IRCS Technical Reports Series

 

Search results

Now showing 1 - 10 of 199
  • Publication
    Computational Methods for Realistic Image Synthesis
    (1996-12-01) Shao, Min-Zhi
    In this thesis, we investigate the computational methods for both diffuse and general reflections in realistic image synthesis and propose two new approaches: the overrelaxation solution and the Bernstein polynomial solution. One of the major concerns with the radiosity method is its expensive computing time and memory requirements. In this thesis, we analyze the convergence behavior of the progressive refinement radiosity method and propose two overrelaxation algorithms: the gathering and shooting solution and the positive overshooting solution. We modify the conventional shooting method to make the optimal use of the visibility information computed in each iteration. Based on a concise record of the history of the unshot light energy distribution, a solid convergence speed-up is achieved. Though a great effort has been made to extend the radiosity method to accommodate general non-diffuse reflection, the current algorithms are still quite limited to simple environment settings. In this thesis, we propose using the piecewise spherical Bernstein basis functions over a geodesic triangulation to represent the radiance function. The representation is intrinsic to the unit sphere, and can be efficiently stored, evaluated, and subdivided by the numerically stable de Casteljau algorithm. We demonstrate that the computation of other fundamental radiometric quantities such as vector irradiance and reflected radiance can be reduced to the integration of the piecewise spherical Bernstein basis functions. A novel geometric integration algorithm based on adaptive domain subdivision is presented for the Bernstein-B´ezier polynomials over a geodesic triangle on the unit sphere.
  • Publication
    Parallel Programming Languages for Collections
    (1995-08-01) Suciu, Dan
    The thesis discusses the design, expressive power, and implementation of parallel programming languages for collections, the fragment dealing with collections of an object-oriented query language. The Relational Algebra has a simple, intrinsic parallel semantics, which enabled the successful development of parallel relational database systems. But the implementation techniques of these systems do not carry over to the more complex object-oriented databases. In order to develop efficient parallel object-oriented database systems, one needs to (1) design their query languages with parallelism in mind, and (2) find new implementation techniques, specially designed for these languages. Here we pursue these goals for parallel languages for collections. The collections of interest for us are sets, bags, and sequences (lists). We start by describing a basic collection calculus and additional forms of recursion on collections. They have an idealized parallel "execution", assuming unbounded resources and instant communication, which gives us high-level parallel complexity measures. An interesting fragment of the calculus expresses exactly the queries in the parallel complexity class NC. Here the salient construct is divide and conquer recursion on sets. Sublanguages obtained by imposing a bound k on the number of recursion nesting correspond to the subclasses ACk, for k≥1. We break the implementation of the calculus into three steps. First, sets and bags are implemented on sequences, using high-level parallel algorithms we express such algorithms in a high-level language for sequences called MAP, built around a new form of recursion. Second, we describe a complexity-preserving compilation of MAP on a simple vector-parallel model. Third, we implement the vector model on a parallel multiprocessor. Here we choose as target the LogP model, which can be instantiated to simulate various multiprocessors. All but one of the vector model instructions require only restricted forms of communication patters on LogP, called monotone communications. These in turn admit efficient implementations on LogP. We ran two simple benchmarks on a LogP simulator, measuring the speedup and the scaleup. We report conditions under which good speedup and scaleup can be expected.
  • Publication
    An Image-Based Framework For Global Illumination In Animated Environments
    (1997-12-01) Nimeroff, Jeffry S.
    Interacting with environments exhibiting the subtle lighting effects found in the real world gives designers a better understanding of the scene’s structure by providing rich visual cues. The major hurdle is that global illumination algorithms are too inefficient to quickly compute their solutions for reasonably sized environments. When motion is allowed within the environment, the problem becomes even more intractable. We address the problem of sampling and reconstructing an environment’s time-varying radiance distribution, its spatio-temporal global illumination information, allowing the efficient generation of arbitrary views of the environment at arbitrary points in time. The radiance distribution formalizes incoming chromatic radiance at all points within a constrained view space, along all directions, at all times. Since these distributions cannot, in general, be calculated analytically, we introduce a framework for specifying and computing sample values from the distribution and progress through a series of sample-based approximations designed to allow easy and accurate reconstruction of images extracted from the distribution. The first approximation is based on storing time-sequences of images at strategic locations within the chosen view space. An image of the environment is constructed by first blending the images contained in the individual time-sequences to get the desired time and then using view interpolation to merge the proximate views. The results presented here demonstrate the feasibility and utility of the method but also show its major drawback. An inability to accurately model the temporal radiance variations using image sequences without resorting to a high sampling rate leads to the replacement of the image sequences by a sparse temporal image volume representation for storing randomly, or adaptively, placed radiance samples. Triangulation techniques are then used to reconstruct the radiance distribution at the desired time from a proximate set of stored spatio-temporal radiance samples. The results presented here show that temporal image volumes allow for more accurate and efficient temporal reconstruction requiring less sampling than the more traditional time-sequence approach.
