Structures and strings by Virginia Savova A dissertation submitted to The Johns Hopkins University in conformity with the requirements for the degree of Doctor of Philosophy. Baltimore, Maryland October, 2006 © Virginia Savova 2006 All rights reserved UMI Number: 3240793 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted.
Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3240793 Copyright 2007 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company 300 North Zeeb Road P. Box 1346 Ann Arbor, MI 48106-1346 Abstract A central question for linguistics is how strings and structures relate to one another. The majority of frameworks that exist today assume that the hidden structures con- tain complete information about word order. Hence, the theory of grammar is only a theory of the structure-generating component.
However, it is possible to design a the- ory of grammar which views the structure-generating component as separate from the linearization component. As a result, structural descriptions can directly incorporate multi-dominance, and consequently eliminate the need for transformational devices like movement. Another benefit is that linearization can take into account any combination of structural, morpho-phonological and discourse features, paving the way for a word or- der typology in the style of Optimality Theory. This dissertation presents an analysis of basic word order typology in this formalism, and suggests that cross-linguistic frequency of word orders can be explained if constraints are viewed as priors for Bayesian iterative learning.
Finally, it discusses the role of information structure and morpho-phonology for language-internal word order. Author: Virginia Savova Adviser: Dr. Robert Frank, Professor of Cognitive Science, Johns Hopkins University ii Acknowledgements The following acknowledgments are due: To my grandmother, for teaching me the English language, and playing silly card games to motivate me. To my mother, for teaching me the intricacies of linguistic expression.
To my father, for teaching me about the human psyche. To my husband, for introducing me to AI and machine learning. To my Greek and Latin teachers in highschool, for letting me know how syntax really works. To my professor at Harvard, who insisted that I should study linguistics.
To my professor at Harvard, who taught me that “what doesn’t happen is interesting”. To my professor at MIT, who started it all. To my professor at Johns Hopkins, who made it all fit together. To my adviser, for letting me find my own way, while gently pushing in the right direc- tion.
You taught me well: Vesela Kotchemidova, Christina Kotchemidova, Zdravko Savov, Leon Peshkin, Ogi Radev, Lili lordanova, Calvert Watkins, Sam Epstein, Noam Chom- sky, Paul Smolensky, Robert Frank. Thank you! iil Contents Abstract ii Acknowledgements iii List of Tables vii List of Figures I Introduction 0.2 Organization of the dissertation. eee ee ee H Word Order in Syntactic Theory 1 Conflicts of structure and order 10 1. và kg kg k KV 10 12 Mulidomimance.
Qua kg ng 1. Transformations and the Linear Correspondence Axiom 26 2 Complexity and Universality 34 iv III Generation and Linearization 39 3 Separating structure from order 40 3.1 The hierarchical component .2 Overview of dependency grammars .3 Dependency grammar with multdominance.4 The linearization component.5 The parsing component. ee 60 IV Cross-linguistic word order variation 63 4 Basic Word Order 64 4.1 What is basic word order? 2 ww 2 .2 Dependency structure ofdeclaratves.3 Constraints on linearization of discourse-neutral declaratives .4 Order of simple-tense clauses (S, V,O).5 Order of complex-tense clauses (S,O, Aux,V).6 Order and main/subordinate clause status .7 Order correlations across constructions.0004 92 5 Cross-linguistic frequency of word order types 5.1 The role of frequency facts in linguistic theory.2 The number-of-rankings hypothesis .3 The learning bias hypothesis .4 Alternative hypothesis spaces. ee V_ Word order variation within a language 152 6 Discourse and morpho-phonological effects on linearization 153 6.1 Formation of yes-no interrogatives.
