not explicitly represented, thoughit can possibly be retrieved from morphological annotation
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 7 / 29
Background DG and PSG The conversion Some applications
What do PSGs represent
Immediate constituents, i.e.words that stand together andbelong together
Constituents are typicallylabelled with their grammaticalcategoryOften, they are enriched (e.g.X-bar theory) with a notion ofheadand a distinction betweencomplementation andadjunction
the girl killed Max
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 8 / 29
Background DG and PSG The conversion Some applications
What do PSGs represent
Immediate constituents, i.e.words that stand together andbelong togetherConstituents are typicallylabelled with their grammaticalcategory
Often, they are enriched (e.g.X-bar theory) with a notion ofheadand a distinction betweencomplementation andadjunction
S
Dthe
Ngirl
Vkilled
NPMax
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 8 / 29
Background DG and PSG The conversion Some applications
What do PSGs represent
Immediate constituents, i.e.words that stand together andbelong togetherConstituents are typicallylabelled with their grammaticalcategoryOften, they are enriched (e.g.X-bar theory) with a notion ofhead
and a distinction betweencomplementation andadjunction
S
NP
DP
the
N
girl
VP
V
killed
NP
N
Max
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 8 / 29
Background DG and PSG The conversion Some applications
What do PSGs represent
Immediate constituents, i.e.words that stand together andbelong togetherConstituents are typicallylabelled with their grammaticalcategoryOften, they are enriched (e.g.X-bar theory) with a notion ofheadand a distinction betweencomplementation andadjunction
S
NP
DP
the
N
girl
VP
VP
V
killed
NP
N
Max
AdvP
brutally
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 8 / 29
Background DG and PSG The conversion Some applications
The choices to make
1 What are the categories?
2 What kind of projections do a category X have?
3 If a category Y depends on a category X in a dependency structure,how far should Y project before it attaches to Xs projection?
4 If a category Y depends on a category X in a dependency structure,to what position in Xs projection should Ys projection attach?
Our answers will be guided by LFG’s particular theory of c-structure,but the result should be useful for other theoretical approaches as well
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 9 / 29
Background DG and PSG The conversion Some applications
The choices to make
1 What are the categories?
2 What kind of projections do a category X have?
3 If a category Y depends on a category X in a dependency structure,how far should Y project before it attaches to Xs projection?
4 If a category Y depends on a category X in a dependency structure,to what position in Xs projection should Ys projection attach?
Our answers will be guided by LFG’s particular theory of c-structure,but the result should be useful for other theoretical approaches as well
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 9 / 29
Background DG and PSG The conversion Some applications
The choices to make
1 What are the categories?
2 What kind of projections do a category X have?
3 If a category Y depends on a category X in a dependency structure,how far should Y project before it attaches to Xs projection?
4 If a category Y depends on a category X in a dependency structure,to what position in Xs projection should Ys projection attach?
Our answers will be guided by LFG’s particular theory of c-structure,but the result should be useful for other theoretical approaches as well
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 9 / 29
Background DG and PSG The conversion Some applications
The choices to make
1 What are the categories?
2 What kind of projections do a category X have?
3 If a category Y depends on a category X in a dependency structure,how far should Y project before it attaches to Xs projection?
4 If a category Y depends on a category X in a dependency structure,to what position in Xs projection should Ys projection attach?
Our answers will be guided by LFG’s particular theory of c-structure,but the result should be useful for other theoretical approaches as well
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 9 / 29
Background DG and PSG The conversion Some applications
The choices to make
1 What are the categories?
2 What kind of projections do a category X have?
3 If a category Y depends on a category X in a dependency structure,how far should Y project before it attaches to Xs projection?
4 If a category Y depends on a category X in a dependency structure,to what position in Xs projection should Ys projection attach?
Our answers will be guided by LFG’s particular theory of c-structure,but the result should be useful for other theoretical approaches as well
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 9 / 29
Background DG and PSG The conversion Some applications
Categories
Category POSAdj adjectives, attributive participles, numerals, pronounsAdv adverbsC subjunctionsConj (superfluous) conjunctionsD articleI finite verbsInt interjectionsN nouns + nominalized adjectives, numerals and pronounsP prepositionsVinf infinitive verbsVptcp participle verbs
Category inferences are configurable
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 10 / 29
Background DG and PSG The conversion Some applications
Categories
Category POSAdj adjectives, attributive participles, numerals, pronounsAdv adverbsC subjunctionsConj (superfluous) conjunctionsD articleI finite verbsInt interjectionsN nouns + nominalized adjectives, numerals and pronounsP prepositionsVinf infinitive verbsVptcp participle verbs
Category inferences are configurable
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 10 / 29
Background DG and PSG The conversion Some applications
What is a nominalization?