  • Publication
    Complexity of Lexical Descriptions and its Relevance to Partial Parsing
    (1997-06-01) Bangalore, Srinivas
    In this dissertation, we have proposed novel methods for robust parsing that integrate the flexibility of linguistically motivated lexical descriptions with the robustness of statistical techniques. Our thesis is that the computation of linguistic structure can be localized if lexical items are associated with rich descriptions (supertags) that impose complex constraints in a local context. However, increasing the complexity of descriptions makes the number of different descriptions for each lexical item much larger and hence increases the local ambiguity for a parser. This local ambiguity can be resolved by using supertag co-occurrence statistics collected from parsed corpora. We have explored these ideas in the context of Lexicalized Tree-Adjoining Grammar (LTAG) framework wherein supertag disambiguation provides a representation that is an almost parse. We have used the disambiguated supertag sequence in conjunction with a lightweight dependency analyzer to compute noun groups, verb groups, dependency linkages and even partial parses. We have shown that a trigram-based supertagger achieves an accuracy of 92.1‰ on Wall Street Journal (WSJ) texts. Furthermore, we have shown that the lightweight dependency analysis on the output of the supertagger identifies 83‰ of the dependency links accurately. We have exploited the representation of supertags with Explanation-Based Learning to improve parsing effciency. In this approach, parsing in limited domains can be modeled as a Finite-State Transduction. We have implemented such a system for the ATIS domain which improves parsing eciency by a factor of 15. We have used the supertagger in a variety of applications to provide lexical descriptions at an appropriate granularity. In an information retrieval application, we show that the supertag based system performs at higher levels of precision compared to a system based on part-of-speech tags. In an information extraction task, supertags are used in specifying extraction patterns. For language modeling applications, we view supertags as syntactically motivated class labels in a class-based language model. The distinction between recursive and non-recursive supertags is exploited in a sentence simplification application.
  • Publication
    A Computational Approach to Aspectual Composition
    (1994-08-01) White, Michael
    In recent years, it has become common in the linguistics and philosophy literature to assume that events and processes are ontologically distinct entities, on a par with objects and substances. At the same time, the idea that time-based episodic knowledge should be represented as a collection of interrelated eventualities has gained increasing acceptance in the computational linguistics and articial intelligence literature. Contrary to what one might expect, a search through the prior literature in linguistics and philosophy reveals no account in which these sortal distinctions play a direct role in adequately explaining the problem of aspectual composition and the closely related imperfective paradox. In the computational linguistics and artificial intelligence literature, moreover, relatively little attention has been paid to either problem. In the first part of the dissertation, I investigate the hypothesis that the parallel ontological distinctions introduced above may be directly employed in an explanatory formal account of the problem of aspectual composition and the imperfective paradox. In so doing, I develop a synthesis of proposals by Hinrichs (1985), Krifka (1989; 1992) and Jackendo (1991) which makes correct predictions in many cases not considered by these authors. In particular, the account is the first to adequately explain the syntactic and semantic behavior of non-individuating accomplishment expressions, such as Jack pour some amount of wort into the carboy, which are too vague to individuate a single event but nevertheless behave like other Vendlerian accomplishments. In the second part of the dissertation, I explore the potential computational applications of the linguistic account, by way of two case studies. In the first one, I follow Moens (1987) in showing how a calculus of eventualities can facilitate the implementation of a simple statement verifier which allows for a much greater range of natural language queries than is usually the case with temporal databases. In the second, more preliminary study, I examine the relevance of the model-theoretic analysis to discourse interpretation, within the context of devising a program which produces simple microworld animations using short narrative descriptions as input specifications.
