Luận án tiến sĩ: Understanding the figurative language of tropes in natural language processing using a brain-based organization for ontologies

When Shakespeare writes, we see the intent in Romeo’s words, but it is lost againwhen we attempt to express it using a computer model for language; a model with anability to handle trope

Understanding the Figurative Language of Tropes in Natural Language Processing Using a Brain-based Organization for Ontologies

by

Christine M Keuper

A dissertation submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy

Graduate School of Computer and Information Sciences

Nova Southeastern University

2007

1

3244325 2007

Copyright 2007 by Keuper, Christine M.

UMI Microform Copyright

All rights reserved This microform edition is protected against unauthorized copying under Title 17, United States Code.

ProQuest Information and Learning Company

300 North Zeeb Road P.O Box 1346 Ann Arbor, MI 48106-1346 All rights reserved.

by ProQuest Information and Learning Company

We hereby certify that this dissertation, submitted by Christine M Keuper, conforms

to acceptable standards and is fully adequate in scope and quality to fulfill the tation requirements for the degree of Doctor of Philosophy

disser- _

Chairperson of Dissertation Committee

_

Dissertation Committee Member

_

Dissertation Committee Member

Approved:

_

Dean, Graduate School of Computer and Information Sciences

Graduate School of Computer and Information Sciences

Nova Southeastern University

2007

2

Processing Using a Brain-based Organization for Ontologies

byChristine M Keuper

2007

Look, love, what envious streaks

Do lace the severing clouds in yonder east;

Night's candles are burnt out, and jocund day Stands tiptoe on the misty mountain tops.

“Romeo and Juliet,” Shakespeare

Language communication is the successful interpretation of the speaker’s communicativeintent When Shakespeare writes, we see the intent in Romeo’s words, but it is lost againwhen we attempt to express it using a computer model for language; a model with anability to handle tropes (metaphor, metonymy, synecdoche and irony) is needed The goal

of this model is to correctly interpret the nouns that occur within these tropes

Early computer language models had not worked well when they encountered tropes, yetthe brain handled them easily These early models concentrated on the language functions

of the left temporal lobes of the brain; perhaps the models worked poorly because theyhad limited themselves to modelling only the parts of the brain that handled propositionallanguage The designs of these models were also influenced by the assumption that thehuman brain understood language using a grammar-based Language Acquisition Device

In examining human language acquisition however, grammar does not even show up untilthe third year

In addition to the common taxonomic and mereologic structures that occur in most

language models, the current model also recreates the brain’s thematic, perceptual andfunctional categorizations Words no longer occur at a single location: words defined bytheir perceptual features, whether nouns or adjectives, occur within perceptual categoriza-tions, and those defined by functional features, whether nouns or verbs, occur withinfunctional categorizations Tenor-vehicle connections then expand these perceptual andfunctional categories with metaphor Words occurring within thematic categories are used

to understand metonymy; and words occurring in the taxonomic and mereologic tures are used to understand synecdoche

struc-Classifiers, such as the Japanese hon, indicate membership in a category Marked

percep-tual and functional classifiers in ASL, Japanese and Swahili made it easier to identify theoccurrences of unmarked perceptual and functional categories in English Likewise, themythos-based categories in Dyirbal, French and German made the remnants of mythos-based categories still occurring in English understandable

This is one or two pages, page iii or pages iii and iv The page number(s) should not be

printed The abstract should be written in the past tense It should contain the problemstatement, method(s) employed, results/findings, conclusions, and recommendations Itshould not exceed 350 words The abstract is single-spaced

a college diploma the same year I did as well He is also missed.

To my daughter Francie, who was still an infant when this journey began, the day I went across town to the Polytechnic University in San Luis Obispo, California and became part of a very small group of women who wanted to study engineering amongst the thousands of men there She taught me about child language acquisition, she was a joyful part

of my life and gave me a reason to get up every morning, and she was emotional support to me many years later when

we were both in graduate school at the same time To my older brother Robert who followed me to the university, also

to study engineering, but who died after developing a fatal cancer He always believed in me To my youngest brother Phillip, who I raised from infancy, who is also no longer here To my younger sister Karen, who was always a safety net for me.

To my professors at Cal Poly: To Dr Peter Litchfield, who taught me experimental psychology methodology To Dr Barbara Cook, who taught me cultural anthropology To Dr Robert Lint, my linguistics professor, for the wonderful sense of déjà vu that occurred when I walked into my first compiler design class, for all of the questions he asked, some

of which I am still trying to answer here many years after his death To Dr Jay Bayne, my advisor, and Dr Emile Attala, my thesis advisor, for encouraging my love of computer science, and all of the fantastical directions I wanted to

go with the computer.

