Automatization Using FDG

Embed Size (px)

Citation preview

  • 8/3/2019 Automatization Using FDG

    1/89

    Automated Natural Language ProcessingUsing Functional Discourse Grammar

    Reinier LamersApril 21, 2009

    head

    head

    operators modifiers

    operators

    multi

    recipeN

    fi

    cxi

    modifiers

    multi

    Supervisors:dr. F.P.M. Dignumdrs. N.L. Vergunstprof. dr. Kees Hengeveld

    Submitted as a masters thesis in Cognitive Artificial Intelligence atUniversiteit Utrecht for 30 European credits

    1

  • 8/3/2019 Automatization Using FDG

    2/89

    Preface

    The road leading up to the thesis you are reading has been longer and bumpierthan the graduation manual predicts, although this is hardly a surprise, espe-cially to a Cognitive Artificial Intelligence student.

    There are numerous people who supported me while I was cruising this road

    and its surroundings. I will take the opportunity to thank them here. Firstamong them are my parents Diny and Leo. I also wish to thank my supervisorsFrank and Nieske. Kees Hengeveld and Evelien Keizer in Amsterdam, whoalways had the patience to answer my questions about Functional DiscourseGrammar. My officemates in the computer science building, especially Barbaraand Eelco. And of course CAI students in general, for always being in the moodto discuss anything remotely related to AI in depth.

    The tree diagrams in this thesis, one of which made it to the cover page,were drawn using the dot2tex and dot2texi packages by Kjell Magne Fauske.I thank him, and the authors of all other open source tools used in my mastersthesis project, for sharing their work.

    2

  • 8/3/2019 Automatization Using FDG

    3/89

    Contents

    1 Introduction 6

    1.1 From Syntax to Semantics Using Functional Discourse Grammar 61.1.1 From Syntax to Semantics . . . . . . . . . . . . . . . . . . 7

    1.1.2 Why FDG? . . . . . . . . . . . . . . . . . . . . . . . . . . 71.2 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Notation conventions . . . . . . . . . . . . . . . . . . . . . . . . . 81.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

    2 Functional Discourse Grammar 9

    2.1 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.1 Discourse Act . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.2 No Transformations or Filters . . . . . . . . . . . . . . . . 102.1.3 A Single Theory of Language . . . . . . . . . . . . . . . . 112.1.4 Maximal Depth Principle . . . . . . . . . . . . . . . . . . 122.1.5 Typology-based . . . . . . . . . . . . . . . . . . . . . . . . 12

    2.1.6 Not a Theory of Discourse . . . . . . . . . . . . . . . . . . 132.2 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

    2.2.1 Levels, Formulation and Encoding . . . . . . . . . . . . . 132.2.2 Use of the Structure for Natural Language Processing . . 15

    2.3 Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.4 Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

    2.4.1 Formulation Primitives . . . . . . . . . . . . . . . . . . . 162.4.2 Morphosyntactic Encoding Primitives . . . . . . . . . . . 172.4.3 Phonological Encoding Primitives . . . . . . . . . . . . . 17

    2.5 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . . . . 172.6 Notation Conventions for FDG Layers . . . . . . . . . . . . . . . 18

    3 The Morphosyntactic and Representational Levels 20

    3.1 The Morphosyntactic Level . . . . . . . . . . . . . . . . . . . . . 203.1.1 The Formal Structure of the Morphosyntactic Level . . . 203.1.2 Positioning and Alignment . . . . . . . . . . . . . . . . . 24

    3.2 The Representational Level . . . . . . . . . . . . . . . . . . . . . 253.2.1 Semantic Categories . . . . . . . . . . . . . . . . . . . . . 253.2.2 Propositional Contents . . . . . . . . . . . . . . . . . . . . 263.2.3 States of Affairs . . . . . . . . . . . . . . . . . . . . . . . 283.2.4 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

    3

  • 8/3/2019 Automatization Using FDG

    4/89

    3.2.5 Individuals . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2.6 Coindexation . . . . . . . . . . . . . . . . . . . . . . . . . 363.2.7 Relation to the Morphosyntactic Level . . . . . . . . . . . 36

    4 Language-Independent Aspects 39

    4.1 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . 394.2 Overview of the Approach . . . . . . . . . . . . . . . . . . . . . . 40

    4.2.1 FDG Structures as Trees . . . . . . . . . . . . . . . . . . 404.2.2 The Three Stages . . . . . . . . . . . . . . . . . . . . . . . 42

    4.3 The Treewalk Stage . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.1 Form of Tree Chunks . . . . . . . . . . . . . . . . . . . . . 454.3.2 The Treewalk Procedure . . . . . . . . . . . . . . . . . . . 454.3.3 Implementation in Prolog . . . . . . . . . . . . . . . . . . 464.3.4 The Lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . 46

    4.4 The Composition Stage . . . . . . . . . . . . . . . . . . . . . . . 474.4.1 Form of Combination Criteria Nodes . . . . . . . . . . . . 474.4.2 The Composition Process . . . . . . . . . . . . . . . . . . 48

    4.5 Coreference Resolution . . . . . . . . . . . . . . . . . . . . . . . . 514.5.1 Implicit Subjects . . . . . . . . . . . . . . . . . . . . . . . 52

    5 Language-Dependent Aspects for a Fragment of English 54

    5.1 The Test Dialog . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545.1.1 Morphosyntactic Analyses . . . . . . . . . . . . . . . . . . 555.1.2 Representational Analyses . . . . . . . . . . . . . . . . . . 56

    5.2 The Accepted Fragment of English . . . . . . . . . . . . . . . . . 575.3 The Node Function and Matching Function . . . . . . . . . . . . 58

    5.3.1 Grammatical Words . . . . . . . . . . . . . . . . . . . . . 585.3.2 Noun Phrases . . . . . . . . . . . . . . . . . . . . . . . . . 585.3.3 Verb Phrases . . . . . . . . . . . . . . . . . . . . . . . . . 635.3.4 Adjective Phrases . . . . . . . . . . . . . . . . . . . . . . 675.3.5 Adposition Phrases . . . . . . . . . . . . . . . . . . . . . . 67

    5.4 Representational Frames . . . . . . . . . . . . . . . . . . . . . . . 685.5 Initial open spot . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

    6 An Example 71

    6.1 The Treewalk Stage . . . . . . . . . . . . . . . . . . . . . . . . . 726.2 The Composition Stage . . . . . . . . . . . . . . . . . . . . . . . 79

    7 Discussion and Conclusion 82

    7.1 Discussion and Future Work . . . . . . . . . . . . . . . . . . . . . 827.1.1 Limitations of the Node Function . . . . . . . . . . . . . . 827.1.2 Universal Combination Criteria . . . . . . . . . . . . . . . 827.1.3 Towards a Complete FDG-based Understanding System . 837.1.4 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . 837.1.5 Coreference Resolution . . . . . . . . . . . . . . . . . . . . 837.1.6 Use High-Level Information . . . . . . . . . . . . . . . . . 84

    7.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

    4

  • 8/3/2019 Automatization Using FDG

    5/89

    A Samenvatting in het Nederlands 87

    A.1 Inleiding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87A.2 Functionele discourse-grammatica . . . . . . . . . . . . . . . . . . 87

    A.3 De aanpak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

    5

  • 8/3/2019 Automatization Using FDG

    6/89

    Chapter 1

    Introduction

    Imagine a clever, friendly machine in your kitchen. Imagine that its there

    when you need it, and that it listens to your calls for help in ordinary Englishor Dutch. It can tell you what you can cook with the available ingredients. Itcan tell you how you can replace a missing ingredient. It can tell you how thinyou should slice the tomatoes. Again, it tells you all this in ordinary naturallanguage. And best of all, this clever machine is aware when youre tired or ina hurry, and it adapts its behavior accordingly.

    To build such a friendly machine from dead materials, one needs a lot oftechnology from the field of Artificial Intelligence. One needs language andspeech generation. One needs a model of the environment including the emo-tions and beliefs of the users. One needs reasoning systems to come up withsolutions to the users problems. And last but not least, one needs language

    and speech recognition to figure out what a user is saying to the machine.Actually, people are now trying to build this machine. This project is known

    as the Dutch Companion Project, which has a website at http://www.decis.nl/content/view/50/41/. The aim of this thesis is to describe a part of anatural language processing system that could be used in this machine.

    The machine will have to be able to make appropriate deductions from aspeech signal. It receives an acoustic signal from its sensors, and finally thespeech in that signal must bring about changes in the machines model of theenvironment and the users. Between those two endpoints lie many processingsteps, from noise reduction to parsing the linguistic utterance present in thesignal. This thesis focuses on one of those processing steps: the step from the

    syntax of a linguistic utterance to the semantics of that utterance.

    1.1 From Syntax to Semantics Using Functional Dis-

    course Grammar

    This thesis will describe a computer program that takes a representation of thesyntax of an utterance, and produces a representation of the semantics of theutterance. These representations of syntax and semantics follow the schemeof Functional Discourse Grammar (FDG) [Hengeveld and Mackenzie, 2008][Hengeveld and Mackenzie, 2006], a grammar theory described in chapter 2.

    6

    http://www.decis.nl/content/view/50/41/http://www.decis.nl/content/view/50/41/http://www.decis.nl/content/view/50/41/http://www.decis.nl/content/view/50/41/http://www.decis.nl/content/view/50/41/
  • 8/3/2019 Automatization Using FDG

    7/89

    FDG has also inspired the design of the approach that is used to compute thesemantics from the syntax.