0002 eee eee eee 159 6.3 The linearization ofclitics .1 The position of the clitic interrogative marker in Bulgarian. 174 7 Relating the current proposal to existing OT syntax literature. 180 8 Conclusion 190 Bibliography 194 Vita 207 vi List of Tables 3.1 Apparent interaction of word order and discourse-motivated linear prece- denceinGerman .1 Frequency of base word orders [Dryer, 2005a] .2 Basic word order typology .3 Order in complex tense clauses.4 SAuxOV is harmonically bounded by other forms .5 Complex-tense main clauses .7 SVO/SAuxVO ranking .8 VSO/AuxVSO ranking .9 VOS/AuxVOS ranking 2. Note: Wan has no overt complementizer .16 Predicted language 1: SVO/SAuxVO and VSO/AuxVSO .17 Predicted language 2:SVO/SAuxOV in the main and VSO/AuxVSO in the subordinate clause.18 Predicted language 3: VOS/AuxVOS in the subordinate, SVO/SAuxOV in the main clause 4.19 Predicted language 4 (Miya: VOS/AuxVOS in the subordinate, SVO/SAuxVO inthemainclause.
eee ee eee 110 4.20 Typology predicted by the same analysis without subject multi-headedness 111 5.1 Number of winner rankings for each word order predicted in Chapter4.2 Frequency of base word orders [Dryer,2005a]. Number of winner rankings for each word order predicted in Chapter 4, grouping by simple main order .4 Number of winner rankings for each word order predicted in Chapter 4, grouping by simple subordinateorder.5 Number of winner rankings for each word order predicted in Chapter 4, grouping by complex main order.6 Number of winner rankings for each word order predicted in Chapter 4, grouping by complex subordinate order .7 Basic word ordertypology .8 Logistic regression fttedparameters.9 Hypothesis space withp= .10 Hierarchical hypothesis space.1 Order in complex-tense, main, yes-no questions with no overt comple- mentizer(q-particle) .2 Order in complex-tense yes-no questions with complementizer (q-particle)156 6.3 Tensed verb Inversioninquesions.4 Order in complex-tense object wh-questions: QEdge inactive .5 Order in complex-tense object wh-questions: QEdge active .6 Predicted word orders for complex-tense, main, object wh-questions in SVO languages .7 Order in complex-tense object wh-questions: QEdge active .8 The position of interrogative clitic in Bulgarian yes-no questions: focus- free 2.9 The position of interrogative clitic in Bulgarian yes-no questions: focus- free with clitic auxiliary.10 The position of interrogative clitic in Bulgarian yes-no questions: object 6.11 71 Apparent interaction of word order and discourse-motivated linear prece- dence inGerman .2 A modular account of interactioninGerman. 73 A modular account of interaction in English subordinate clauses 7.4 A modular account of interaction in English subordinates 1X List of Figures 0.1 Wh-questions involve multidominance.2 Control as mulidominance.3 Minimalist structure of example (1). ch So 6 11 Chomsky’s early view of splitconstituency .2 VSO word order in OJedasaccount.3 Sampson’s solution to the Bach-Peters paradox .4 Multidominance presents problems for linearization.5 Object-raising according to Anderson .6 Chomsky’s view of displacement.7 The Head Parameter.8 SVO versus VSO structure: the flat structureapproach .9 SVO versus VSO structure: the movement approach.10 SOV: the movement approach .1 SOV versus SVO: the movement approach.1 Simple dependency structure.2 Dependency structure accordingtoKreps.4 Structured set of category labels .5 Interpreting dependency relations.9 Competitor dependency structure(CDS]I).10 Competitor dependency structure(CDS2).