Pronouns, numerals and adjectives are ambiguous between N and Adjand are counted as N when
they have a nominal relation (subject, object, oblique, adverbial,apposition, nominal argument or partitive)
they are reflexive, personal and reciprocal pronouns, as these do nothave a determinative usethey are attributes, but genitive and the head is not genitive or has adifferent number
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 11 / 29
Background DG and PSG The conversion Some applications
What is a nominalization?
Pronouns, numerals and adjectives are ambiguous between N and Adjand are counted as N when
they have a nominal relation (subject, object, oblique, adverbial,apposition, nominal argument or partitive)they are reflexive, personal and reciprocal pronouns, as these do nothave a determinative use
they are attributes, but genitive and the head is not genitive or has adifferent number
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 11 / 29
Background DG and PSG The conversion Some applications
What is a nominalization?
Pronouns, numerals and adjectives are ambiguous between N and Adjand are counted as N when
they have a nominal relation (subject, object, oblique, adverbial,apposition, nominal argument or partitive)they are reflexive, personal and reciprocal pronouns, as these do nothave a determinative usethey are attributes, but genitive and the head is not genitive or has adifferent number
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 11 / 29
Background DG and PSG The conversion Some applications
Some theoretical issues
Notice that in LFG, even functional projections should have lexicalheads
So CP and IP are projections of complementizers and finite verbs, notof V, though they belong to the functional domain of VP
Clauses with only a finite verb will only have IP, no VP
The status of DP in Greek is unclear; for simplicity I just assume DPin spec, NP.
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 12 / 29
Background DG and PSG The conversion Some applications
Some theoretical issues
Notice that in LFG, even functional projections should have lexicalheads
So CP and IP are projections of complementizers and finite verbs, notof V, though they belong to the functional domain of VP
Clauses with only a finite verb will only have IP, no VP
The status of DP in Greek is unclear; for simplicity I just assume DPin spec, NP.
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 12 / 29
Background DG and PSG The conversion Some applications
Some theoretical issues
Notice that in LFG, even functional projections should have lexicalheads
So CP and IP are projections of complementizers and finite verbs, notof V, though they belong to the functional domain of VP
Clauses with only a finite verb will only have IP, no VP
The status of DP in Greek is unclear; for simplicity I just assume DPin spec, NP.
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 12 / 29
Background DG and PSG The conversion Some applications
Some theoretical issues
Notice that in LFG, even functional projections should have lexicalheads
So CP and IP are projections of complementizers and finite verbs, notof V, though they belong to the functional domain of VP
Clauses with only a finite verb will only have IP, no VP
The status of DP in Greek is unclear; for simplicity I just assume DPin spec, NP.
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 12 / 29
Background DG and PSG The conversion Some applications
X-bar theory
LFG assumes a version of X-bar theory where
all categories X project two levels X’ and XPfunctional projections have specifiers, to which they assign discoursefunctionsadjunction can occur to all levels
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 13 / 29
Background DG and PSG The conversion Some applications
Phrase structure rules underlying the conversionI and C have specifier positions, and so does N (but only for DPs)
Adjunction is either to X’ or to XP (appositions)
Coordination is always phrasal
A sample rule
N::phrase adjuncts:
- NP- AdjP
:specifier:- DP
:bar adjuncts:- AdjP- NP
:complements:- NP
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 14 / 29
Background DG and PSG The conversion Some applications
Phrase structure rules underlying the conversionI and C have specifier positions, and so does N (but only for DPs)
Adjunction is either to X’ or to XP (appositions)
Coordination is always phrasal
A sample rule
N::phrase adjuncts:
- NP- AdjP
:specifier:- DP
:bar adjuncts:- AdjP- NP
:complements:- NP
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 14 / 29
Background DG and PSG The conversion Some applications
Adjunction site
The conversion allows adjunction both to X’ and to XP - but how dowe know?