To Lisa Krasna, who let me adopt and raise her deaf, autistic son, Jeremy To Jeremy, who taught me what I didn’t know, and who has become a great joy in my life in his adulthood To Dr Edward Ritvo, for his medical research that opened the doors for Jeremy, and for introducing me to Bill Christopher To Bill Christopher, who also has an adopted, autistic son, Ned, and who introduced me to Dr Art Schawlow and his wife Aurelia, who had an autistic son, Artie To Art and Aurelia who encouraged me to continue the development of the methodology I used to teach Jeremy language, and who also both encouraged me to continue my studies in computer science They are both missed To Alan Alda, who encouraged me to continue development of the sign language dictionary I was working on, and who encouraged

me to return to graduate school.

To Dr Graham Chalmers, my friend and advisor of many years To Mark Lucas and Scott Simon, who were always there with new language features for the development environment To Dr John Bonvillian, whose emails helped me refine my thoughts and theories about language models To Dr Bill Stokoe, who spent years encouraging me via email

to continue with my linguistic and computer science studies, and who I finally met in person shortly before his death.

To Dr Jerry Keuper who, after hearing I was interested in computational linguistics, sent me a copy of his book on Chinese idiom as well as a few chapters of a book he was writing on Spanish idiom, and who also called me on my first day as a new doctoral student at Nova to encourage me Drs Stokoe and Keuper are both missed as well.

To my daughter Meagan, who is now away at college studying industrial design, for her love and support, and for as a young child being proud to tell her friends that her mother was studying for a “doctorette.”

And finally last, but certainly not least, to my professors at Nova Southeastern University, all of whom supplied me with a quality education To Dr Rollie Guild, who started working with me when I was a new student at Nova, directing my early research To Dr Lee Leitner, who continued after Dr Guild’s death, helping me take a vague idea and start to turn it into a dissertation To my dissertation committee, Dr Michael Laszlo, Dr James Cannady and Dr Amon Seagull, for their interminable patience, and for the excellent direction and feedback they provided me with while working on this dissertation.

Abstract iii

List of Tables x

List of Figures xii

Chapters 1 Introduction 1

Issues 4

Can Something “Not in the Real World” be Represented in a Classic Taxonomy? 4 Can There be More than One Conceptual System? 5