61 41 Subcategorization dependencies of the main declarative constituents .2 Subcategorization and morphological dependencies of the main declar- ative constituents 2. 0à k g g kvna 68 43 Constituent structure corresponding to Flgure42.4 Constituent structure of simple tense clauses.5 Constituent structure of simple tense clauses - head movement and affix lowering.6 Simple-tense declaraiVe.7 Structure violating LHM and implicitely disallowed by the current for- malism. Q Q Q Q Q nu kg và kg kia 71 4.8 Abstract dependency structure 2.10 Lack of head-movement to CinEnglish. 88 411 Structure of sentences with or without overt complementizers.1 Kirby’s iterative learning scheme.2 Griffiths and Kalish’s Model: Graphical model of iterated Bayesian learn- 5.3 Evolution of the posterior over 200 iterations: HIGH and LOW initial distibution .4 Evolution of the posterior over 200 iterations: uniform initial distribution 128 XI 3.5 Distribution of word orders on the 150th iteration (LOW skewed distri- bution).6 Actual current word order distribution .7 Percent trials consistent with current word order distribution plotted against iteration (100 trials per iteration, uniformly initialized) .8 Hasse diagram of pure SOV word order .9 Hasse diagram of pure SVO wordorder.10 Hasse diagram of pure VSO word order.11 Hasse diagram of pure VOS wordorder.12 Hasse diagramof Welsh .13 Hasse điagram ofBreton.15 Hasse diagram of Gaelic 2.16 Hasse diagram of Miya.17 Hasse diagram ofDinka.18 Hasse diagram of Predicted languagel.19 Hasse diagram of Predicted language2.20 Hasse diagram of Predicted language3.1 Underlying representation of yes-no interrogatives.2 Possble underlying representations of object wh-questions .3 Wh-movement account of object wh-questons.1 Possible structures for wurde in Geriman.2 Possible structures for stative verbsinEnglish.
187 XI Part I Introduction 0.1 Overview Linguistic expressions are like icebergs: the string of words we hear is the small observable part above water, while the crucial syntactic and semantic information hides beneath. The objective of syntactic theory is to capture the shape of the hidden object and the nature of its correspondence with the observable string. Thus, a central question for linguistics is how strings and structures relate to one another. The majority of frameworks that exist today assume that word order must be accounted for exclusively in structural terms.
In phrase-structure grammar and its successors, structures are trees on which dominance and precedence relations are ex- haustively specified. The words of the linguistic expression constitute the leaf nodes of the tree, and the precedence relations between them determine the string. Hence, the theory of grammar is only a theory of the structure-generating component. A theory of structure-string mapping is not required, beyond defining the latter as lower-dimensional projections of the former.
However, it is possible to design a theory of grammar which views the structure-generating component as separate from the linearization component, which could take into account any combination of structural, phonological and discourse facts. Under this view, structural relationships influence the string form, but may be overwritten by more powerful considerations. This dissertation will show that separating the theory of structures from the theory of strings is particularly fruitful in conjunction with an Optimality Theoretic view of word order variation. I will explore a theory of grammar in which structural descriptions capture the abstract syntactic relationships among words but make no direct reference to word order.
Instead, the word order corresponding to the structure emerges through the interaction of mutually contradicting ranked constraints. While such a system can be built on the basis of any number of existing formalisms, this work is situated within the dependency grammar framework. The choice is moti- vated by a desire to avoid unnecessary theory-driven structural complexity at this early stage, particularly because a significant portion of the massive theoretical machinery accumulated by constituent-based formalisms is due to the fusion of hiearchical struc- ture with precedence. In defining a system that treats these components separately, it is preferable to start with the most transparent type of structures.
What is stucture? No pre-theoretical reason forces us to confine this notion to any particular set of abstract objects. There is general agreement that conditional and se- mantic dependencies among words, such as thematic roles and subcategorization frames, should be reflected at the structural level, be it configurationally or substantively. While some approaches treat agreement as a conditional dependency, other choose to attribute it to psycholinguistic factors external to the grammar. The situation is similar with re- spect to discourse features, such as prominence or novelty.
In some frameworks these are presented on a par with selectional properties of lexical items, while others leave them to external components. Some of these questions are partly contingent on the extent to which word order should treated as part of the strunctural description. If word order is taken to be a purely structural consequence, any difference in word order implies a structural difference.