Proceedings with the revised papers of the FSMNLP (Finite-state Methods and Natural Language Processing) 2007 Workshop in Potsdam / Tagungsband mit den Beiträgen der FSMNLP (Finite-state Methods and Natural Language Processing) 2007 in Potsdam

In this thesis we will introduce multioperator weighted monadic datalog (mwmd), a formal model for specifying tree series, tree transformations, and tree languages. This model combines aspects of multioperator weighted tree automata (wmta), weighted monadic datalog (wmd), and monadic datalog tree transducers (mdtt). In order to develop a rich theory we will define multiple versions of semantics for mwmd and compare their expressiveness. We will study normal forms and decidability results of mwmd and show (by employing particular semantic domains) that the theory of mwmd subsumes the theory of both wmd and mdtt. We conclude this thesis by showing that mwmd even contain wmta as a syntactic subclass and present results concerning this subclass.

Beschreibungslogiken (Description logics, DLs) sind eine Klasse von Wissensrepraesentationsformalismen mit wohldefinierter, logik-basierter Semantik und entscheidbaren Schlussfolgerungsproblemen, wie z.B. dem Erfuellbarkeitsproblem. Zwei wichtige Entscheidungsverfahren fuer das Erfuellbarkeitsproblem von DL-Ausdruecken sind Tableau- und Automaten-basierte Algorithmen. Diese haben aufgrund ihrer unterschiedlichen Arbeitsweise komplementaere Eigenschaften: Tableau-Algorithmen eignen sich fuer Implementierungen und fuer den Nachweis von PSPACE- und NEXPTIME-Resultaten, waehrend Automaten sich besonders fuer EXPTIME-Resultate anbieten. Zudem ermoeglichen sie eine vom Standpunkt der Theorie aus elegantere Handhabung von unendlichen Strukturen, eignen sich aber wesentlich schlechter fuer eine Implementierung. Ziel der Dissertation ist es, die Gruende fuer diese Unterschiede zu analysieren und Moeglichkeiten aufzuzeigen, wie Eigenschaften von einem Ansatz auf den anderen uebertragen werden koennen, um so die positiven Eigenschaften von beiden Ansaetzen miteinander zu verbinden. Unter Anderem werden Methoden entwickelt, mit Hilfe von Automaten PSPACE-Resultate zu zeigen, und von einem Tableau-Algorithmus automatisch ein EXPTIME-Resultat abzuleiten. / Description Logics (DLs) are a family of knowledge representation languages with well-defined logic-based semantics and decidable inference problems, e.g. satisfiability. Two of the most widely used decision procedures for the satisfiability problem are tableau- and automata-based algorithms. Due to their different operation, these two classes have complementary properties: tableau algorithms are well-suited for implementation and for showing PSPACE and NEXPTIME complexity results, whereas automata algorithms are particularly useful for showing EXPTIME results. Additionally, they allow for an elegant handling of infinite structures, but they are not suited for implementation. The aim of this thesis is to analyse the reasons for these differences and to find ways of transferring properties between the two approaches in order to reconcile the positive properties of both. For this purpose, we develop methods that enable us to show PSPACE results with the help of automata and to automatically derive an EXPTIME result from a tableau algorithm.

The specification of a decoder, i.e., a program that translates sentences from one natural language into another, is an intricate process, driven by the application and lacking a canonical methodology. The practical nature of decoder development inhibits the transfer of knowledge between theory and application, which is unfortunate because many contemporary decoders are in fact related to formal-language theory. This thesis proposes an algebraic framework where a decoder is specified by an expression built from a fixed set of operations. As yet, this framework accommodates contemporary syntax-based decoders, it spans two levels of abstraction, and, primarily, it encourages mutual stimulation between the theory of weighted tree automata and the application.

In this thesis we will introduce multioperator weighted monadic datalog (mwmd), a formal model for specifying tree series, tree transformations, and tree languages. This model combines aspects of multioperator weighted tree automata (wmta), weighted monadic datalog (wmd), and monadic datalog tree transducers (mdtt). In order to develop a rich theory we will define multiple versions of semantics for mwmd and compare their expressiveness. We will study normal forms and decidability results of mwmd and show (by employing particular semantic domains) that the theory of mwmd subsumes the theory of both wmd and mdtt. We conclude this thesis by showing that mwmd even contain wmta as a syntactic subclass and present results concerning this subclass.