Can an Interlingua Represent Concepts Independent of Language? 7

Limitations 9

The Autonomy Hypothesis and the Lexical Independence Hypothesis 9

Delimitations 10

Pre- and Post-editing to Resolve Ambiguity 10

Pragmatic Ambiguity 10

2 Relevance, Significance, and Brief Review of the Literature 12

Introduction 12

Early Attempts at Machine Translation of Natural Language 13

Is There a Language Acquisition Device? 15

The Development of Tropes 16

Language Stimuli 18

Perceptual Conceptualization and Lexicalization 20

Basic-level Perceptual Categorization, Prototypes, and Radial Structures 25

Perceptual Categorization in Navaho, Japanese, and ASL 29

Morphology and Categorization 31

Action Verbs 34

Thematic Categorization 36

Thematic Roles 38

Arbitrary “one criterion” Categorization and Ad-hoc Categorization 39

Metaphoric Categorization 40

Orientational Metaphor 42

Ontological Metaphor 43

Mythos-based Categorization in Dyirbal 44

Mereologic Categorization 48

Part-whole Hierarchies Across Languages 50

Taxonomic Categorization 52

Contrastive Ambiguity and Taxonomic Categorization 52

Taxonomic Categorization in German 53

Category Markedness and Taxonomic Ambiguity 55

Grammar 56

Grammatical Ambiguity 58

v

Metonymy 60

Synecdoche 61

Irony 62

Context Switching 64

Where’s the Syntax? 65

Summary 68

3 Methodology 69

The Proposed Model 70

Paradigm and Syntagm 71

Paradigmatic Language 74

Time Metaphor and Orientational Metaphor 74

Perceptual Metaphor 76

Ontological Metaphor 80

Tenor-vehicle Metaphor 81

Contrastive Ambiguity 84

Paradigmatic Selection 87

Mereologic and Taxonomic Ambiguity 90

Mereologic and Taxonomic Synecdoche 91

Syntagmatic Language 92

Chunking, Idiom, and Irony 92

Thematic Categorization 93

Functional Categorization 94

Grammatical Inflection 97

Complementary Ambiguity 97

Grammatical Inflection in Idiom 98

Syntactic Ambiguity 98

Thematic-and Function-based Metonymy 99

Format for Presenting Results 99

Evaluation of the Results 100

Summary 103

4 Results 104

Data Analysis 104

Input 104

Internal Representation 106

Brain Structure Modules 107

Language Processing 109

The Right Anterior Temporal Lobe Module 111

Parsing 111

Idioms and Collocations 112

Agglutinative and Derivational Languages 113

Irony 114

Echolalia 114

The Right Frontal Lobe Module 116

Switching Languages 117

vi

The Left Anterior Temporal Lobe Module 118

Grammatical Inflection and Function Words 118

The Right Posterior Temporal Lobe Module 120

Perceptual Categorization and Perceptual Classifiers 120

Time and Orientational Metaphor 132

Ontological Metaphor 134

Thematic Categorization and Metonymy 136

The Left Motor Cortex Module 137

Functional Categorization and Contrastive Ambiguity 138

Functional Categorization and Complementary Ambiguity 139

Verb-Noun Pairs and Subject-Verb-Object Groupings 140

Functional Ambiguity 145

Retention of S-V-O in Broca’s Aphasia 146

Verb Loss in ALS 147

The Left Posterior Temporal Lobe Module 147

Hierarchical Categorization and Hierarchical Ambiguity 147

Mereology-based Interlingua 149

Mereology-based Synecdoche 150

Taxonomy-based Synecdoche 152

Wernicke’s Aphasia 153

The Right Motor Cortex Module 155

Functional Categorization and Tenor-vehicle Metaphor 155

Asperger’s Syndrome 156

The Left Frontal Lobe Module 157

The Impact of “Not Implemented” 157

Syntactic Ambiguity 158

Broca’s Aphasia 159

Findings 159

Comparison to Language Acquisition, Aphasiology, and Autism Models 160

Comparison to Learning Models 160

Comparison to Propositional Models 160

Comparison to Grammatical Models 161

Comparison to Statistical Models 162

Comparison to Interlingual Models 163

Comparison to Cruse’s Examples of Taxonomic Ambiguity 163

Comparison to Pustejovsky’s Contrastive & Complementary Ambiguity 163

Comparison to Examples of Functional Ambiguity 164

Comparison to Examples in Fillmore’s Case Theory 165

Comparison to Jackendoff’s Examples of Thematic-based Metonymy 167

Comparison to Chandler’s Examples of Synecdoche 168

Comparison to Examples of Tenor-vehicle Metaphor 169

Comparison to Narayanan’s Examples of Metaphor 169

Comparison to Lakoff’s Examples of Classifiers and Categorization 172

vii

Comparison to Lakoff’s Examples of Ontological Metaphor 173

Summary of the Results 174

5 Conclusions, Implications, Recommendations, and Summary 177

Conclusions 177

Evaluation of Error 178

Limitations of the Findings 181

Implications 182

Recommendations 184

Summary 186

The Implementation 189

Computer Models of Mental Processes 190

Appendices A Some History of Computers and Natural Language 192

Rule-based Direct Translations 192

Transfer Approaches 192

Interlingua Approaches 193

Corpus-based Systems—Statistical Methods and Example-based Translation 195

Knowledge-based Systems 195

B Autism 197

The Triad of Impairment 197

The Rates of Autism in Neurocutaneous Disorders 198

C Classical Conditioning 200

D The Midbrain 202

E A Mereologic Structure from the MeSH 204

F Mereologic Groups for Animals 207

G Strong Verbs 209

H Three Possible Selections From Fabeln 211

I The Sentences 212

Sentences by Language 213

ASL 213

Dani 213

Danish 213

Dyirbal 214

English 214

French 250

German 250

Hasau 252

Hawaiian 253

Hebrew 253

Inuit/Yupik 253

Irish 253