  • Publication
    The Computational Analysis of the Syntax and Interpretation of "Free" Word Order in Turkish
    (1995-06-01) Hoffman, Beryl
    In this dissertation, I examine a language with “free” word order, specifically Turkish, in order to develop a formalism that can capture the syntax and the context-dependent interpretation of “free” word order within a computational framework. In “free” word order languages, word order is used to convey distinctions in meaning that are not captured by traditional truth-conditional semantics. The word order indicates the “information structure”, e.g. what is the “topic” and the “focus” of the sentence. The context-appropriate use of “free” word order is of considerable importance in developing practical applications in natural language interpretation, generation, and machine translation. I develop a formalism called Multiset-CCG, an extension of Combinatory Categorial Grammars, CCGs, (Ades/Steedman 1982, Steedman 1985), and demonstrate its advantages in an implementation of a data-base query system that interprets Turkish questions and generates answers with contextually appropriate word orders. Multiset-CCG is a context-sensitive and polynomially parsable grammar that captures the formal and descriptive properties of “free” word order and restrictions on word order in simple and complex sentences (with discontinuous constituents and long distance dependencies). Multiset-CCG captures the context-dependent meaning of word order in Turkish by compositionally deriving the predicate-argument structure and the information structure of a sentence in parallel. The advantages of using such a formalism are that it is computationally attractive and that it provides a compositional and flexible surface structure that allows syntactic constituents to correspond to information structure constituents. A formalism that integrates information structure and syntax such as Multiset-CCG is essential to the computational tasks of interpreting and generating sentences with contextually appropriate word orders in “free” word order languages.
  • Publication
    Formal and Computational Aspects of Natural Language Syntax
    (1994-05-01) Rambow, Owen
    This thesis explores issues related to using a restricted mathematical formalism as the formal basis for the representation of syntactic competence and the modeling of performance. The specific contribution of this thesis is to examine a language with considerably freer word-order than English, namely German, and to investigate the formal requirements that this syntactic freedom imposes. Free word order (or free constituent order) languages can be seen as a test case for linguistic theories, since presumably the stricter word order can be subsumed by an apparatus that accounts for freer word order. The formal systems investigated in this thesis are based on the tree adjoining grammar (TAG) formalism of Joshi et al. (1975). TAG is an appealing formalism for the representation of natural language syntax because its elementary structures are phrase structure trees, which allows the linguist to localize linguistic dependencies such as agreement, subcategorization, and filler-gap relations, and to develop a theory of grammar based on the lexicon. The main results of the thesis are an argument that simple TAGs are formally inadequate, and the definition of an extension to TAG that is. Every aspect of the definition of this extension to TAG, called V-TAG, is specifically motivated by linguistic facts, not by formal considerations. A formal investigation of V-TAG reveals that (when lexicalized) it has restricted generative capacity, that it is polynomial parsable, and that it forms an abstract family of languages. This means that it has desirable formal properties for representing natural language syntax. Both a formal automaton and a parser for V-TAG are presented. As a consequence of the new system, a reformulation of the linguistic theory that has been proposed for TAG suggests itself. Instead of including a transformational step in the theory of grammar, all derivations are performed within mathematically defined formalisms, thus limiting the degrees of freedom in the linguistic theory, and making the theory more appealing from a computational point of view. This has several interesting linguistic consequences; for instance, functional categories are expressed by feature content (not node labels), and head movement is replaced by the adjunction of heads. The thesis sketches a fragment of a grammar of German, which covers phenomena such as scrambling, extraposition, topicalization, and the V2 effect. Finally, the formal automaton for V-TAG is used as a model of human syntactic processing. It is shown that this model makes several interesting predictions related to free word order in German.
  • Publication
    Covert Modality in Non-Finite Contexts
    (2000-01-01) Bhatt, Rajesh
    This dissertation investigates the distribution and interpretation of covert modality. Three environments where covert modality appears are analyzed. These environments are infinitival relative clauses, infinitival questions, and ability modals. Infinitival relative clauses are shown to not form a unified class structurally. Subject infinitival relative clauses are assimilated to the class of reduced relative clauses. They lack a CP projection. Non-subject infinitival relative are assimilated to the class of full relative clauses. Like full relatives, they have a CP projection. The infinitival [+wh] Cº is argued to be the source of the modality in a non-subject infinitival relative clause. All nonsubject infinitival relative clauses and infinitival questions involve modality because of the obligatory presence of the infinitival [+wh] Cº. Since subject infinitival relative clauses do not involve the infinitival [+wh] Cº, they are not necessarily modal. If they are modal, the source of the modality lies within the infinitival clause. The conditions under which a subject infinitival relative can receive a non-modal interpretation are analyzed. It is shown that a non-modal interpretation must be licensed and that only a limited class of modifiers (superlatives, ordinals, and only) in a particular configuration can license the non-modal interpretation. The licensing configuration obtains under reconstruction of the head NP of the relative clause. The basic result regarding themodality in infinitival questions is that despite the apparent variability in the nature (deontic vs. circumstantial) and the force of this modality, we really have just one modality. The apparent variation in force and nature falls out from the interaction between the semantics of the infinitival [+wh] C0 and contextual factors. The apparent variation in the force and nature of the infinitival questionmodality is also found in non-subject infinitival relative clauses. The proposal for capturing variable modality effects in infinitival questions is extended to non-subject infinitival relatives. A covert modal (the Generic operator) is responsible for the modality in an ability modal. The ability modal itself has the semantics of an implicative verb like manage.