Beschreibungslogiken (Description logics, DLs) sind eine Klasse von Wissensrepraesentationsformalismen mit wohldefinierter, logik-basierter Semantik und entscheidbaren Schlussfolgerungsproblemen, wie z.B. dem Erfuellbarkeitsproblem. Zwei wichtige Entscheidungsverfahren fuer das Erfuellbarkeitsproblem von DL-Ausdruecken sind Tableau- und Automaten-basierte Algorithmen. Diese haben aufgrund ihrer unterschiedlichen Arbeitsweise komplementaere Eigenschaften: Tableau-Algorithmen eignen sich fuer Implementierungen und fuer den Nachweis von PSPACE- und NEXPTIME-Resultaten, waehrend Automaten sich besonders fuer EXPTIME-Resultate anbieten. Zudem ermoeglichen sie eine vom Standpunkt der Theorie aus elegantere Handhabung von unendlichen Strukturen, eignen sich aber wesentlich schlechter fuer eine Implementierung. Ziel der Dissertation ist es, die Gruende fuer diese Unterschiede zu analysieren und Moeglichkeiten aufzuzeigen, wie Eigenschaften von einem Ansatz auf den anderen uebertragen werden koennen, um so die positiven Eigenschaften von beiden Ansaetzen miteinander zu verbinden. Unter Anderem werden Methoden entwickelt, mit Hilfe von Automaten PSPACE-Resultate zu zeigen, und von einem Tableau-Algorithmus automatisch ein EXPTIME-Resultat abzuleiten. / Description Logics (DLs) are a family of knowledge representation languages with well-defined logic-based semantics and decidable inference problems, e.g. satisfiability. Two of the most widely used decision procedures for the satisfiability problem are tableau- and automata-based algorithms. Due to their different operation, these two classes have complementary properties: tableau algorithms are well-suited for implementation and for showing PSPACE and NEXPTIME complexity results, whereas automata algorithms are particularly useful for showing EXPTIME results. Additionally, they allow for an elegant handling of infinite structures, but they are not suited for implementation. The aim of this thesis is to analyse the reasons for these differences and to find ways of transferring properties between the two approaches in order to reconcile the positive properties of both. For this purpose, we develop methods that enable us to show PSPACE results with the help of automata and to automatically derive an EXPTIME result from a tableau algorithm.

In this thesis, we investigate different theoretical questions concerning weighted automata models over tree-like input structures. First, we study exact and approximated determinisation and then, we turn to Kleene-like and Büchi-like characterisations. We consider multiple weighted automata models, including weighted tree automata over semirings (Chapters 3 and 4), weighted forest automata over M-monoids (Chapter 5), and rational weighted tree languages with storage (Chapter 6). For an explanation as to why the last class can be considered as a weighted automaton model, we refer to page 188 of the thesis. We will now summarise the main contributions of the thesis. In Chapter 3, we focus on the determinisation of weighted tree automata and present our determinisation framework, called M-sequentialisation, which can model different notions of determinisation from the existing literature. Then, we provide a positive M-sequentialisation result for the case of additively idempotent semirings or finitely M-ambiguous weighted tree automata. Another important contribution of Chapter 3 is Theorem 77, where we provide a blueprint theorem that can be used to find determini- sation results for more classes of semirings and weighted tree automata easily. In fact, instead of repeating an entire determinisation construction, Theorem 77 allows us to prove a determinisation result by finding certain finite equivalence relations. This is a very potent tool for future research in the area of determinisation. In Chapter 4, we move from exact determinisation towards approximate determini- sation. We lift the formalisms and the main results from one approach from the literature from the word case to the tree case. This successfully results in an approximated determinisation construction for weighted tree automata over the tropical semiring. We provide a formal mathematical description of the approximated determinisation construction, rather than an algorithmic description as found in the related approach from the literature. In Chapter 5, we turn away from determinisation and instead consider Kleene-like and Büchi-like characterisations of weighted recognisability. We introduce weighted forest automata over M-monoids, which are a generalisation of weighted tree automata over M-monoids and weighted forest automata over semirings. Then, we prove that our recognisable weighted forest languages can be decomposed into a finite product of recognisable weighted tree languages. We also prove that the initial algebra semantic and the run semantic for weighted forest automata are equivalent under certain conditions. Lastly, we define rational forest expressions and forest M-expressions and and prove that the classes of languages generated by these formalisms coincide with recognisable weighted forest languages under certain conditions. In Chapter 6, we consider rational weighted tree languages with storage, where the storage is introduced by composing rational weighted tree languages without storage with a storage map. It has been proven in the literature that rational weighted tree languages with storage are closed under the rational operations. In Chapter 6, we provide alternative proofs of these closure properties. In fact, we prove that our way of introducing storage to rational weighted tree languages preserves the closure properties from rational weighted tree languages without storage.:1 Introduction 2 Preliminaries 2.1 Languages 2.2 WeightedLanguages 2.3 Weighted Tree Automata 3 A Unifying Framework for the Determinisation of Weighted Tree Automata 3.1 Introduction 3.2 Preliminaries 3.3 Factorisation in Monoids 3.3.1 Ordering Multisets over Monoids 3.3.2 Cayley Graph and Cayley Distance 3.3.3 Divisors and Rests 3.3.4 Factorisation Properties 3.4 Weighted Tree Automata over M_fin(M) and the Twinning Property 3.4.1 Weighted Tree Automata over M_fin(M) 3.4.2 The Twinning Property 3.5 Sequentialisation of Weighted Tree Automata over M_fin(M) 3.5.1 The Sequentialisation Construction 3.5.2 The Finitely R-Ambiguous Case 3.6 Relating WTA over M_fin(M) and WTA over S 3.7 M-Sequentialisation of Weighted Tree Automata 3.7.1 Accumulation of D_B 3.7.2 M-Sequentialisation Results 3.8 Comparison of our Results to the Literature 3.8.1 Determinisation of Unweighted Tree Automata 3.8.2 The Free Monoid Case 3.8.3 The Group Case 3.8.4 The Extremal Case 3.9 Conclusion 4 Approximated Determinisation of Weighted Tree Automata 125 4.1 Introduction 4.2 Preliminaries 4.3 Approximated Determinisation 4.3.1 The Approximated Determinisation Construction 4.3.2 Correctness of the Construction 4.4 The Approximated Twinning Property 4.4.1 Implications for Approximated Determinisability 4.4.2 Decidability of the Twinning Property 4.5 Conclusion 5 Kleene and Büchi Theorems for Weighted Forest Languages over M-Monoids 5.1 Introduction 5.2 Preliminaries 5.3 WeightedForestAutomata 5.3.1 Forests 5.3.2 WeightedForestAutomata 5.3.3 Rectangularity 5.3.4 I-recognisable is R-recognisable 5.4 Kleene’s Theorem 5.4.1 Kleene’s Theorem for Trees 5.4.2 Kleene’s Theorem for Forests 5.4.3 An Inductive Approach 5.5 Büchi’s Theorem 5.5.1 Büchi’s Theorem for Trees 5.5.2 Büchi’s Theorem for Forests 5.6 Conclusion 6 Rational Weighted Tree Languages with Storage 6.1 Introduction 6.2 Preliminaries 6.3 Rational Weighted Tree Languages with Storage 6.4 The Kleene-Goldstine Theorem 6.5 Closure of Rat(S¢,Σ,S) under Rational Operations 6.5.1 Top-Concatenation, Scalar Multiplication, and Sum 6.5.2 α-Concatenation 6.5.3 α-Kleene Star 6.6 Conclusion 7 Outlook References