Italian 253

Japanese 254

viii

Spanish 254

Swahili 255

Tarahumara 258

Examples by Order of Occurrence 259

Examples by Topic 299

Parsing 299

Ambiguity 300

Metaphor 307

Metonymy 309

Synecdoche 310

Classifiers 310

Switching S-V-O/S-O-V/V-S-O 312

Meaning and Grammar 313

J The Rapid Application Development Prototyping Environment 315

Reference List 317

ix

List of Tables

Table 1 The Development of Tropes 17

Table 2 Morphology and Categorization in Swahili 32

Table 3 Bantu Roots 33

Table 4 Thematic Roles 39

Table 5 Literal Meaning and Metaphor 44

Table 6 Dyirbal Classification (in English) 45

Table 7 Dixon’s Dyirbal Classification System 45

Table 8 Part/whole and Class Inclusion Hierarchies 50

Table 9 Part/whole Relationships Across Languages 51

Table 10 Part/whole Relationships Applied to Colors 52

Table 11 Das Gemüse Superordinate Category 53

Table 12 Das Tier Superordinate Category 54

Table 13 Superordinate Neuter Classification of Animals 54

Table 14 Neuter Classification of Non-indigenous Animals 54

Table 15 Category Markedness 55

Table 16 Grammatical Inflection in Nouns, Adjectives and Verbs 58

Table 17 ASL vs English 59

Table 18 The Ambiguity of fly 60

Table 19 Metonymy 60

Table 20 Synecdoche 61

Table 21 The Paradigmatic and Syntagmatic Aspects of Language 72

Table 22 Time Metaphor 75

Table 23 Orientational Metaphor 75

Table 24 Perceptual Metaphor 79

Table 25 Chinese Phoneme ma 84

Table 26 Banke, Baunke and Banque Merge into Bank 84

Table 27 The Meanings of Baunke, Banke and Banque 86

Table 28 Sample Languages by Language Family 105

Table 29 Sample S-O-V, S-V-O, & V-S-O Languages, with Number of Speakers 106

Table 30 Functionality Implemented in the Modules 108

Table 31 Language Disorders 108

Table 32 Perceptual Classifiers in ASL, English, Japanese and Swahili 120

Table 33 -dege 121

Table 34 -tabu 122

Table 35 Perceptual Descriptors 130

Table 36 Thematic Definitions 154

Table 37 Disambiguating bill 156

Table 38 Choice of Prototypical Examples 175

Table 39 Complementary Ambiguity 180

x

Table 41 Transfer Systems 193

Table 42 Direct Translation vs Interlingua With 10,000 Entries 194

Table 43 Interlingua Systems 194

Table 44 Corpus-based Systems 195

Table 45 Knowledge-based Systems 196

Table 46 Reported Cases of Autism 1943-1994 198

Table 47 Reported Cases of Autism 1997 and 2004 198

Table 48 Neurotransmitters Important for Language 203

xi

List of Figures

Figure 1 La Peau Rouge vs Le Peau-rouge 8

Figure 2 Metaphorical Deep Structure 10

Figure 3 Pragmatic Ambiguity 11

Figure 4 Language Stimuli 19

Figure 5 The Cerebellum 19

Figure 6 Tonal Rhythm and Lexicalization 21

Figure 7 Perceptual Categorization 26

Figure 8 Japanese hon 30

Figure 9 Action Verbs 35

Figure 10 Thematic Categorization 37

Figure 11 Arbitrary and Ad Hoc Categorization 40

Figure 12 Metaphoric Categorization 41

Figure 13 Creating Metaphor 42

Figure 14 The Back of an Object 43

Figure 15 A Dyirbal Classification System 46

Figure 16 Mereologic and Taxonomic Categorization 49

Figure 17 Pustejovsky’s Category and Genus of bank 53

Figure 18 Grammatical Inflection 57

Figure 19 Metonymy 61

Figure 20 Synecdoche 62

Figure 21 Irony 63

Figure 22 Context Switching 64

Figure 23 Cultural Ontologies 65

Figure 24 Complex planning 66

Figure 25 Classic Language Analysis 69

Figure 26 Nirenburg’s Ontology 70

Figure 27 Proposed Language Analysis 70

Figure 28 Paradigmatic and Syntagmatic Language 75

Figure 29 The Meanings of pike and pikestaff 76

Figure 30 The Meanings of pike 76

Figure 31 A Perceptual pike Ontology 77

Figure 32 The Meaning of pikestaff 77

Figure 33 A Perceptual pikestaff Ontology 78

Figure 34 pikestaff vs pike Ontologies 80

Figure 35 Metaphor Evoking the Underlying Mythos 81

Figure 36 The Metaphoric Meaning of pike 82

Figure 37 The time is-a pikestaff Metaphor 82

Figure 38 A snow blankets the ground Metaphor 83

xii

Figure 40 Paradigmatic Selection 87

Figure 41 Paradigmatic Selection of Cooking Terms 89

Figure 42 A Ceramics Oven and a Dutch Oven 90

Figure 43 Taxonomic and Mereologic Ontology 90

Figure 44 Examples of Taxonomic Ambiguity 91

Figure 45 Examples of Synecdoche 91

Figure 46 A Taxonomic computer Ontology 91

Figure 47 A Mereologic credit card Ontology 92

Figure 48 Cuéntaselo a tu tía 93

Figure 49 A Thematic bedroom Ontology 94

Figure 50 A Thematic pastrami Ontology 94

Figure 51 An eat Functional Ontology 94

Figure 52 The essen and fressen Functional Ontologies 94

Figure 53 Classic husband Ontology 95

Figure 54 Functional husband Ontology 96

Figure 55 The to husband Ontology 96

Figure 56 The English Word bill 96

Figure 57 Inconsistent Grammatical Inflection 97

Figure 58 Grammatical Inflection 98

Figure 59 Animal Crackers 98

Figure 60 Time Flies 99

Figure 61 Examples of Metonymy 99

Figure 62 The Structures of the Ontologies 100

Figure 63 Evaluating Compilers 101

Figure 64 Language Models Represent Small Portions of Language 102

Figure 65 The Internal Representation Window 106

Figure 66 The Brain Structures Window 107

Figure 67 The RATL module 111

Figure 68 The Sign Language Input Screen 112

Figure 69 You see red 112

Figure 70 We drove to New York 113

Figure 71 Rechtsschutzversicherungsgesellschaften ist das größte Wort im deutschen Wörterbuch 113