    1.1.1 From Syntax to Semantics

    The step from syntax to semantics is not as well-researched as the step fromtext to syntax. While there are many well-known parsing algorithms for naturallanguage, like CYK, Earley or HPSG-based parsers, I do not know of anysyntax-to-semantics technology that is equally well-known.

    The work of Richard Montague [Montague, 1970], [Montague, 1973] on sys-tematically extracting the meaning of a text out of its form has been extremelyinfluential. It has been taken up by the categorial grammar community amongothers, as explained in [Moortgat, 2002].

    When using this Montague method, the semantics of an utterance are com-

    monly represented as formulas in a logic formalism. While this allows one tocompute the truth of an utterance, it also has a number of drawbacks. Mostnotably, some linguistic utterances, like exclamations or imperatives, are notnaturally analyzed as logical sentences that can be true or false.

    Therefore, in this thesis, I use data structures postulated by FDG to repre-sent the semantics of an utterance instead of logical formulas.

    1.1.2 Why FDG?

    One reason to choose FDG as the linguistic foundation to build the computerprogram on, was that FDG is a complete linguistic theory. FDG covers thephonology, the morphosyntax, the semantics and the pragmatics of an utterance

    in one coherent theory. Using one theory that covers all these aspects is muchmore convenient than using four different theories.

    Also, FDG already describes the structure of linguistic data to some extentwith its notion oflayers, as described in chapter 2. This made it easier to designthe data structures for the program.

    The most important reason to use FDG however, is that FDG promisesto make it easy to use data from different sources to find out the speakersintention. The kitchen machine will have a user model, that it can use topredict the response of the user. It will also have a model of the environment,that contains many objects that the user may refer to in his or her utterances.FDG offers us ways to integrate these sources of knowledge with the language

    processing faculty of the machine. This point is explained in 2.2.2.

    1.2 Research Question

    The research question that this thesis will attempt to answer is: How doesone compute the semantics of an utterance, given its morphosyntax? In thisquestion, semantics and morphosyntax are understood as defined and rep-resented according to Functional Discourse Grammar.

    Note that I am not going to investigate how this system could discriminatebetween grammatical and ungrammatical input. The system is not a syntax

    7

  • 8/3/2019 Automatization Using FDG

    8/89

    checker. It will give a meaning to the syntactic representations of utterancesthat cannot occur in correct English for example, it will accept The chaircan I prepare?, treating it much like What can I prepare?.

    In addition to this main question, I also briefly discuss computing the prag-matics of an utterance, where pragmatics is again understood as defined andrepresented according to Functional Discourse Grammar.

    1.3 Notation conventions

    I have used the following conventions about notation in this thesis:

    The notation for glosses and FDG formulas follows[Hengeveld and Mackenzie, 2008], unless stated otherwise.

    While the thesis describes a computer program, I have avoided displayingprogram code as much as possible to keep the thesis readable. Readersthat are interested in the source code of the program, can find it at http://reinier.de/thesis/program-source.zip .

    In English, I use male pronouns when writing about humans of whom thegender is not known

    Glosses have translations below them in single quotes. FDG formulaexamples have the orthographic form of the constituent they representbelow them in double quotes.

    Linguistic examples are given in double quotes, quoted names are givenin single quotes.

    1.4 Outline

    After this introduction to the thesis comes an introduction to Functional Dis-course Grammar in chapter 2. This chapter will give the reader an understand-ing of the structure and principles of FDG in general.

    Chapter 3 is about the structure of the representations that FDG employsto describe morphosyntax and semantics.

    Chapter 4 describes the structure of the computer program that I developed

    to compute semantics from morphosyntax, and the data structures this programuses.Chapter 5 describes how the program was applied to a fragment of English.

    This fragment of English was determined by an actual dialog between twopeople.

    Chapter 6 concludes the thesis by discussing the results and their relevance.

    8

    http://reinier.de/thesis/program-source.ziphttp://reinier.de/thesis/program-source.ziphttp://reinier.de/thesis/program-source.ziphttp://reinier.de/thesis/program-source.zip
  • 8/3/2019 Automatization Using FDG

    9/89

    Chapter 2

    Functional Discourse

    Grammar

    Functional Discourse Grammar (FDG) [Hengeveld and Mackenzie, 2008][Hengeveld and Mackenzie, 2006] is a grammar framework for natural language,based on the Functional Grammar framework which is described in [Dik, 1989].Functional Discourse Grammar explains the shape of natural language utter-ances in terms of the goals and knowledge of natural language users. By doingthat, Functional Discourse Grammar follows a functional approach to model-ing language. This functional approach contrasts with the formal approach tomodeling language, which views language as an abstract entity, with no regardto how language is used.

    In the following section I explain the principles that underly Functional

    Discourse Grammar. Then I proceed to explain the components and primitivesthat FDG supposes are involved in processing natural language. Finally I givean example of how FDG can be used to analyze a simple constituent of anutterance in English.

    2.1 Principles

    Functional Discourse Grammar was created as a revision of Functional Gram-mar, to achieve greater pragmatic and psychological accuracy. In this section, Iexplain some distinguishing principles of Functional Discourse Grammar, thatset it apart from Functional Grammar or other linguistic theories.

    2.1.1 Discourse Act

    A first principle is that the basic unit of analysis in FDG is the discourse act.This sets FDG apart from many other theories that choose either the clauseor the sentence as the basic unit of analysis. FDG claims, though, that not alllinguistic utterances are in the form of clauses, and that the sentence is not aconcept that is found in all the worlds languages.

    Discourse acts may be manifested in language as clauses, fully grammat-ical clause fragments, phrases or words. Discourse acts are mapped to those

    9

  • 8/3/2019 Automatization Using FDG

    10/89

    manifestations by a top-down grammar.Discourse acts are defined in [Kroon, 1995] as the smallest identifiable units

    of communicative behavior, and FDG uses this definition. Discourse acts can

    be combined into larger structures, called moves. A move is the minimal freeunit of discourse that is able to enter into an exchange structure. An exchangeconsists of moves, ideally but not necessarily two of them. Larger structures likemoves and exchanges account for the patterns of organization that are largerthan individual clauses that are found in some languages.

    As an example, consider this dialog example from [Kroon, 1995]:

    (1) A Ive got an extra ticket for the Santa Fe Chamber Orchestra tonight.Are you interested?

    B Yes, wonderful.

    In this example, the whole dialog is a single exchange. There are two moves,one corresponding to (1A) and another corresponding to (1B). The former moveis built up from two acts: one corresponding to Ive got an extra ticket for theSanta Fe Chamber Orchestra tonight, and the other corresponding to Areyou interested?. The latter move consists of a single discourse act.

    2.1.2 No Transformations or Filters

    The top-down grammar processes that map discourse acts to language utter-ances may never employ processing steps that alter or remove structures thathave already been built up. Such steps are known in the Functional Grammarand Functional Discourse Grammar literature as transformations and filters

    respectively.The motivation for disallowing such steps is that doing so constrains the

    number of possible hypotheses explaining a linguistic phenomenon, and that itmakes it possible to recover underlying structures from their outward manifes-tations.

    As concrete examples of the restrictions this requirement imposes on lin-guistic explanations, consider the following examples from [Dik, 1989], pages18-19. An example of a linguistic phenomenon that one could analyze with atransformation is:

    (2) A John doesnt like pancakes

    B PANCAKES John doesnt like

    Any analysis of (2B) that derives it from (2A) by moving the constituent pan-cakes through the underlying structure, is a transformation and is must beavoided in FDG.

    And an example of a phenomenon that one could analyze with a filter is:

    (3) A I met a boy who was carrying a green uniform

    B I met a boy carrying a green uniform

    Any analysis of (3B) that derives it from (3A) by deleting a part of its analysiscorresponding to who was, employs a filter and must be avoided in FDG.

    10

  • 8/3/2019 Automatization Using FDG

    11/89

    2.1.3 A Single Theory of Language

    Another principle is that FDG aims to explain all grammatical aspects of an

    utterance in one theory. An FDG analysis of an utterance includes a phono-logical, a morphosyntactic, a representational and an interpersonal analysis.These explain the phonology, morphosyntax, semantics and pragmatics of theutterance respectively.

    One of the reasons for distinguishing those four levels of analysis is thatanaphoric reference is possible to any one of the them. Consider these examplesfrom [Hengeveld and Mackenzie, 2008] page 5:

    (4) A Get out of here!

    B Dont talk to me like that!

    (5) A There are lots of traffic lights in this town.

    B I didnt notice that.(6) A I had chuletas de cordero last night.

    B Is that how you say lamb chops in Spanish?

    (7) A I had /tSu"letasdekor"dero/ last night.

    B Shouldnt that be /tSu"letasdeTor"dero/?

    In (4B) the speaker uses the pronoun that to refer to the communicativestrategy employed in (4A). The communicative strategy is an interpersonal as-pect of the utterance, so this example shows that interpersonal aspects of theutterance are available for reference. FDG claims that interpersonal aspects ofan utterance are available for reference, because they are stored in an Interper-sonal Level of organization for the utterance.

    In (5B) the speaker uses the pronoun that to refer to the situation inthe external world that (5A) describes. Thus, this occurrence of that is apurely semantic reference and shows that there is a Representational Level oforganization.

    In (6B) that refers to the phrase chuletas de cordero from (6A), andnot to its denotation. That phrase is morphosyntactic in nature, so there mustbe a Morphosyntactic Level to refer to.