  • Publication
    Tree Description Grammars and Underspecified Representations
    (1999) Kallmeyer, Laura
    In this thesis, a new grammar formalism called (local) Tree Description Grammar (TDG) is presented that generates tree descriptions. This grammar formalism brings together some of the central ideas in the context of Tree Adjoining Grammars (TAG) on the one hand, and approaches to underspecified semantics for scope ambiguities on the other hand. First a general definition of TDGs is presented, and afterwards a restricted variant called local TDGs is proposed. Since the elements of a local TDG are tree descriptions, an extended domain of locality as in TAGs is provided by this formalism. Consequently, local TDGs can be lexicalized, and local dependencies such as filler gap dependencies can be expressed in the descriptions occurring in the grammar. The tree descriptions generated by local TDGs are such that the dominance relation (i.e. the reflexive and transitive closure of the parent relation) need not be fully specified. Therefore the generation of suitable underspecified representations for scope ambiguities is possible. The generative capacity of local TDGs is greater than the one of TAGs. Local TDGs are even more powerful than set-local multicomponent TAGs (MC-TAG). However, the generative capacity of local TDGs is restricted in such a way that only semilinear languages are generated. Therefore these languages are of constant growth, a property generally ascribed to natural languages. Local TDGs of different rank can be distinguished depending on the form of derivation steps that are possible in these grammars. This leads to a hierarchy of local TDGs. For the string languages generated by local TDGs of a certain rank, a pumping lemma is proven that allows to show that local TDGs of rank n can generate a language Li := {a1k···a1k|k ≥ 0} iff i ≤ 2n holds. In order to describe the relation between two languages, synchronous local TDGs are introduced. The synchronization with a second local TDG does not increase the generative power of the grammar in the sense that each language generated by a local TDG that is part of a synchronous pair of local TDGs, also can be generated by a single local TDG. This formalism of synchronous local TDGs is used to describe a syntax-semantics interface for a fragment of French which illustrates the derivation of underspecified representations for scope ambiguities with local TDGs.
  • Publication
    Description Based Parsing in a Connectionist Network
    (1994-09-01) Henderson, James
    Recent developments in connectionist architectures for symbolic computation have made it possible to investigate parsing in a connectionist network while still taking advantage of the large body of work on parsing in symbolic frameworks. This dissertation investigates syntactic parsing in the temporal synchrony variable binding model of symbolic computation in a connectionist network. This computational architecture solves the basic problem with previous connectionist architectures, while keeping their advantages. However, the architecture does have some limitations, which impose computational constraints on parsing in this architecture. This dissertation argues that, despite these constraints, the architecture is computationally adequate for syntactic parsing and that these constraints make significant linguistic predictions. To make these arguments, the nature of the architecture's limitations are first characterized as a set of constraints on symbolic computation. This allows the investigation of the feasibility and implications of parsing in the architecture to be investigated at the same level of abstraction as virtually all other investigations of syntactic parsing. Then a specific parsing model is developed and implemented in the architecture. The extensive use of partial descriptions of phrase structure trees is crucial to the ability of this model to recover the syntactic structure of sentences within the constraints. Finally, this parsing model is tested on those phenomena which are of particular concern given the constraints, and on an approximately unbiased sample of sentences to check for unforeseen difficulties. The results show that this connectionist architecture is powerful enough for syntactic parsing. They also show that some linguistic phenomena are predicted by the limitations of this architecture. In particular, explanations are given for many cases of unacceptable center embedding, and for several significant constraints on long distance dependencies. These results give evidence for the cognitive significance of this computational architecture and parsing model. This work also shows how the advantages of both connectionist and symbolic techniques can be unified in natural language processing applications. By analyzing how low level biological and computational considerations influence higher level processing, this work has furthered our understanding of the nature of language and how it can be efficiently and effectively processed.