syntactic variation in russian adversative conjunctions implications for sentence production...
Post on 20-Dec-2015
234 views
TRANSCRIPT
Syntactic variation in Russian adversative conjunctions
Implications for sentence production
Vsevolod KapatsinskiIndiana University
Linguistics and Cognitive ScienceSpeech Research [email protected]
The problem
• Adversative conjunctions– Da– No– Odnako
• Он ушёл, да скоро вернулся• Он ушёл, но скоро вернулся• Он ушёл, однако скоро вернулся
– ‘He left but soon returned’
X Y
The problem
• No contexts that categorically determine conjunction choice
• The aims:– First corpus-based multivariate study of the
problem– Implications for sentence production in
general
What is adversative? (after Shvedova et al. 1980)
• Y prevents X from running to completion– He would have left but they detained him.– And everything would be fine but the guy
complained.
• Conjoined events are independent but Y contradicts implications of X– He left but soon returned. – They were disliked but no-one looked good on
their background.
The irrealis-realis construction
X {conditional} or {past/present imperfective}+infinitive}
+
{{da} or {no} or {odnako}}
+
Y {perfective}
Examples
– И всё бы обошлось, да один старичок написал сыну в город жалобное письмо.
– ‘And everything would be fine but one old man wrote to his son in the city a complaint.’
– Он хотел уйти, но его задержали.
– ‘He wanted/was going to leave but [they] detained him.’
Conditional Adversative conjunction
Perfective
Past imperfective
Adversative conjunction
Perfective
Previous work
• No consensus on the effect of semantics– Serebrjanaja (1976): da is favored when Y interrupts X – Lekant et al (1982): da is less adversative than no and
odnako – Kruchinina (1988): in X da Y, speaker believes that
the hearer believes X, Y contradicts X, da softens the contradiction
Factors: Discourse
• Topicality of Y– Koolemans Beynen (1976): odnako introduces non-topical
information • Register
– Krilova (1980), Shvedova et al (1980), Kruchinina (1988): da is unproductive, restricted to colloquial usage
– Lekant et al (1982): odnako is restricted to written discourse
– Here: articles vs. interviews• Following unit length relative to median length of
such units• Anaphoric linkage between X and Y
Factors: syntactic
• Constituent type – Types of X and Y:
• A, NP or PP nominal• V(P) or clause verbal• asymmetric
Asymmetric constructions
– When X and Y are of the same type (e.g. A conj A), there is no way to know whether type of X, type of Y or both influence conjunction choice.
– However, there are cases when X and Y are of different types (e.g. NP conj A).
Examples
• Adj conj VP – А обычно какой-нибудь ветер, да дует.– ‘While usually some wind but blows.’
• Adj conj NP – С другой стороны, пока Земля ещё
вертится: слабое, да утешение.– ‘On the other hand, the earth is still spinning –
weak but a reassurance.’
Asymmetric constructions and sentence production
• If NP conj A has o A different effect than A conj A o And a different effect than NP conj NP both X type and Y type matter
information about the types of both conjoined constituents is available at the time when the conjunction is chosen
OR the type of the constituent preceding the conjunction
and the type of concept following the conjunction
Asymmetric constructions and sentence production
• This would imply that either – language production does not proceed in a
strictly serial, constituent-by-constituent manner
or– syntax and semantics are not informationally
encapsulated and can co-determine lexical item choice at a single processing stage.
Data
• 234 tokens of da, 247 tokens of no, and 246 tokens of odnako from the 7,600,000 word Ogonek Corpus (Berger 2003)
• The Ogonek Corpus contains the full text of a Russian magazine by the same name including articles and interviews
Exclusions
• Non-unit-initial conjunctions: – may not be chosen at the same point in the language
production process: e, non-unit-initial uses of odnako
• Conjunction clusters– might be a single choice; – there are too few tokens of each cluster type to enter
them into the analysis– no odnako
• Fixed Expressions
Analysis
• Varbrule– Add factors one by one, see if the larger
model explains more variation than the smaller model, keep factors whose addition increases predictiveness and whose deletion decreases it
– Output: factor weights
Results of multivariate analysis: Length of Y
00.10.20.30.40.50.60.70.80.9
1
Long Short
Y is...