    In (7B) that refers to the way the phrase chuletas de cordero was pro-nounced in (7A). This kind of reference shows that phonological material is alsoavailable for reference, thus that there is a Phonological Level of organization.

    To get some feeling about what the various levels represent, I will now say ofsome well-known linguistic concepts and data structures, at which level of Func-tional Discourse Grammar they apply. The Phonological Level deals with suchconcepts as intonational phrase and phonological words. The Morphosyn-tactic Level contains the well-known syntactic categories like noun phrase,adverb, concepts like finite or plural, and functions like subject and ob-ject. The Representational Level is about semantic concepts like individual,property and state of affairs, and functions like Actor or Instrument. TheInterpersonal Level, finally, deals with concepts like speech act and partici-pant, and functions like topic and focus.

    11

  • 8/3/2019 Automatization Using FDG

    12/89

    2.1.3.1 Relations Between the Levels

    The four levels of analysis (interpersonal, representational, morphosyntactic and

    phonological) of an utterance need not have the same structure. In other words,there is not a one-to-one mapping between entities of one level to entities ofanother. As already mentioned above, a discourse act (an interpersonal entity)may be manifested at the Morphosyntactic Level as a clause, a clause fragment,a phrase or a word.

    As an illustration of the variability of the relation between different levels,consider the following examples from [Hengeveld and Mackenzie, 2008]. Sen-tences with the same structure at the Representational Level (a predicate ap-plied to two individuals, an actor and an undergoer) have a different mor-phosyntactic structure in English (8) and Southern Tiwa (9). In the gloss,1.sg.sbj>pl.obj means first person singular subject acting on plural object,

    andpst

    means past tense.(8) I made shirts.

    (9) Te-shut-pe-ban1.sg.sbj>pl.obj-shirt-make-pst

    I made (the) shirts.

    While Southern Tiwa expresses with an affix on the verb that there is a firstperson singular Actor, English uses a separate word to express it. And whileSouthern Tiwa incorporates the Undergoer argument of the predicate into theverbal word, in English it becomes another separate word. So in English thismeaning is expressed with a clause of three constituents, in Southern Tiwa the

    meaning is expressed using a clause with only one constituent.

    2.1.4 Maximal Depth Principle

    Functional Discourse Grammar has a principle of Maximal Depth, which is thatonly those levels that are relevant for the production of a certain utterance, areused in the production of that utterance. So if for example an utterance doesnot have any representational content that influences its appearance, it doesnot have a Representational Level.

    That is the case for the utterance Okay (expressing agreement) for ex-ample. Its meaning only relates to previous discourse, and not to the world

    outside the conversation and its direct environment. Therefore it has no Rep-resentational Level.

    2.1.5 Typology-based

    Functional Discourse Grammar is a typology-based theory: it aims to be appli-cable to every natural language in existence. The aim of this thesis, however,is to describe a way to process utterances in English. While it is conceivablethat the program described in this thesis can be applied to a wide variety oflanguages, that was not a goal during its development.

    12

  • 8/3/2019 Automatization Using FDG

    13/89

  • 8/3/2019 Automatization Using FDG

    14/89

    Figure 2.1: General layout of Functional Discourse Grammar. From[Hengeveld and Mackenzie, 2008].

    14

  • 8/3/2019 Automatization Using FDG

    15/89

    formulation stage creates two data structures: the data structure at the Inter-personal Level that represents the pragmatical aspects of the utterance, and thedata structure at the Representational Level that represents the semantics of

    the utterance. The rules of the formulation process are all language-dependent.FDG does not presuppose any universal semantic or pragmatic notions.

    The output of the formulation is fed into the morphosyntactic encodingprocess. It takes the interpersonal and representational data structures createdby the formulation, and creates a data structure at the Morphosyntactic Levelthat represents the morphosyntax of the utterance.

    The data of the Morphosyntactic Level, the Representational Level and theInterpersonal Level are used by the phonological encoding process to createa data structure at the Phonological Level that represents the phonologicalaspects of the utterance. This Phonological Level data structure is then fedinto the output component, which actually expresses the utterance.

    As has already been shown in examples (4-7), all intermediate data struc-tures at the four levels are stored in the contextual component to enable thediscourse participants to refer back to parts of them later on in discourse.

    The structure of the four components, complete with the levels and pro-cesses, is shown in Figure 2.1.

    2.2.2 Use of the Structure for Natural Language Processing

    The structure of Functional Discourse Grammar makes it a good candidate forforming the basis of the natural language understanding faculty of an intelligentkitchen assistant machine. The model of language production already incorpo-rates links with external components. FDGs Conceptual Component mapsonto the kitchen machines belief base, FDGs Contextual Component mapsonto the machines user and environment models, and the output componentmaps onto the actuator control system of the kitchen robot.

    Additionally, one can use the information from the Contextual and theConceptual Component to make predictions about the Representational andInterpersonal Level of the utterance of the user. These predictions can be usedto check and correct the Representational and Interpersonal Level as obtainedfrom the users utterance.

    2.3 Layers

    Now that I have explained what the principles behind FDG are, and whichcomponents it presupposes in the natural language user, I can explain the formof the four levels that the grammatical component creates for an utterance.

    All of the four levels have an organization that is hierarchical in nature,and displayed as a layered structure ([Hengeveld and Mackenzie, 2008], page14). So a level consists of layers, which may contain lower layers, which in turnmay contain yet lower layers, and so on. In its maximal form, the structure ofan FDG layer is:

    (10) ( v1: [head (v1)]: [(v1)X])

    15

  • 8/3/2019 Automatization Using FDG

    16/89

    In this form, stands for a set of operators that restrict the variable v1.v1 stands for an identifier that uniquely identifies the layer and indicates thecategory of what the layer represents [Smit et al., 2008]. The head is another

    layer or lexical item that restricts the variable, and so are the modifiers . andXare functions, that determine in what way the head and respectively restrictthe variable v1. In general, operators and functions stand for grammaticalmeans to restrict the variable, while heads and modifiers are lexical means torestrict the variable.

    The difference between operators and functions is that functions specify therelation between this layer and other layers that are part of the same containinglayer, while operators only apply to the layer itself. An example of a conceptthat is an operator at the Morphosyntactic Level would be plural; an exampleof a function at the Morphosyntactic Level would be subject.

    The difference between the head and the modifiers is that a head is the

    most important restrictor, and that the head must always be present. In ap-parently headless constructions like Spanish la verde or Dutch de groene,both meaning the green one, the head is still present in the context.

    The square brackets ([ and ]) in the form indicate that all the things thatare directly enclosed by them are not in a hierarchical relation to each other.The directly enclosed constituents are called equipollent constituents in FDG.As the form shows, heads and modifiers may consist of a number of equipollentconstituents.

    The following section will describe what kinds of building blocks can beinserted into this general form. In 2.5 I give an example of how layers are builtup from those building blocks according to the general form.

    2.4 Primitives

    The building blocks of FDG layers are the so-called primitives. There aredistinct sets of primitives for every level in the grammatical component.

    The differences between the four levels are motivated by the principle thatno level contains more distinctions than is necessary to explain the propertiesof the utterance at that level; i.e., the Interpersonal Level of an utterancedoes not contain more information than is necessary to explain the pragmaticalproperties of that utterance. Also, the levels never contain more distinctionsthan are relevant to the grammar of a given language.

    I will now discuss the different kinds of primitives. I start with primitivesused in the formulation stage, then discuss primitives used in morphosyntacticencoding, and finish with the primitives for phonological encoding.

    2.4.1 Formulation Primitives

    The kinds of primitives used in formulation to build the Interpersonal Leveland Representational Level are frames, lexemes and operators.

    Frames are blueprints for layers, that determine the ways that other buildingblocks of structures may be combined. What set of frames a language has is

    16

  • 8/3/2019 Automatization Using FDG

    17/89

  • 8/3/2019 Automatization Using FDG

    18/89

    (11) a IL (+id RI)

    b RL (prox m cxi: [(fi: /b@"nA:n@/N(fi))(xi)])

    c ML (Npi: [(Gwi: this-pl(Gwi))(Nwi: /b@"nA:n@/-pl (Nwi))](Npi))d PL (ppi: [(pwi: /Di:z/ (pwi)) (pwj: /b@"nA:n@z/ (pwj))](ppi))

    In the interpersonal analysis (11a) we see that this constituent is analyzedat this level as a single layer, which has only an operator (+id) and an identifier(RI). There are no head, modifiers or function to be shown. The identifier letter(R) tells us that this layer is a referential subact. The interpersonal operator(+id) signifies that the constituent is identifiable in the discourse context. Thesubscript I on the identifier is used to make the identifier unique in the presenceof other referential subacts.

    In (11b) we see that at the Representational Level, the constituent is ana-lyzed as a layer of the entity category (x). This entity layer contains another

    layer of the property category (f). In the head of the property layer, we see thelexeme for the word banana: it is a noun (N) with the phonological represen-tation /b@"nA:n@/. The property layer has no operators, modifiers or function.The entity layer also lacks modifiers, but it has two operators and a function.The operators signify that the layer designates more than one (m) entity interms of the location of its referent (prox). The function on the entity layeris a variable that holds the place of the actual function, which we dont knowwithout context.

    The Morphosyntactic Level analysis in (11c) shows that the Morphosyntac-tic Level of these bananas consists of one noun phrase (Np). The head ofthis noun phrase in turn consists of a grammatical word (Gw) and a nominal

    word (Nw). The operators prox and m from the Representational Level areexpressed as the morphosyntactic operators this and pl. The operator pl isapplied to both the grammatical word and the nominal word because it will beexpressed on both.