Figure 72 That’s fine with me 114

Figure 73 Deactivating the RATL module 115

Figure 74 lion 115

Figure 75 It’s plain as a pikestaff 115

Figure 76 The lion is big 116

Figure 77 La chair crue a une texture croquante et une saveur poivree 117

Figure 78 The RFL module 117

Figure 79 Le piquant du gout peut etre retire en retirant la peau rouge 118

Figure 80 Oumpah Pah le Peau Rouge is a comic book 118

Figure 81 Bill saw the bank of clouds 119

xii

Figure 82 Bill saw the bank of the river 119

Figure 83 Kitambaa kidogo kitatosha 121

Figure 84 Kitabu 122

Figure 85 Kijitabu 122

Figure 86 Empitsu wa gohon desu 122

Figure 87 Biiru o nihon kudasai 123

Figure 88 long, straight road hierarchy 123

Figure 89 long, thin fish hierarchy 124

Figure 90 pikestaff: long, rigid, thin, straight 124

Figure 91 The boy caught a pike 125

Figure 92 spike: short, rigid, thin, straight, pointed 126

Figure 93 high, narrow heel hierarchy 126

Figure 94 Father drove the car 127

Figure 95 The fireman fell to the ground 128

Figure 96 ASL Signs 128

Figure 97 Defining Characteristics of the Signs 129

Figure 98 ASL Classifiers 129

Figure 99 Father put the pencil on the table 130

Figure 100 floor, ground Classifier 131

Figure 101 Fireman fall down on the floor/ground 131

Figure 102 ASL Brain Structures 131

Figure 103 Saa mbili asubuhi 132

Figure 104 Saa kumi na moja jioni 133

Figure 105 Rokuji 133

Figure 106 Rokujikan 133

Figure 107 Ga cokali can baya da kwarya 134

Figure 108 Mahali pa hatari 134

Figure 109 Bill is blue 135

Figure 110 RPTL module 135

Figure 111 The pastrami wants the check 136

Figure 112 The White House vetoed the bill 137

Figure 113 White House Brain Structure 137

Figure 114 Pike catches boy 138

Figure 115 Bill wants the bill 139

Figure 116 The boy went through the door 140

Figure 117 The ghost went through the door 140

Figure 118 Bill baked the cake 141

Figure 119 The clay pot was baked in the oven 141

Figure 120 Bill baked a cake in the dutch oven 142

Figure 121 The house is an oven 142

Figure 122 Bill took a cake of soap 143

Figure 123 The car is caked with mud 143

Figure 124 The makeup is caked 144

Figure 125 The husband baked the cake 144

xiv

Figure 127 Bill husbanded his resources 145

Figure 128 The boy runs the course 146

Figure 129 The water runs its course 146

Figure 130 Bill washed the dishes 148

Figure 131 Bill washed the dishes Brain Structure 148

Figure 132 The lion ate the dog 149

Figure 133 Wir essen und fressen und tanzen und trinken 150

Figure 134 The recycling center takes plastic 151

Figure 135 The store takes plastic 151

Figure 136 UCLA won the game 152

Figure 137 Bill used the Hoover 152

Figure 138 Brain Structure Wernicke’s Aphasia 153

Figure 139 Taxonomic grandmother 153

Figure 140 Broken LPTL 154

Figure 141 Thematic grandmother 154

Figure 142 The words cut Bill 155

Figure 143 The words cut Bill Brain Structure 156

Figure 144 The words cut Bill—Brain Structure Without RMC 157

Figure 145 The words cut Bill—Internal Representation Without RMC 157

Figure 146 Bill will diet and exercise if his doctor approves 158

Figure 147 Fruit flies like a banana 159

Figure 148 Mereologic and Taxonomic Categorization 165

Figure 149 Grandmother baked for an hour 165

Figure 150 Grandmother and the cake baked for an hour 166

Figure 151 Banggun ganibarragu budin bangun gujarra 167

Figure 152 The ham wants the bill 168

Figure 153 The policy backs the track 171

Figure 154 Policy back on track 171

Figure 155 Narayanan’s Metaphoric Mappings 171

Figure 156 Noun Mappings 172

Figure 157 Ja antwortet der Löwe 174

Figure 158 The ham wants the bill 179

Figure 159 Translation Into an Interlingua 193

Figure 160 Language in Autism 197

Figure 161 The Reticular Formation in the Midbrain 202

xv

Chapter 1 Introduction

“The strategy I’m adopting here is to build more lamps.”