The current approach uses a mixture of category and dependencylabel
Only apos-dependents are allowed as phrasal adjunctions, if admittedby the category filter
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 15 / 29
Background DG and PSG The conversion Some applications
The algorithm
1 Fix up the dependency graph so as to have a more intuitive input fora PS analysis
2 Assign a category to each node
3 Find those of the node’s dependents that should go into its phrasalprojection and add traces for the others
4 If the node projects a clause, pick up loose constituents that have notreceived head yet
5 Assign positions (phrase adjunction, specifier, bar adjunction,complement) to the nodes in the phrasal projection
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 16 / 29
Background DG and PSG The conversion Some applications
The algorithm
1 Fix up the dependency graph so as to have a more intuitive input fora PS analysis
2 Assign a category to each node
3 Find those of the node’s dependents that should go into its phrasalprojection and add traces for the others
4 If the node projects a clause, pick up loose constituents that have notreceived head yet
5 Assign positions (phrase adjunction, specifier, bar adjunction,complement) to the nodes in the phrasal projection
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 16 / 29
Background DG and PSG The conversion Some applications
The algorithm
1 Fix up the dependency graph so as to have a more intuitive input fora PS analysis
2 Assign a category to each node
3 Find those of the node’s dependents that should go into its phrasalprojection and add traces for the others
4 If the node projects a clause, pick up loose constituents that have notreceived head yet
5 Assign positions (phrase adjunction, specifier, bar adjunction,complement) to the nodes in the phrasal projection
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 16 / 29
Background DG and PSG The conversion Some applications
The algorithm
1 Fix up the dependency graph so as to have a more intuitive input fora PS analysis
2 Assign a category to each node
3 Find those of the node’s dependents that should go into its phrasalprojection and add traces for the others
4 If the node projects a clause, pick up loose constituents that have notreceived head yet
5 Assign positions (phrase adjunction, specifier, bar adjunction,complement) to the nodes in the phrasal projection
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 16 / 29
Background DG and PSG The conversion Some applications
The algorithm
1 Fix up the dependency graph so as to have a more intuitive input fora PS analysis
2 Assign a category to each node
3 Find those of the node’s dependents that should go into its phrasalprojection and add traces for the others
4 If the node projects a clause, pick up loose constituents that have notreceived head yet
5 Assign positions (phrase adjunction, specifier, bar adjunction,complement) to the nodes in the phrasal projection
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 16 / 29
Background DG and PSG The conversion Some applications
Fixing up the dependency graph
DG analyses often takes the lexical word as the head, where a PSanalysis would take the functional word as the head
Shared elements in coordinations are hung on the conjunction in theDG-analysis, but on a conjunct in the PS
Otherwise we’d need X’ conjunction (possibly future work)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 17 / 29
Background DG and PSG The conversion Some applications
Fixing up the dependency graph
DG analyses often takes the lexical word as the head, where a PSanalysis would take the functional word as the head
Shared elements in coordinations are hung on the conjunction in theDG-analysis, but on a conjunct in the PS
Otherwise we’d need X’ conjunction (possibly future work)
subpred
[pred ‘men‘
],[
pred ‘women‘]
adj
{[pred ‘good’
]}
NP
NP
AdjP
Adj
good
NP
N
men
Conj
and
NP
N
women
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 17 / 29
Background DG and PSG The conversion Some applications
Fixing up the dependency graphDG analyses often takes the lexical word as the head, where a PSanalysis would take the functional word as the headShared elements in coordinations are hung on the conjunction in theDG-analysis, but on a conjunct in the PSOtherwise we’d need X’ conjunction (possibly future work)
subpred
[pred ‘men‘
],[
pred ‘women‘]
adj
{[pred ‘good’
]}
N’
AdjP
Adj
good
N’
N’
N
men
Conj
and
N’
N
women
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 17 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Every node in the DG heads a phrase in the PS
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
To find the phrases which attaches to it we look at its dependents
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
We map nodes to the continuous domains in their dependency graphs
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
For uion (7), this is [6,7,8,9]
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
For einai (5), it is [1] and [5,6,7,8,9]
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
For each such domain, we can find the words within it whosedependency graphs ‘cover’ the domain - 1 covers [1] and 7 covers[6,7,8,9]
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Those which are in the same domain as their head will be the head ofcomplements - so (7) heads a complement of (5)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Those which are not in the same domain leave traces - so (5) governsthe trace of (1)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
We repeat the procedure at the next level
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
The dependents of (2) map on to the domains [1], [3,4] and[5,6,7,8,9], which we can cover with the nodes (1), (4) and (5)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
All of these are in the domain of (2), which is the whole sentence, sothey are phrasal dependents of (2)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
However, (1) is not a direct dependent of (2) - so we identify it withthe trace left under (5) and resolve the trace
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
We only allow such resolution under nodes of category I and V
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
(1), (4) and (5) head projections within the projection of (2) - butwhere?
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
First we look for possible phrasal adjunctions (appositions) - nonehere
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Next, we look for whether the leftmost constituent can be a specifier,which it can here, given our phrase structure rules
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Third, we look for bar-level adjunctions, of which there are none here
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
Finding the phrases
Finally, we take the rest of the projections as complements under (2)
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 18 / 29
Background DG and PSG The conversion Some applications
The result
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 19 / 29
Background DG and PSG The conversion Some applications
What gets topicalized?