    At the Phonological Level (11d), the morphosyntactic operators have beenturned into phonological material. The phonological layers of this utterancehave no operators, modifiers or functions; they have only identifiers and heads.

    2.6 Notation Conventions for FDG Layers

    I use the following conventions when writing FDG layer formulas:

    Sometimes I write down an FDG layer formula that contains constituentsthat are not analyzed. In such cases, I write the orthographical rep-resentation of those constituents between dashes inside the formula,like (Npi: - the garlic - (Npi)). This formula shows that there is announ phrase that has a further analysis of the garlic in its head,but that further analysis is not given. This notation is borrowed from[Hengeveld and Mackenzie, 2008].

    Operators on the surrounding layer that are deduced from the materialin dashes are not given explicitly.

    18

  • 8/3/2019 Automatization Using FDG

    19/89

    I use . . . for pieces of FDG formula forms that are not filled in. So (Np1: . . . (Np1)) stands for a generic noun phrase.

    , X, , and are used as variables over functions of representationallayers.

    (Npk) is the notation for a layer with no operators, head nor modifiers.Npk is the notation for an identifier of a layer with any constitution.

    19

  • 8/3/2019 Automatization Using FDG

    20/89

    Chapter 3

    The Morphosyntactic and

    Representational Levels

    To be able to describe my computer program, I must first clarify the structureof its input and its output. Its input is the Morphosyntactic Level, and itsoutput is the Representational Level. In this section I explain as much of themas is necessary to understand the approach explained in the next chapter.

    The Morphosyntactic Level holds information on the formal aspects of theform of the utterance: it tells what morphemes it is built up from and how thesemorphemes are connected. The Representational Level holds the semantics ofthe utterance: it tells how its meaning relates to the external world, outside ofthe discourse participants and the utterance itself. The Morphosyntactic Leveldeals with concepts like noun, adposition phrase or clause for example, while

    the Representational Level deals with concepts like property, simultaneousor deontic.

    In the following two subsections, I delineate the scope of what the structuresat the Morphosyntactic and Representational Levels describe in more detail. Indoing so, I also present the form of the Morphosyntactic and the Representa-tional Level in more detail.

    3.1 The Morphosyntactic Level

    3.1.1 The Formal Structure of the Morphosyntactic Level

    For any Morphosyntactic Level, the outermost layer is of the category of Lin-guistic Expression (Le). So the general form of a Morphosyntactic Level is:

    (12) ( Le1: [head (Le1)])

    Morphosyntactic layers are usually written without the square bracketsaround the application of the head to the identifier, and without functions.The functions are empty when omitted. So in the usual notation, example (12)becomes example (13):

    (13) ( Le1: head (Le1))

    20

  • 8/3/2019 Automatization Using FDG

    21/89

    As one can see in example (12), morphosyntactic layers do not have mod-ifiers. In the head of a Linguistic Expression, a number of different layer cat-egories may occur. In many cases it will be a clause (Cl), which is a kind of

    layer that is found in every language in the world. But it may consist of one ormore words or one or more phrases in case of a simple Linguistic Expression.

    A clause, in turn, has words and/or phrases in its head, and a phrase hasone or more words in its head. A phrase is a layer that is of a category Xp,where the X is a variable for the type of phrase. All the possible values for theX are N for noun phrases, V for verb phrases, Adj for adjective phrases,Ad for adposition phrases and Adv for adverb phrases.

    A word is a layer that is of a category Xw, where the X is a variable forthe type of word. All the possible values for the X are N for nouns, V forverbs, Adj for adjectives, G for grammatical words, and Adv for adverbs.

    The smallest unit that can occur on its own in a discourse is the Linguistic

    Expression. Any unit that can occur freely in discourse is considered a Linguis-tic Expression. Only when two morphosyntax units can be shown to belongtogether morphosyntactically, they belong together in a single Linguistic Ex-pression.

    We can wrap up these structures of clauses, phrases and words in the form(14):

    (14) ( Le1 : [(Xw1)(Xp1 : [(Xw2)(Xp2)(Cl1)](Xp1))(Cl2 :[(Xw3)(Xp1)(Cl3)](Cl2))](Le1))

    where Xp stands for a phrase of any type and Xw stands for a word of anytype. Every clause and every phrase in this form can be expanded further as

    explained in the paragraphs above. Every clause and phrase or word in theform may occur any number of times, or not at all.

    As some examples of linguistic expressions with a different constitution, lookat the examples in (15-17):

    (15) A In the refrigerator, obviously

    B (Lei : [(Adpi : - in the refrigerator - (Adpi)) (Advpi : - obviously -(Advpi))] (Lei))

    (16) A He is asleep

    B (Lei : (Cli : - he is asleep - (Cli)) (Lei))

    (17) A Not again!B (Lei : (Advpi : [(Advwi : not (Advwi)) (Advwj : again (Advwj))]

    (Advpi)) (Lei))

    In (15B), we see that the Morphosyntactic Level of (15A) comprises twophrases, one adposition phrase and one adverb phrase. In (16B), the Mor-phosyntactic Level of (16A), we see the usual case of a linguistic expressionthat has a single clause. And in (17B) finally, we see a linguistic expressionthat consists of one adverb phrase as the Morphosyntactic Level of (17A).

    21

  • 8/3/2019 Automatization Using FDG

    22/89

    3.1.1.1 The Morphosyntactic Form of Words

    Words can sometimes be decomposed into smaller constituents. An example is

    the word apples which consists of the stem apple and the affix -s.Words that cannot be decomposed further are represented by a layer with

    the phonological material of the word in the head, for example:

    (18) (Vwi : /Iz/ (Vwi))is

    for the verbal word is. I write such layers with the orthographic form of theword instead of the phonological form, if that does not cause any confusion.

    The layers of words that can be decomposed follow the following template:

    (19) (Xw1 : [(Xm)(Xw)(Xp)(Cl)](Xw1))

    That is, such words can have a configuration of morphemes (Xm), words (Xw),phrases (Xp) and/or clauses (Cl) in its head. For English, only morphemes arerelevant.

    There are three different kinds of morphemes: stems, roots and affixes.Stems have layer identifiers of the form Xs1, roots have layer identifiers ofthe form Xr1, and affixes have layer identifiers of the form Aff1. Stems aremorphemes that have lexical content and that can occur as a word on theirown (like apple in the example apples above). Roots are morphemes thathave lexical content and that cannot occur as a word on their own. Affixes aremorphemes that have grammatical content and that need a root or a stem toform a word (like the -s in the example apples).

    3.1.1.2 Examples of Morphosyntactic Layers

    I now illustrate the structure of the Morphosyntactic Level explained abovewith some morphosyntactic analyses of fragments of English. These fragmentsare taken from the kitchen dialog I attempted to parse with my program.

    The first utterance of the dialog that is interesting to us 1 is Im hungry.The Morphosyntactic Level of this utterance is given in example (20):

    (20) (Lei : (Cli : [(Npi : (Nwi : i (Nwi)) (Npi))Subj (fin Vwi : am (Vwi))(Adjpi : (Adjwi : hungry (Adjwi)) (Adjpi)) ] (Cli)) (Lei))Im hungry.

    In this example, we see that the Morphosyntactic Level of this utteranceconsists of a single Linguistic Expression, which consists of a single clause.The clause has three constituents: a noun phrase Npi, a verbal word Vwi andan adjective phrase Adjpi. The first two constituents will be merged at thePhonological Level to form Im, but at the Morphosyntactic Level they aretwo separate constituents. The verbal word am is not part of a verb phrase.Verbal words are only grouped in verb phrases when there is a lexical verb

    1Utterances are only interesting to us if they have a Representational Level. Greetings likeHello! can be argued to have only Interpersonal and Phonological Levels.

    22

  • 8/3/2019 Automatization Using FDG

    23/89

    heading a phrase, as stated in 5.1.1.1. The noun phrase Npi and the adjectivephrase Adjpi both have a lexical word as their only constituent.

    As another example, consider the Morphosyntactic Level of Can you do

    something with those ingredients?, given in (21). The example is renderedwith indentation for easy reading:

    (21) (Lei : (Cli :[ (fin Vwi : can (Vwi))

    (Npi : (Nwi : you (Nwi)) (Npi))Subj(Vpi : (Vwj : do (Vwj)) (Vpi))(Npj : (Nwj : something (Nwj)) (Npj))Obj(Adpi :

    [ (Gwi : with (Gwi))(Npk :

    [ (Gwj : those (Gwj))(Nwk :

    [ (Nsi : ingredient (Nsi))(Affi : s (Affi))

    ] (Nwk))] (Npk))

    ] (Adpi))] (Cli)) (Lei))

    Can you do something with those ingredients?

    In this example we see that this linguistic expression has one clause in itshead. That clause in turn has five constituents in an equipollent configuration

    in its head. Those are a verb word Vwi, a noun phrase Npi, a verb phrase Vpi,a noun phrase Npj and an adposition phrase Adpi.

    A first thing that one can see in this example, is that according to FDG,a verb phrase does not contain the morphosyntactic realizations of the repre-sentational arguments of the verb. So, in this example, the morphosyntacticconstituents corresponding to something and with those ingredients are notpart of the verb phrase Vpi. A verb phrase in FDG is simply a group of verbsand grammatical particles headed by a verb.

    The verbal word Vwi has an operator fin, which indicates that it is a finiteverb.