Ray Jackendoff

Science fiction often offers us a glimpse into what the human mind considers doable,sometimes centuries before the details of developing the necessary technology will beworked out.1 One of the common characters in science fiction has been the robot capable

of carrying on a conversation with us, the computer capable of processing and usingnatural language: Asimov’s robots, HAL in 2001, the Universal Translator on the Enter-prise Even though we struggle at how to accomplish this goal we still believe in ourhearts that computers will eventually have the ability to process natural language

The brain is currently our only working model2 for language understanding WhenBroca published his first brain study in 1865, he postulated that language was based in theleft anterior temporal lobe In 1874 Wernicke discovered the left posterior temporal lobewas also involved in language; Broca’s area was the seat of grammar and Wernicke’s areathe source of the lexicon Language models of that time followed this lead and assumed

1 As long as science has been around, futuristic thinkers have recorded “ what might be.” That Leonardo da Vinci could conceive of ideas 400-500 years before the necessary science and engineering were available makes it less surprising that writers also spoke of future technology, (though the term “science fiction” was only coined in 1954): Cyrano de Bergerac’s “Voyage to the Moon” (1649) was 320 years before our voyage to the moon, Margaret Cavendish’s “Description of a New World, Called the Blazing World” in

1666 was 340 years ago, and of course Mary Shelly’s Frankenstein dates back to 1818, and will be 200 years old in a dozen years.

2 A model in the sense of “a thing that serves as a pattern” rather than as “a set of plans”

that a grammar combined with a lexicon could describe language completely In thecontext-free world of computer languages this model might have worked, but when it wasapplied to explaining natural languages the results were disappointing.

The nature and structure of language is not well understood Attempts to understandlanguage through defining a grammar go back over 2000 years but these attempts haveonly been partially successful (Manning & Schütze, 1999) As Edward Sapir (1971) said,

“All grammars leak.” Even then, the grammaticality of a sentence, the judgment ofwhether it is structurally well-formed, does not guarantee that the sentence carries anymeaning3 (Huck & Goldsmith, 1995); language communication is not the exchange ofsymbolic expressions; it is the successful interpretation of the speaker’s communicativeintent (Green, 1996)

The field of Generative Semantics, which attempted to define a path from externallanguage form to meaning, grew out of this concern To the Generative Semanticists,semantics initially meant logic, and early attempts to create semantic language modelsused first order predicate calculus However, logic-based language models failed miser-ably in dealing with the extensive occurrence of figurative language (Lakoff, 1987), andthe formal semantic theories did not addressed the use of words in novel contexts (Puste-jovsky, 1998) The field of objectivism also wanted to avoid the use of figurative lan-guage to grasp the world, but it was not possible:

3 In 1957, Chomsky created a sample sentence, colorless green ideas sleep furiously, whose grammar is

correct but which has no meaning.

2

An attempt to avoid figurative language became closely allied to the realist ideology of objectivism Language and reality, thought and language, and form and content are regarded by realists as separate,

or at least as separable Realists favor the use of the 'clearest', most 'transparent' language for the accurate and truthful description of 'facts' However, language isn't 'glass' (as the metaphorical refer- ences to clarity and transparency suggest), and it is unavoidably implicated in the construction of the world as we know it Banishing metaphor is an impossible task since it is central to language (Chan- dler, 2001).

Figurative language, or tropes, are often treated as though they are anomalous

lan-guage forms, but they are instead an integral part of lanlan-guage (Kittay, 1987) Because thefigurative use of language is so ubiquitous, Jonathan Culler referred to tropes as “a

system, indeed the system, by which the mind comes to grasp the world conceptually in

language” (Chandler, 2001)

Giambattista Vico (1668 — 1744) was the first to identify the four basic tropes as

metaphor, metonymy, synecdoche, and irony (Chandler, 2001) Metaphor is connecting a

thing to something else that has a similarity to it but is unrelated:4 leeway has a literal meaning of the lateral drift of a vessel due to the force of the wind; freedom becomes connected to lateral drift through metaphor, and the meaning of leeway is expanded to include room for freedom Metonymy is the use of the name of a thing for something else

that has a thematic association5 with it: the White House said means the President said Synecdoche is the use of the name of a thing for something else that has a taxonomic or

mereologic6 association with it: Did you see the new wheels I got means did you see the

4 Metaphor is often used as an all-inclusive term for figurative expressions, and likewise metonymy is often

used to include both metonymy and synecdoche Each of these terms will only be used here in their most restrictive sense.

5 A thematic relationship is between things that occur in the same place and at the same time, the tion between the dog, his leash, and his food dish.

connec-6 A mereologic association is a part/whole relationship.

3

3

new car I got Irony is the use of an opposite for the intended meaning: This is a fine

situation you have gotten us into means this is a bad situation you have gotten us into.