Fa
cto
r w
eig
ht
Da
Odnako
Odnako Da
Factor weight > .5 this factor setting favors this pronoun
Results of multivariate analysis: Register
Odnako
0
0.2
0.4
0.6
0.8
1
Article Interview
Register
Fac
tor
wei
gh
t
NOT odnako
Odnako
0
0.2
0.4
0.6
0.8
1
Yes No
Y contains a referent that's mentioned later?
Fa
cto
r w
eig
ht
Results of multivariate analysis: Topicality of Y
Odnako tends to introduce non-topical referents
Results of multivariate analysis: Anaphoric linkage
00.10.20.30.40.50.60.70.80.9
1
Yes No
Y is anaphorically linked to X?
Fa
cto
r w
eig
ht
Da
No
Odnako
NOT da
Da signals a larger break in coherence
Results of multivariate analysis: Irrealis-Realis
00.10.20.30.40.50.60.70.80.9
1
Yes No
Irrelis-realis?
Fa
cto
r w
eig
ht
Da
No
Odnako
DaDa is semantically distinct from the other adversative conjunctions
Results of multivariate analysis: Constituent type
00.10.20.30.40.50.60.70.80.9
1
Nominal Verbal Asymmetric
Constituents linked
Fa
cto
r w
eig
ht
Da
No
Odnako
No DaOdnako
Summary
Relative factor importance
0
10
20
30
40
50
60
70
Da No Odnako
Ra
ng
e
Topicality of Y
Register
X-Y anaphoric linkage
Length of Y
Irrealis-Realis
Constituent type
Asymmetric Nominal VerbalArticles
Larger range more important factor
Disfavors
Favors
Conclusion• Da is semantically distinct from other adversative
conjunctions and is used to signal that the event denoted by X is prevented from running to completion by the event denoted by Y
• Odnako is the most restricted conjunction. It is influenced by register more than the other conjunctions.
• Conjunction choice is gradient but systematic. Not free variation. Supports stochastic grammar.
• Demonstrates the value of corpus-based multivariate approaches to the description of linguistic phenomena.– Semantic controversy resolved– New factors discovered
• Constituent type – the most important factor!• Anaphoric linkage between X and Y
Processing implications
• Asymmetric constructions behave differently than symmetric constructions
Types of both of the conjoined constituents must be available when the conjunction is chosen
Sentence production does not proceed in strictly serial, constituent-by-constituent manner
Or, syntax and semantics jointly determine conjunction choice at a single processing stage.
ReferencesKoolemans Beynen, G. (1976). Semantic differences between no and odnako. Slavic and
East European Journal, 20 (2), 167-73. Krilova, G. (1980). Ruskijat sojuz da I negovite funkcionalni ekvivalenti v xudozhestveni
prevodi na bolgarski ezik. Sopostavitelno Ezikoznanie, 5 (3), 18-25. Kruchinina, I. N. (1988). Struktura i Funkcii Sochinitel’noj Svjazi v Russkom Jazyke.
Moscow: Nauka. Lekant, P. A. et al. (1982). Sovremennyj Russkij Literaturnyj Jazyk. Moscow: Vysshaja
Shkola. Rand, D., and D. Sankoff. (1990). GoldVarb Version 2: A Variable Rule Application for
Macintosh. On-line Manual. http://www.crm.umontreal.ca/~sankoff/GoldVarbManual.Dir
Serebrjanaja, F. I. (1976). Nekotorye nabljudenija nad upotrebleniem sojuza da. Russkij Jazyk v Shkole, 4.
SFB 441, Project B1. Ogonek 1996-2002. http://heckel.sfb.uni-tuebingen.de/cgi-bin/cqp.pl?sprache=en&trans=lat
Shvedova, N. Ju, N. D. Arutjunova, A. V. Bondarko, V. V. Ivanov, V. V. Lopatin, I. S. Uluxanov, F. Filin, and the Institute for the Russian Language, The Academy of Sciences of the USSR. (1980). Russkaja Grammatika. Vol. 2: Sintaksis. Moscow: Nauka.