    One will also notice that the word ingredients is decomposed further into

    the stem ingredient (Nsi) and the affix -s (Affi). This structure is repre-sented in the formalism by giving the word a head with an equipollent config-uration of the stem and the affix.

    A final important thing to notice is that morphosyntactic functions havebeen assigned to some of the noun phrases. Npi (you) has been assignedthe function Subj (for Subject) and Npj) has been assigned the functionObj (for Object). These functions, together with the chosen representationalframe, can help us determine the representational function of arguments of theverb when building up the Representational Level from the MorphosyntacticLevel.

    23

  • 8/3/2019 Automatization Using FDG

    24/89

    3.1.2 Positioning and Alignment

    Two important processes that happen during Morphosyntactic Expression (the

    process that builds up the Morphosyntactic Level from the Representationaland Interpersonal Levels) are positioning and alignment. Positioning deter-mines the position that hierarchically related units will have, relative to eachother, in the resulting utterance. Alignment determines the relative order ofnon-hierarchically related units. Positioning proceeds in a top down fashion,so the highest layers of the Representational and Interpersonal Level are posi-tioned first. Alignment happens after positioning is complete, but the structurescreated by alignment and positioning may interlock.

    While it is beyond the scope of this thesis to discuss and motivate all theprinciples of positioning and alignment in general and all the rules that governtheir implementation in English, I will give a short overview of both to show

    the way they shape English utterances.

    3.1.2.1 Positioning

    As already mentioned, positioning determines the position that hierarchicallyrelated units will have in the utterance. What the positioning process leadsto, is that the lower in the hierarchy of layers a piece of interpersonal or rep-resentational information is, the closer to the center of the utterance it will berealized. Interpersonal constituents tend to be realized further from the centerthan representational constituents.

    As an example, consider example (22) (taken from[Hengeveld and Mackenzie, 2008]):

    (22) Finally, she honestly reportedly probably has been drinking continu-ously again recently

    The example is contrived but grammatical. And it shows that the interpersonalinformation (Finally, honestly and reportedly) goes to the beginning ofthe utterance. The interpersonal speech act modifier Finally is even posi-tioned outside the clause. The representational information (probably, con-tinuously, again and recently) stays within the clause and occurs later inthe utterance than the interpersonal information. And at the end of the ut-terance, the representational modifiers are expressed in hierarchical order, withthe hierarchically lowest modifiers first. The hierarchically highest representa-

    tional modifier, probably, goes to the beginning of the utterance, where itoccurs later than the interpersonal modifiers.

    3.1.2.2 Alignment

    Alignment is the way in which non-hierarchically related pragmatic and seman-tic units map onto morphosyntactic ones ([Hengeveld and Mackenzie, 2008],page 316). It is during alignment that morphosyntactic functions such as Sub-ject and Object are assigned to constituents.

    FDG discerns three types of alignment: interpersonal alignment, represen-tational alignment and morphosyntactic alignment. These types of alignment

    24

  • 8/3/2019 Automatization Using FDG

    25/89

    are named after the level from which functions and operators are used to alignconstituents.

    English uses the morphosyntactic functions Subject and Object for mor-

    phosyntactic alignment. The contextual component plays in important role indeciding which constituent gets which morphosyntactic function; that cannotbe determined from the Representational and Interpersonal Levels alone.

    Morphosyntactically complex constituents are often aligned to the end ofan utterance. So we have the singing man versus the man who sings:though both singing and who sings are modifiers of an individual at theRepresentational Level, and they mean the same, who sings is realized at theend of the phrase because it is more complex than singing.

    3.2 The Representational Level

    While the Morphosyntactic Level describes the hierarchical form of the utter-ance, the Representational Level describes the relation between the utteranceand the world outside the discourse and its participants. Or in FDG terminol-ogy: the Representational Level describes the designation of the utterance.

    It is important to realize what the Representational Level is not about. It isnot about the meaning of an utterance when it relates to the utterance itself (likethe meaning of the word briefly). Nor is it about the meaning that linguisticunits have in a particular context: the Representational Level is restricted tothe meanings of lexical units (. . . ) and complex units (. . . ) in isolation fromthe ways these are used in communication ([Hengeveld and Mackenzie, 2008],page 129).

    The latter point has important consequences for the notion of reference.When we consider the sentence I saw a lion, the expression a lion be con-sidered a referring expression in two distinct ways: firstly, the speaker refersto an animal of the lion-class by using this expression; and secondly, the ex-pression itself refers to an animal of the lion-class. The first kind of referringfrom this example, in which the referring is done by the speaker, is handledby FDG at the Interpersonal Level. The second kind of referring is what theRepresentational Level is about, and it is called designation in FDG.

    3.2.1 Semantic Categories

    Representational layers are of the general form for layers that has been givenin the previous chapter in example (10).

    The layer identifiers of representational layers indicate the semantic categoryof the layer. A semantic category is an ontological category that is reflectedby the system of a language. Distinctions that are made solely by differentlexemes, expression of operators or expression of functions are not considereddistinctions of semantic categories. Only distributional criteria are valid as evi-dence to consider two layers as belonging to a different semantic category. Thatis, two layers have different semantic categories if and only if their morphosyn-tactic realizations behave differently with regard to in what semantically basedmorphosyntactic configurations they are allowed.

    25

  • 8/3/2019 Automatization Using FDG

    26/89

    Because different languages allow different sorts of configurations, it followsfrom this definition of semantic category that different languages have differentsets of semantic categories.

    There are four semantic categories that are found in every language: propo-sitional contents (layer identifier of the form p1), states of affairs (layer identifiere1), individuals (layer identifier x1) and properties (layer identifier f1). Englishhas some more categories: episodes (layer identifier ep1), locations (layer iden-tifier l1), times (layer identifier t1), manners (layer identifier m1) and quantities(layer identifier q1). Possibly, there is also a separate semantic category forreasons, with the layer identifier r1.

    The outermost layer of the Representational Level of a declarative or aninterrogative utterance is always a propositional contents layer. The outermostlayer of the Representational Level of imperative utterances is a state of affairslayer.

    3.2.2 Propositional Contents

    Propositional content layers typically have the following form:

    (23) (p1 p1 :[ (ep1 ep1 :

    [[ (e1 e1 : . . . : . . . )

    . . .] (ep1)

    ] : [ ep1(ep1)X ])

    (p1)] : [ p1 (p1) ])

    3.2.2.1 The head of the propositional content

    The most typical propositional content head is that which contains a singleepisode, as displayed in (23). The episode typically contains one or more statesof affairs in its head. If there are multiple states of affairs in the head of anepisode, those states of affairs belong together because of unity or continuity ofthe time at which they take place, the location where they take place, or theindividuals involved.

    In English, the episode layer is important for temporal operators and mod-ifiers. In English, the operators ep1 and modifiers ep1 of the episode specifythe moment in time at which the states of affairs take place absolutely, or rela-tive to the moment of speaking. Further temporal operators and modifiers onthe states of affairs may indicate the time of those states of affairs relative tothe time of the whole episode.

    3.2.2.2 Other propositional content head structures

    Though most propositional content layers have follow the structure given inexample (23), there are propositional content layers that do not. Propositional

    26

  • 8/3/2019 Automatization Using FDG

    27/89

    content layers may have lexical heads, as in the Morphosyntactic Level con-stituent that corresponds with the noun phrase a belief:

    (24) (pi : [ (f i : [ belief (fi) ]) (pi) ])a belief

    If the word yes or no is used as the full answer to a yes/no-question, theRepresentational Level of the utterance will be as in (25) or (26) respectively:

    (25) (pi : [ yes (pi) ])Yes.

    (26) (pi : [ no (pi) ])No.

    So in these two cases the propositional content layer has a lexical constantdirectly in its head. The lexical constant is not wrapped in a property layer filike in (24).

    3.2.2.3 Propositional content operators

    Propositional content layers, like all layers, have a set of operators, called p1in the formula (23) above. These operators specify the subjective epistemicmodality and the evidential modality of the propositional content. The subjec-tive epistemic modality operators include operators expressing belief, doubt andhypothesis. The evidential modality operators can indicate that the proposi-tional content is based on perception, that it is based on deduction from existing

    knowledge, or that it is based on common knowledge.

    3.2.2.4 Propositional content modifiers

    The modifiers of the propositional content, p1 in formula (23) above, spec-ify the propositional attitude that the speaker has toward the content. Thepropositional attitude is the kind and degree of commitment that a speakerhas toward the content. Words like probably, undoubtedly or hopefullyexpress propositional attitudes. They are analyzed as properties that modifypropositional contents at the Representational Level.

    So the sentence Evidently he is working is analyzed at the Representa-tional Level as:

    (27) (pi : [ (pres epi :[ [ (sim ei : [ - he is working - (e i)]) ]

    (epi)]) (pi)

    ] :[ (fi : [ evidently (fi) ]) (pi) ])

    Evidently, he is working.

    In this example, the propositional content has a modifier that expresses themeaning of the word evidently. This modifier is on the last line of the formula.

    27

  • 8/3/2019 Automatization Using FDG

    28/89

    In this modifier, a property fi, the property of being evident, is applied to thelayer identifier pi.

    The propositional content has in its head an episode epi that expresses the

    state of affairs of he is working. The operator pres on the episode specifiesthat this episode happens at the time of speaking. The operator sim on thestate of affairs ei specifies that that state of affairs happens simultaneously withthe episode.