The field of statistical machine translation has made tremendous progress due to the rise of statistical methods, making it possible to obtain a translation system automatically from a bilingual collection of text. Some approaches do not even need any kind of linguistic annotation, and can infer translation rules from raw, unannotated data. However, most state-of-the art systems do linguistic structure little justice, and moreover many approaches that have been put forward use ad-hoc formalisms and algorithms. This inevitably leads to duplication of effort, and a separation between theoretical researchers and practitioners. In order to remedy the lack of motivation and rigor, the contributions of this dissertation are threefold: 1. After laying out the historical background and context, as well as the mathematical and linguistic foundations, a rigorous algebraic model of machine translation is put forward. We use regular tree grammars and bimorphisms as the backbone, introducing a modular architecture that allows different input and output formalisms. 2. The challenges of implementing this bimorphism-based model in a machine translation toolkit are then described, explaining in detail the algorithms used for the core components. 3. Finally, experiments where the toolkit is applied on real-world data and used for diagnostic purposes are described. We discuss how we use exact decoding to reason about search errors and model errors in a popular machine translation toolkit, and we compare output formalisms of different generative capacity.

The field of statistical machine translation has made tremendous progress due to the rise of statistical methods, making it possible to obtain a translation system automatically from a bilingual collection of text. Some approaches do not even need any kind of linguistic annotation, and can infer translation rules from raw, unannotated data. However, most state-of-the art systems do linguistic structure little justice, and moreover many approaches that have been put forward use ad-hoc formalisms and algorithms. This inevitably leads to duplication of effort, and a separation between theoretical researchers and practitioners. In order to remedy the lack of motivation and rigor, the contributions of this dissertation are threefold: 1. After laying out the historical background and context, as well as the mathematical and linguistic foundations, a rigorous algebraic model of machine translation is put forward. We use regular tree grammars and bimorphisms as the backbone, introducing a modular architecture that allows different input and output formalisms. 2. The challenges of implementing this bimorphism-based model in a machine translation toolkit are then described, explaining in detail the algorithms used for the core components. 3. Finally, experiments where the toolkit is applied on real-world data and used for diagnostic purposes are described. We discuss how we use exact decoding to reason about search errors and model errors in a popular machine translation toolkit, and we compare output formalisms of different generative capacity.

The specification of a decoder, i.e., a program that translates sentences from one natural language into another, is an intricate process, driven by the application and lacking a canonical methodology. The practical nature of decoder development inhibits the transfer of knowledge between theory and application, which is unfortunate because many contemporary decoders are in fact related to formal-language theory. This thesis proposes an algebraic framework where a decoder is specified by an expression built from a fixed set of operations. As yet, this framework accommodates contemporary syntax-based decoders, it spans two levels of abstraction, and, primarily, it encourages mutual stimulation between the theory of weighted tree automata and the application.