The work being described is a computer-based model that deals with the extensiveoccurrence of tropes in natural language The model is limited in its scope to the forma-tion of structures to represent nouns, and the new ontologies that can better represent thefigurative use of those nouns These structures are based on how the brain acquireslanguage at a functional level,7 how the basis of that language is conceptual and meta-phoric, and how the creation of perceptual, thematic, and metaphoric categorizationoccurs at an early stage of language development, continues throughout language acquisi-tion, and is integral to language use.|

Issues

Before creating an underlying structure to represent the nouns, there are severalconcerns that need to be addressed:

° Can something “not in the real world” be represented in a classic taxonomy?

° Can there be more than one conceptual system?

° Can an interlingua8 represent concepts independent of language?

Each of these issues will be discussed in turn

Can Something “Not in the Real World” be Represented in a Classic Taxonomy?

The meanings of words have to be reflected in the structures created to represent them

(Pustejovsky, 1998) If these structures are limited to classic taxonomies then only objects _

7 Brain descriptions are at a functional level for each area of the brain rather than at a lower, cellular level, with the exception of the cerebellum, which is broken down in enough detail to show the difference in paths between conditioned and unconditioned stimuli.

8 An interlingua is a layer below the surface level of a language, used to represent concepts for translation into another language See Appendix A for more details on interlingua.

4

5that are part of objective reality are categorizable Langacker (1990) states that meaningdoes not reside in objective reality, nor is it represented in terms of truth conditions, but iswithin the realm of cognitive processing So, the interpretations of some “things” are

based on perceptions rather than being objective and “in the real world.”

An extreme example of how ontologically relevant entities depend on our perceptive and cognitive structures is the notion of constella- tion: is a constellation a genuine thing, different from a collection of stars? This is not so clear at first sight But, if we distinguish be- tween stars and their specific arrangements, we are able to under- stand how constellations may be considered as cognitive things de- pendent on states of mind (Gangemi et al., 2001).

When languages such as Dyirbal (the aboriginal language of Australia) are examined,the underlying cultural belief that some things possess a spirit will become crucial tounderstanding why the Dyirbal language is structured as it is, and this will have to bereflected in the structures created to represent concepts in Dyirbal (there is additionaldiscussion of mythos-based categorization in Dyirbal on pg 44)

Can There be More than One Conceptual System?

Early attempts to create semantic-based translation models used first order predicatecalculus Lakoff (1987) states that predicate calculus “assumes an a priori view of catego-rization, namely, the classical theory that categories are sets defined by common proper-ties of objects Such an assumption makes it impossible to ask, as an empirical question,whether the classical view of categorization is correct.” Classical categorization hasbecome the background assumption, the unquestioned truth, at the basis of all otherdisciplines (Lakoff, 1987)

55

5

6Classical categorization, a view which has existed from the time of Aristotle, is based

on categories as abstract containers.9 Objects were either inside the “container” or side, and all the objects inside shared the properties of the category equally (Lakoff,1987) Since the properties defining the category are shared by all of the members inclassical theory, if classical categorization was complete, then no member of a categorywould have any kind of special status (Lakoff, 1987) Without this special status, lan-guages not using a classical categorization, such as Dyirbal, appear unintelligible anduntranslatable

out-Two important claims are made when trying to understand and translate conceptualsystems that may not use classical categorization The first claim is, if two languageshave radically different conceptual systems, then understanding and translation betweenthem is impossible All conceptual systems vary as to the “fineness of grain” of theconcepts they contain; the differences have to be fundamental, such as how space andtime are dealt with (Lakoff, 1987) Minor domain-of-experience vocabulary is not aproblem, as with the claimed large number of words for snow in Eskimo;10 all languageshave this phenomena (Lakoff, 1987), with seafaring terms, or technology, or even linguis-tics

The second claim is, if two languages have different conceptual systems then learning

the other language is not possible, and if people can learn radically different languages,

_

9 A category seen as a “container” is itself a metaphor (Lakoff, 1987).

10 This is a story that has grown with retelling, starting at “4 words for snow in Eskimo” (Boas, 1911) and growing to “100 words for snow” (New York Times, 1984), with other sources going as high as 200 and

400 English has snow, sleet, slush, blizzard, pack, powder, etc (and facetiously, water for melted snow,

rain for melted snow falling from the sky, steam for gaseous snow, and flood for fast moving melted snow

coming out of the mountains in the Spring) The actual four Inuit/Yupik words are aput (snow on the ground), gana (falling snow), piqsirpoq (drifting snow), and qimuqsuq (snow drift) (Pullum, 1991).