    3.2.3 States of Affairs

    States of affairs are entities that can be situated in relative time, and that canbe said to be true or false. And as we have seen above, they can be groupedto form episodes. In the sentence He mounted his horse and rode toward thesunset for example, there are two states of affairs, one corresponding to He

    mounted his horse and one corresponding to rode toward the sunset. Thesetwo states of affairs are joined in the head of one episode.The typical structure of a state of affairs layer is given in ( 28). This structure

    will be explained in the following paragraphs:

    (28) (ei ei :[ (fi fi :

    [ [ (fj fj : . . . )(xi xi : . . . ). . .

    ] (fi)X] : [ fi(fi) ]) (ei)

    ] : [ ei(ei) ])

    3.2.3.1 Heads of States of Affairs

    In this typical structure, we find a property fi in the head of the state of affairs.This property fi has a configurational head itself

    2. In the head of fi we find aproperty fj followed by any number of participants in the state of affairs. Herean individual xi has been listed as a participant. Every participant has a certainrole in the state of affairs, which is coded into the formalism as the function of the participant.

    The property fj is the predicate of the state of affairs. This property is

    predicated over the participants of the state of affairs. In other words, fj specifiesthe relation that holds between the participants.I will not go deeper into the role of the operators in fi and fj and the mod-

    ifiers in fi at this point. What operators can be found in xi will be explainedin 3.2.5 on individuals. I explain more about properties with configurationalheads, like fi in (28), in 3.2.4.2.

    2The notation that I use for configurational properties is slightly different from the onethat [Hengeveld and Mackenzie, 2008] uses. In [Hengeveld and Mackenzie, 2008], the outerpair of square brackets around the head restrictor (the pair that is opened by the secondsquare bracket in 28) is omitted for configurational properties.

    28

  • 8/3/2019 Automatization Using FDG

    29/89

    3.2.3.2 Operators of states of affairs

    The set of operators on the state of affairs itself, ei, may specify location,

    perception, and negation. They may also specify the relative tense of the stateof affairs, relative to the time of the episode.

    3.2.3.3 Modifiers of states of affairs

    There is a wide range of things that modifiers on states of affairs can specify.Among them are all the things that operators may specify, except negation.Furthermore, modifiers may specify frequency, reality (as with the word re-ally), cause and goal.

    3.2.3.4 An example of a typical state of affairs

    As an example of a state of affairs layer, I will discuss the Representational Levelof the sentence Sheila may eat a bagel tomorrow. The Representational Levelof it is given in (29):

    (29) (pi :[ (pres epi :

    [ [ (poss ei :[ (fi :

    [ [ (f j : [ eatV (fj) ])(xi)Actor(1 cxj : [ (fk : [ bagelN (fk) ]) (xj) ])Undergoer

    ] (fi)

    ]) (ei)] :[ (ti : [ (fl : [ tomorrowAdv (fl) ]) (ti)]) (ei) ])

    ] (epi)]) (pi)

    ])Sheila may eat a bagel tomorrow.

    The state of affairs layer ei in the example illustrates every part of the typicalstate of affairs layer as discussed above. Firstly, it has an operator poss thatsignifies that the meaning of this state of affairs is possibly, but not necessarily,true. This operator is realized at the Morphosyntactic Level as the verb wordmay.

    Secondly, this state of affairs layer has a property fi in its head. In thisexample, the property fi does not have any operators or modifiers of its own.The property fi in turn has a configurational head which contains a predicatefj and its arguments xi and xj. These arguments have functions (Actor andUndergoer) that indicate what their role in the state of affairs is. xi, whichstands for Sheila, is Actor, and xj, which stands for a bagel, is Undergoer.A list of possible roles is given in 3.2.4 on properties.

    The meaning of the proper name Sheila is analyzed as a completely inter-personal matter. Thus the individual xi for Sheila does not have any further

    29

  • 8/3/2019 Automatization Using FDG

    30/89

    information about its meaning at the Representational Level. Layers for indi-viduals will be explained further in 3.2.5 on individuals.

    Finally, the state of affairs has a relative tense modifier t i, which stands for

    the phrase tomorrow. It specifies the time of the state of affairs, relative to thetime of the containing episode, which in this case has the absolute tense operatorpres, for present. The operator pres is realized at the MorphosyntacticLevel as the present tense of the verb form may.

    3.2.3.5 Other Layer Structures for States of Affairs

    Besides the typical structure given above, state of affairs layers with lexicalheads may also occur. An example is given in (30):

    (30) (ei : [ (f i : [ meetingN (fi) ]) (ei) ])a meeting

    So there is a state of affairs layer ei, which has in its head a property layer fi,which in turn has in its head the lexical constant meetingN. The subscript Non the lexical constant signifies that it is a nominal lexeme.

    3.2.4 Properties

    Property layers are a fundamental building block of FDG semantics. Non-configurational property layers (layers where the head does not contain anyother layers) form the set of lexemes of a language. And configurational prop-erty layers (layers of which the head does contain other layers) are found in the

    heads of most states of affairs.

    3.2.4.1 Lexical Properties

    The form of a lexical property is given in (31) 3.

    (31) ( f1 : [ (f1) ] : [ (f1) ])

    and , as usual, stand for sets of operators and modifiers respectively. The in the example stands for a lexical constant. FDG does not further analyze thenature of these lexical constants; in this thesis I regard them as atomic bits ofsemantics stored in the lexicon.

    More complex combinations of lexical constants are described in[Hengeveld and Mackenzie, 2008], but they are beyond the scope of this the-sis.

    Every lexical constant is member of a certain lexeme class. A lexeme class islike a part-of-speech, except that it is a semantic instead of a morphosyntacticnotion. The word worker, for example, is of the part-of-speech noun, but it

    3There is some variation in the notation of lexical properties:[Hengeveld and Mackenzie, 2008] use ( f1 : (f1) : [ (f1) ]) while[Smit and Van Staden, 2007] use ( f1 : [($1|) (f1) ] : [ (f1) ]). I see no com-pelling reason to use a different notation for heads with lexical constants than for heads withlayers.

    30

  • 8/3/2019 Automatization Using FDG

    31/89

    contains the lexeme work- of the lexeme class verb. English has four lexemeclasses: nouns, verbs, adjectives and adverbs.

    In representational formulas, lexical constants are written as the most typi-

    cal orthographic rendition of the constant, with a lexeme class symbol added insubscript. The lexeme class symbols are N for nouns, V for verbs, Adj foradjectives and Adv for adverbs. Following this convention, the verbal lexemein worker would be written workV. Some more examples of the notationcan be found in example (29) above.

    Modifiers of Lexical Properties Like other layers, lexical properties mayhave modifiers. One kind of modifiers of lexical properties stems from a certainkind of adjectives.

    Some adjectives ascribe a property to an individual, while others changethe meaning of the head noun of the noun phrase they are in. As examples,

    consider (32):

    (32) A A rich doctor

    B A former doctor

    By using the utterance (32A), one evokes an individual who is a doctor andwho is rich. By using the utterance (32B) however, one evokes an individual whois not a doctor, but who was formerly a doctor. So in (32A), the adjective richascribes a property to an individual, while in (32B), the adjective formerchanges the way the property of doctor applies to the individual.

    In FDG, the adjective rich as in (32A) is analyzed as a modifier of anindividual at the Representational Level, while the adjective former as in(32B) is analyzed as a modifier of the lexical property that represents the worddoctor at the Representational Level. So the Representational Levels of (32A)and (32B) are given in (33A) and (33B) respectively:

    (33) A (1 cxi : [ (fi : [ doctorN (fi) ]) (xi) ]: [ (fj : [ richAdj (fj) ]) (xi) ])

    a rich doctor

    B (1 cxi : [ (fi : [ doctorN (fi) ]: [ (fj : [ formerAdj (fj) ]) (fi) ])

    (xi) ])

    a former doctor

    Another kind of lexical property modifiers are directional modifiers on ver-bal lexemes. These appear in English as the verbs go home, come down,etcetera. In such verbs, the directional particle is analyzed as a location layerthat modifies the lexical property of the verb. So come down is analyzed as:

    (34) (fi : [ comeV (fi) ]: [ (li : [ (fj : [ down (fj) ]) (li) ]) (fi) ])

    come down

    31

  • 8/3/2019 Automatization Using FDG

    32/89

    Operators on Lexical Properties Lexical properties may also have oper-ators. The only kind of lexical property operator that is relevant to this thesisis narrow-scope negation.

    In a sentence like She is a not unintelligent girl (taken from[Hengeveld and Mackenzie, 2008]), not is analyzed as an operator on the lex-ical property of unintelligent that negates its meaning. So not unintelligentas in this sentence is analyzed as:

    (35) (neg fi : [ unintelligentAdj (fi) ])not unintelligent

    3.2.4.2 Configurational Properties

    The properties that are not lexical properties are known as configurational prop-erties. In the head of a configurational property layer, there is a configurationof one or more other layers. So the general form of a configurational propertylayer is:

    (36) ( f1 : [ [ (v1) . . . (vn) ] (f1)X ] : [ (f1) ])

    In this form, as usual, and are the operators and modifiers of the configura-tional property, respectively. , X and are variables over semantic functions.(v1) . . . (vn) is a sequence of layers of any semantic category. The semanticcategories of the different (vi)s need not be the same. Typically, (v1) will nothave a function, but the other (vi)s will have one. In that case (v1) is calledthe nucleus and the other (vi)s are called the dependents.

    An important use for configurational properties is in states of affairs. As hasalready been mentioned, the typical head contains a configurational propertythat describes what is the case.