category frequencyNP 2779VptcpP 975PP 527NP[rel] 488AdvP 487CP 297NP[int] 271
category frequencyAdvP[rel] 250IP[rel] 175AdjP 171AdvP[int] 135IP 19IP[int] 5AdjP[int] 4
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 20 / 29
Background DG and PSG The conversion Some applications
Types of fronted PPs
category frequency
CAUSAL 111TIME 119LOCATION 30DIRECTION 28MANNER 18SOURCE 16PURPOSE 10RECIPIENT 5TOPIC 4BENEFICIARY 2
untagged 174
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 21 / 29
Background DG and PSG The conversion Some applications
Topicalizations in CP
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 22 / 29
Background DG and PSG The conversion Some applications
Projectivity
Projectivity is the property of being a continous phrase (no traces)
We can distinguish two types of non-projectivity:
The displaced element occurs in a left-peripheral position(wh-movement, topicalization, relativization)The displaced element occurs ‘somewhere else’ (scrambling)
The degree of non-projectivity varies greatly across categories
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 23 / 29
Background DG and PSG The conversion Some applications
Projectivity
Projectivity is the property of being a continous phrase (no traces)
We can distinguish two types of non-projectivity:
The displaced element occurs in a left-peripheral position(wh-movement, topicalization, relativization)The displaced element occurs ‘somewhere else’ (scrambling)
The degree of non-projectivity varies greatly across categories
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 23 / 29
Background DG and PSG The conversion Some applications
Projectivity
Projectivity is the property of being a continous phrase (no traces)
We can distinguish two types of non-projectivity:
The displaced element occurs in a left-peripheral position(wh-movement, topicalization, relativization)The displaced element occurs ‘somewhere else’ (scrambling)
The degree of non-projectivity varies greatly across categories
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 23 / 29
Background DG and PSG The conversion Some applications
Projectivity across categories
Category Wh-movement Scrambling ProjectiveCP 3 0 1351IP 28 9 10260PP 2 11 4318
VinfP 62 34 676VptcpP 9 59 2753
AdjP 3 41 331AdvP 1 12 211NP 41 212 4706
CP, IP and PP do not allow scrambling
AdvP, NP and AdjP do allow scrambling
VinfP and VptcpP behave like the AdvP, AdjP and NP
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 24 / 29
Background DG and PSG The conversion Some applications
Internal subject?
If we assume that the subject can be internal to the participle phrase,there are only 12 non-projective participles phrases (6 cases ofscrambling and 6 of wh-movement) against 1625 projective ones
All of these look suspicious on closer inspection
Should we assume backward control from adjuncts instead?
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 25 / 29
Background DG and PSG The conversion Some applications
Internal subject?
If we assume that the subject can be internal to the participle phrase,there are only 12 non-projective participles phrases (6 cases ofscrambling and 6 of wh-movement) against 1625 projective ones
All of these look suspicious on closer inspection
Should we assume backward control from adjuncts instead?
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 25 / 29
Background DG and PSG The conversion Some applications
Internal subject?
If we assume that the subject can be internal to the participle phrase,there are only 12 non-projective participles phrases (6 cases ofscrambling and 6 of wh-movement) against 1625 projective ones
All of these look suspicious on closer inspection
Should we assume backward control from adjuncts instead?
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 25 / 29
Background DG and PSG The conversion Some applications
The traditional picture
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 26 / 29
Background DG and PSG The conversion Some applications
A backward control analysis
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 27 / 29
Background DG and PSG The conversion Some applications
Conclusions
Dependency grammar is a interesting formalism for ‘storing’ linguisticdata
But it can and should be converted to ‘real’ linguistic theory ofvarious sorts
We have seen an algorithm to generate phrase structure fromdependencies
The phrase structure view allows more applications from boringstatistics via interesting statistics to the challenging received wisomeon Greek syntax
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 28 / 29
Background DG and PSG The conversion Some applications
Conclusions
Dependency grammar is a interesting formalism for ‘storing’ linguisticdata
But it can and should be converted to ‘real’ linguistic theory ofvarious sorts
We have seen an algorithm to generate phrase structure fromdependencies
The phrase structure view allows more applications from boringstatistics via interesting statistics to the challenging received wisomeon Greek syntax
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 28 / 29
Background DG and PSG The conversion Some applications
Conclusions
Dependency grammar is a interesting formalism for ‘storing’ linguisticdata
But it can and should be converted to ‘real’ linguistic theory ofvarious sorts
We have seen an algorithm to generate phrase structure fromdependencies
The phrase structure view allows more applications from boringstatistics via interesting statistics to the challenging received wisomeon Greek syntax
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 28 / 29
Background DG and PSG The conversion Some applications
Conclusions
Dependency grammar is a interesting formalism for ‘storing’ linguisticdata
But it can and should be converted to ‘real’ linguistic theory ofvarious sorts
We have seen an algorithm to generate phrase structure fromdependencies
The phrase structure view allows more applications from boringstatistics via interesting statistics to the challenging received wisomeon Greek syntax
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 28 / 29
Background DG and PSG The conversion Some applications
Slides available athttp://www.hf.uio.no/ifikk/proiel
Data from the PROIEL corpushttp://foni.uio.no:3000
Dag Haug The relationship between PROIEL dependency annotation and LFGJanuary 8 29 / 29