    Heads of Configurational Properties In most cases, the head is a con-figuration of a nucleus and zero or more dependents, as stated above. In thetypical case of a head with nucleus and dependents, the nucleus is anotherproperty layer, and the dependents are individuals, locations, times or states ofaffairs. In such cases, the nucleus is a relation and the dependents are the thingsbetween which this relation holds. The semantic functions of the dependentsspecify their role in the relation. This part of the Representational Level of theutterance The man sees the woman is an example:

    (37) (fi : [ [ (f j : [ seeV (fj) ])(xi : [ - the man - (xi) ])Actor(xj : [ - the woman - (xj) ])Undergoer

    ] (fi) ])

    the man sees the woman

    This specifies that there is a relation of seeing between an individual x i and anindividual xj, and that the individual xi does the seeing and that the individualxj undergoes the seeing.

    32

  • 8/3/2019 Automatization Using FDG

    33/89

    As an example of a configurational property with no dependents in its head,consider this part of the Representational Level of it is raining:

    (38) (fi : [ [ (f j : [ rainV (fj) ]) ] (fi) ])it is raining

    Semantically, there are no arguments to the verb rain. According to FDG,the word it in the sentence it is raining is inserted during morphosyntacticencoding because all available morphosyntactic frames require a subject.

    Several other layouts of configurational properties are possible besides theusual case with a nucleus and zero or more dependents. Those are beyond thescope of this thesis however.

    Operators of Configurational Properties Operators on configurational

    properties may express aspect, participant-oriented modality and quantificationof duration.

    Operators for Aspect The aspect of a state of affairs reflects its inter-nal temporal constituency ([Hengeveld and Mackenzie, 2008], page 210). Theaspect of a state of affairs appears in FDG as an operator on the configura-tional property layer in the head of the state of affairs at the RepresentationalLevel. Such operators than stand for such things as perfective, imperfectiveor ingressive.

    Operators for Participant-oriented Modality The participant-

    oriented modality of a state of affairs tells in what way a participant partic-ipates in the state of affairs. The three possible participant-oriented modalitiesare facultative (be able to), deontic (have to) and volitive (want to).

    Operators for Quantification of Duration Operators on a configura-tional property may also quantify the duration of the situation described bythe property.

    Modifiers of Configurational Properties Modifiers of a configurationalproperty may add extra participants to the state of affairs that the propertyis the head of. Such extra participants may have the functions Beneficiary,

    Comitative, Instrument or Duration. For an example, see the state of af-fairs layer in (39) (adapted from [Hengeveld and Mackenzie, 2008], example435, page 208):

    33

  • 8/3/2019 Automatization Using FDG

    34/89

    (39) (sim ei: [ (f i: [ [ (f j: [ cutV (fj) ])(xi)Actor(xj: [ - the meat - (xj) ])Undergoer

    ] (fi)]:[(xk: [ - a knife - (xk) ])Instrument (fi)])(ei)

    ])John cut the meat with a knife

    3.2.5 Individuals

    Individuals are representational layers that designate concrete, tangible objectsthat occupy a portion of space. Many languages distinguish countable and

    uncountable individuals. English is one such language.The general form of layers for individuals is given in (40):

    (40) ( Sx1 : [ head (x1) ] : [ (x1)X ])

    In this formula, the as usual stands for operators on the layer. The Sstands for a letter that indicates whether this layer is count, mass, or neither.If the layer is count, the value ofS is c; if the layer is mass, the value ofS ism; and if the layer is neither, the value ofS is empty.

    The x1, as usual, stands for an identifier for the layer.The head position of an individual can be occupied by many different struc-

    tures, as will be explained below. The stands for one or more modifiers on

    the layer. The and X are variables over functions.

    3.2.5.1 Heads of Individuals

    Lexical heads The head position of an individual is most typically occupiedby a lexical head. A phrase like The chair is analyzed as in ( 41):

    (41) (1 cxi : [ (fi : [ chairN (fi) ]) (xi) ])The chair

    In this example, the head position in the individual layer x i is taken by thelexical property layer fi. The fact that there is exactly one chair that is talkedabout, is reflected by the operator 1. The fact that the chair being talked

    about is identifiable (as signaled by the definite article the), is not reflectedat the Representational Level, as it is an interpersonal matter.

    Configurational heads Configurational heads for individual layers are alsofound in English, but not as often as in other languages. In the case of aconfigurational head for an individual layer, the head position of the layer isoccupied by a configurational property layer or a state of affairs layer.

    One kind of construction that gives rise to configurational heads for individ-uals, is the expression of inalienable possession. An utterance like the brotherof the king is analyzed at the Representational Level as in (42):

    34

  • 8/3/2019 Automatization Using FDG

    35/89

    (42) (1 cxi : [ (fi :[ [ (fj : [ brotherN (fj) ])

    (1 cxj : [ (fk : [ kingN (fk) ]) (xj) ])Ref] (fi)

    ]) (xi)])

    The brother of the king

    In this Representational Level we see that there is a lexical property fj thatis the property of being a brother. This property is inside the configurationalhead of property fi as the nucleus. Inside the configurational head of fi thereis also the individual xj as a dependent with the function of Referent. So theconfigurational property fi is the property of being a brother of the king.

    A construction that gives rise to individual layers with a state of affairs

    as their head is the headless relative clause. Such clauses, like what youread in I will read what you read, are analyzed as in (43) (taken from[Hengeveld and Mackenzie, 2008], example 625, page 241) 4.

    (43) (xi : (ei: [ (fi: [ [(fj: [ readV (fj) ])(xj)Actor(xi)Undergoer

    ] (fi)]) (ei)

    ]))what you read

    In this example we see an individual xi headed by a state of affairs xi. Thehead of the state of affairs is a configurational property fi, which says that theyou (xj) is reading (fj) the individual xi. In this way, xi is denoted by theway it participates in the state of affairs e i.

    3.2.5.2 Operators and Modifiers of Individuals

    Operators on individuals may specify location, for example as an analysis ofdemonstratives. Also they may specify quantification: all as in all childrenmust go to school is analyzed as an operator on an individual layer at the Rep-resentational Level. Finally, operators on individuals may specify qualification,most notably with diminutive forms of nouns.

    Individuals may be modified by properties and by states of affairs or episodesin which the individual itself occurs as a participant. Most adjectives on nounphrases are analyzed as modifiers of individuals at the Representational Level.So the Representational Level of an old man becomes:

    (44) (1 cxi: [ (fi: [ manN (fi) ]) (xi) ]: [(fj: [ oldAdj (fj) ]) (xi) ])an old man

    4There is a notation inconsistency in this example: there are no square brackets aroundthe head of xi. Apparently they are omitted because the layer identifier xi already occurs inthe state of affairs in the head. As far as I know there is no rigid specification for this way ofwriting down such heads in FDG: it is not treated in [Smit and Van Staden, 2007].

    35

  • 8/3/2019 Automatization Using FDG

    36/89

    3.2.6 Coindexation

    A special way to use the identifier of a representational layer is coindexation.

    In the case of coindexation, two layers have the same identifier. This is used toindicate that those two layers have the same designation. Layers that expressanaphoric reference have only an identifier, which is the same identifier as theone of the layer that the anaphor refers to.

    So when one considers example (45) (adapted from[Hengeveld and Mackenzie, 2008], page 143), the layer at the Representa-tional Level for the man is given as example (46) and the RepresentationalLevel for he is given in example (47). The fact that the layers in (46) and in(47) have the same identifier, indicates that they have the same designation.

    (45) The man cleaned the windows and he painted the door

    (46) (1

    c

    xi: [ (fi: [ manN (fi) ]) (xi) ])(47) (xi)

    3.2.7 Relation to the Morphosyntactic Level

    Now that I have discussed both the form of the Morphosyntactic Level and theform of the Representational Level, I shall briefly address the way the two formsare related.

    During language production, the Morphosyntactic Level of the utterance iscreated by the process of morphosyntactic encoding. The inputs to the mor-phosyntactic encoding process are the Interpersonal Level, the RepresentationalLevel and context data from the contextual component.

    3.2.7.1 Principles for Morphosyntactic Encoding

    The relation between the structure between any two levels that influence eachother is characterized by the three principles of Iconicity, Domain Integrity andFunctional Stability. The relation between the Morphosyntactic and the Rep-resentational Level is no different. Iconicity, Domain Integrity and FunctionalStability are the three principles that guide morphosyntactic encoding in anylanguage.

    Iconicity means that the order of moves and discourse acts is maintained.Domain Integrity means that what belongs together on higher levels (i.e.

    the Representational Level), will stay together on lower levels (i.e. the Mor-phosyntactic Level).Functional Stability means that constituents that have the same function,

    will go to the same position.None of these three principles is inviolable, and there are many forces that

    may overrule them in certain circumstances. The extent, to which these prin-ciples are enforced in the presence of others forces, differs across languages.

    36

  • 8/3/2019 Automatization Using FDG

    37/89

    3.2.7.2 Ambiguities on the Morphosyntactic and Representational

    Levels

    An important challenge in natural language processing is resolving ambiguities.A single utterance, when taken in isolation, may mean many different things.If one wants to compute the meaning of the utterance, one will have to pick theright meaning based on the available data about the context of the utterance.

    The existing material on Functional Discourse Grammar does not treatambiguity explicitly. However, the descriptions of the various levels of analysisallow us to check if a certain ambiguity, that is found in the form of an utterance,is also present on a certain level of analysis.

    As an example, consider the utterance My neighbor killed the butchertoday. He will be buried tomorrow. The part He will be buried tomorrowis ambiguous: when one views it in isolation, one cannot decide who is meant

    by the word he. Only when one takes into account the information suppliedin the first sentence (My neighbor killed the butcher today), it becomes clearthat the he in the second sentence is the the butcher from the first sentence.

    When we construe the Morphosyntactic Level of the utterance in (48), wefind that the ambiguity is found there too. The Morphosyntactic Level doesnot tell us who is meant by the he in the second sentence.

    (48) (Lei: (Cli: [ (Npi: [ (Gwi: my (Gwi)) (Nwi: neighbor (Nwi)) ] (Npi))(Vpi: (fin Vwi: [ (Vsi: kill (Vsi))

    (Affi: -ed (Affi))] (Vwi)) (Vpi))

    (Npj: [ (Gwj: the (Gwj)) (Nwj: butcher (Nwj)) ] (Npj))

    (Advpi: (Advwi: today (Advwi)) (Advpi))] (Cli)) (Lei))

    (Lej: (Clj: [ (Npk: (Nwk: he (Nwk)) (Npk))(Vpj: [ (fin Vwj: will (Vwj))

    (Vwk: be (Vwk))(Vwl: [ (Vsj: bury (Vsj))

    (Affj : -ed (Affj))] (Vwl))

    ] (Vpj))(Avdpj: (Advwj: tomorrow (Advwj)) (Advpj))

    ] (Clj)) (Lej))

    My neighbor killed the butcher today. He will be buried tomorrow.

    In this Morphosyntactic Level one can see that the layer that representshe, Npk, does not contain any information on what the antecedent of he is.

    In the Representational Level in (49) however, one can see that the an-tecedent of he is indicated by giving the layer representing he the sameindex as the antecedent. So there is no ambiguity at the RepresentationalLevel: he can only refer to the butcher.

    37

  • 8/3/2019 Automatization Using FDG

    38/89

    (49) (pi: [(past epi: [ [(sim ei : [(fi: [ [(fj: [ killV (fj)])(1 cxi: [(fk: [ [(fl: [ neighborN (fl) ]) (xj)Ref] (fk) ]) (xi)(1 cxk: [ (f m : [ butcherN (fm) ]) (xk) ])Undergoer

    ] (fi)]) (ei) ]

    : [(ti: [(fn: [ todayAdv (fn) ]) (ti) ]) (ei) ])] (epi) ]) (pi) ])

    (pj: [(fut epj: [ [(sim ej : [(fo: [ [(fp: [ buryV (fp) ])(xk)Undergoer

    ] (fo)]) (ej) ]

    : [(tj: [(fq: [ tomorrowAdv (fq) ]) (tj) ]) (ej) ])] (epj) ]) (pj) ])

    My neighbor killed the butcher today. He will be buried tomorrow.

    This example shows that at least sometimes, ambiguities that are presentin the Morphosyntactic Level are not present at the Representational Level. Soif one wants to compute the Representational Level from the MorphosyntacticLevel, one will have to resolve those ambiguities using context information, orby making an educated guess.

    There are some other types of ambiguities that one has to resolve whencomputing the Representational Level from the Morphosyntactic Level. Mostimportantly, because the Morphosyntactic Level contains the phonological rep-resentation of words, one cannot distinguish homophones at the Morphosyn-tactic Level. Whether the bank in See you at the bank! is a financialinstitution or a riverside will have to be decided upon when computing the

    Representational Level.Some other types of ambiguities, that are found in the surface form of ut-

    terances, are already resolved at the Morphosyntactic Level. The well-knownPP-attachment ambiguity is one such type of ambiguity. The sentence Theman sees a girl with a telescope is ambiguous, but the two different interpre-tations have different Morphosyntactic Levels. The Morphosyntactic Level ofthe interpretation where the man uses a telescope as an instrument to see thegirl is given in (50), and the Morphosyntactic Level of the interpretation wherethe man sees a girl who carries a telescope is given in (51).

    (50) (Lei: (Cli: [(Npi: - the man - (Npi))(Vpi: - sees - (Vpi))(Npj: - a girl - (Npj))(Adpi: - with a telescope - (Adpi))

    ] (Cli)) (Lei))The man sees a girl with a telescope

    (51) (Lei: (Cli: [(Npi: - the man - (Npi))(Vpi: - sees - (Vpi))(Npj: - a girl with a telescope - (Npj))

    ] (Cli)) (Lei))The man sees a girl with a telescope

    38

  • 8/3/2019 Automatization Using FDG

    39/89

    Chapter 4

    Language-Independent

    Aspects

    Now that I have described the concepts of Functional Discourse Grammar, Ican proceed to explain the approach to compute the Representational Levelfrom the Morphosyntactic Level.

    The approach is presented in three steps. First, I explain the theoreticalbackground of the approach (4.1). Then, I explain the general structure of theprogram (4.2 through 4.5). In 7.1.3, I briefly address the question of howthis approach can be used to build up the Interpersonal Level instead of theRepresentational Level.

    Finally, in a separate chapter, I describe how the program has been used tomodel a fragment of the English language (chapter 5).

    4.1 Theoretical Background

    We can imagine the outline of a procedure for interpreting natural languageaccording to FDG, by inverting the process of language generation that is de-scribed in most FDG and Functional Grammar literature. This way of produc-ing a theory of language understanding is already hinted at by [Dik, 1989] and[Hengeveld and Mackenzie, 2008].

    The procedure, that one obtains by inverting the generation description inFDG, goes like this: when a natural language user hears an utterance, the firstthing that happens is that an inverse of the Output Component turns the per-

    ceived acoustic signal into a Phonological Level of the utterance. Based on thisPhonological Level, the hearer builds up a Morphosyntactic Level of the ut-terance. Based on those Phonological and Morphosyntactic Levels, the hearerbuilds up a Representational Level. The hearer then builds up the Interper-sonal Level based on the Representational, Morphosyntactic and PhonologicalLevels and then derives the communicative intention behind the perceived ut-terance from the Interpersonal and Representational Levels. While building upthe Levels in reverse order this way, the hearer uses the same primitives andContextual Component that he uses when generating language.

    Because my approach analyzes written language, not spoken language, it

    39

  • 8/3/2019 Automatization Using FDG

    40/89

    will not have a Phonological Level as input. It starts with the MorphosyntacticLevel, which I assume can be obtained from a text by conventional parsingtechniques. Without a Phonological Level, and under the assumption that a

    Morphosyntactic Level has already been created, the first step in understand-ing an utterance is building a Representational Level from the MorphosyntacticLevel. That step is what the approach described in this chapter will do. More-over, it will do it in such a way, that the techniques described in this chaptercan also be used to build an Interpersonal Level from the Morphosyntactic andRepresentational Levels.

    4.2 Overview of the Approach

    4.2.1 FDG Structures as Trees

    To understand the approach explained in this chapter, it is necessary to re-gard FDG layer structures as trees. Trees are a fundamental data structure incomputer science, as explained in [Knuth, 1997].

    In FDG layer trees, there are five types of nodes: identifier nodes, identifierchild nodes (the operators, head and modifiers nodes), restrictor nodes, listnodes and constant nodes.

    To regard a layer as a tree, one must regard the identifier as an identifiernode. Such an identifier node has three children: its operators node, its headnode, and its modifiers node. The operators node has the operators of the layeras its children. The children of the operators node are called operator nodes.Operator nodes do not have any children, they are terminal nodes.

    If the layer has no head, the head node has no children. Otherwise, thehead node has a node representing the head restrictor as its child. This noderepresenting a restrictor is labeled with the function that the layer identifierhas with respect to this restrictor. If the restrictor is an FDG layer, the child ofthe restrictor node is the identifier node of that layer. If the head restrictor isa grammatical or lexical constant, a terminal node labeled with that constantis the only child of the restrictor node. If the head restrictor is a list, the onlychild of the restrictor node is a list node that has the elements of the list as itschildren.

    The modifier node has all modifiers of the layer as its children. Modifiersare turned into trees just like heads: the root of the subtree for a modifier isa restrictor node labeled with the function of the layer identifier with respectto the restrictor being represented. That restrictor node has one child. If therestrictor is a lexical constant, that child is a terminal node labeled with thatconstant; a layer is recursively expanded to an identifier node with its operators,head and modifiers nodes; and a list becomes a list node with its elements aschildren.

    The edges that connect a restrictor node to its one child, are labeled withthe function that this restrictor itself has, if it has one. The same goes forlist nodes: the edges to the children are marked with the functions of thosechildren.

    40

  • 8/3/2019 Automatization Using FDG

    41/89

    operators modifiershead

    xi

    Figure 4.1: The tree representation of (xi)

    4.2.1.1 Some Example FDG Trees

    In this section, I give some examples to illustrate how I regard FDG layerstructures as trees. In these examples, I display identifier nodes as circles,identifier child nodes as boxes, restrictor nodes as ovals, list nodes as dots andconstant nodes as diamonds.

    As the simplest possible example, take the layer (x i), as it appears repre-senting an anaphoric reference. This layer has an empty set of operators, it hasno head, and it has an empty set of modifiers. Because (xi) is a layer, it getsan identifier node for its identifier xi. Because this node is an identifier node, italso has a modifiers child, a head child, and a modifiers child. But because thelayer has no operators, head or modifiers, these three children have no children

    of their own. So the tree becomes as given in Figure 4.1.As a second example, that includes list nodes and heads, consider the Mor-

    phosyntactic Level of the constituent those ingredients from (21). The Mor-phosyntactic Level is repeated here as (52). Its rendition as a tree is given inFigure 4.2.