6

7then those languages could not have different conceptual systems (Lakoff, 1987) Sincepeople have a general conceptualizing capability as well as an ability to express conceptsmetalinguistically, understanding different conceptual systems is clearly possible (Lakoff,1987) Learning and translating between different conceptual systems should also bepossible, as a single language can have multiple, and even incompatible conceptualmodels (Lakoff, 1987) Because we deal with incompatible models everyday, they move

out of our awareness When we get up in the morning, the sun rises (part of the folk

model), even though a few hours later in science class, it is the earth that is rotating ratherthan the sun moving (the scientific model) A botanical model gives us a definition of

fruit as the seed bearing part of a plant used for food Using this definition, tomatoes,

eggplant, avocados, cucumbers, and zucchini are fruits In everyday life however, the folkclassification of these foods are according to their use and flavor, and they are put into thecategory with vegetables

Can an Interlingua Represent Concepts Independent of Language?

An interlingua might be used for concepts such as colors since the color spectrum can

be broken down into subdivisions smaller than exists for the naming of colors in any ofthe languages However, the interlingua could not be used by itself for doing a transla-tion Since the interlingua would only include things that are part of the physical world,the metaphoric use of color would be lost, along with any specific cultural informationabout how the color spectrum is divided

What we see as a single concept in English may not be so in another language, as

with the different American Sign Language (ASL) signs for run when referring to a

77

person running vs an animal running, and with the German essen (to eat when applied to

a person) and fressen (to eat when applied to an animal).

Problems would also occur in the French language, as compounds can have different

categorization information from individual words: peau rouge, “skin red,” can be line even though peau is feminine (Figure 1), and cordon bleu, “ribbon blue,” can be

Figure 1 La Peau Rouge vs Le Peau-rouge

feminine even though cordon is masculine (Abeill, Clement, & Toussenel, 2003) But

what has happened is not just a grammatical variation; a switching of categories has

occurred A person who is le peau-rouge would need more than a sunburn to be ered a member of that football team Cordon bleu is not just a ribbon that is blue; when

consid-categorized, it is an award and not a sewing notion When gender information is seen aspart of categorization rather than as simply grammar, it can operate as a signal that anobject might have changed categories

An additional problem occurs with adjectives, where the color in red apple is ent than that intended with red hair, and the amount of money referred to in expensive car

differ-88/15 p 8 I think

the correct

Limitations

The Autonomy Hypothesis and the Lexical Independence Hypothesis

The Autonomy Hypothesis states that syntactic analysis must be done without ence to semantics (Chomsky, 1957), and the Lexical Independence Hypothesis that “the

refer-meanings of words are independent of any grammatical constructions that the wordsoccur in” (Lakoff, 1987) Both of these hypotheses are false; in natural language thegrammar is not free of the meaning, and the meaning is not free of the grammar (Lakoff,1987; Pustejovsky, 1998) Because this natural language computer model does not dealwith most aspects of grammar, some meanings will be lost; because some aspects ofgrammar are irrevocably tied to meaning, they become part of the semantic structure.What is more significant however, is that the proposed model is based on the deepstructure11 of language being metaphorical instead of grammatical (Figure 2) While some

meaning will be lost, the most important functions of language should not be lost if a

grammar component is not implemented This will be discussed in more detail in laterchapters

8/15 p 9 footnote

11, structure is

misspelled

8/15 Fixed

deep structure grammar

deep structure metaphorsurface structure

metaphor

surface structure grammar

Proposed Model Chomskian Model

Figure 2 Metaphorical Deep Structure

Delimitations

Pre- and Post-editing to Resolve Ambiguity

It is common for natural language systems to use some type of pre-editing, editing, or interactive intervention when problems or ambiguities arise Pre-editing caninvolve the identification of proper nouns, the marking of grammatical categories, flag-ging or substituting unknown words, indicating embedded clauses, or even reformulatingthe input text into a “controlled language” (Hutchins, 1992) Post-editing usually involvesidentifying unresolved ambiguity, correcting the output, or making corrections such aswith inconsistent gender An interactive approach is one that resolves syntactic andsemantic ambiguities during the translation process (Hutchins, 1992) Any pre- or post-editing or interactive intervention in the current system will be identified

post-Pragmatic Ambiguity

Pragmatic ambiguity is created when there is insufficient information available aboutthe context in which an utterance occurs (Figure 3) The pragmatic ambiguity of a

1010

He shot a few bucks.12

She ate the hamburger with relish.

Figure 3 Pragmatic Ambiguity

sentence in isolation can be resolved by the computer by inquiring about the missingcontext interactively, in order to determine the larger context,13 or by doing what most

people do — apply stereotypes This means that he shot a few bucks and she shot a few bucks would be interpreted differently; the stereotype would be used to disambiguate the

sentence.14 In she ate the hamburger with relish, the pragmatic ambiguity would either

have to be left unresolved or the disambiguation would have to be purely a matter of

statistical probability, that relish is associated with hamburger more often than it is with eat This could be implemented using statistics in the language model This does not

guarantee that the correct conclusion has been reached, as is the problem with the matic ambiguity that occurs as part of communication even with native users of a lan-guage.15

prag- _

12 This is a rewording of an example used by Pustejovsky (1998)

13 This context can include the general topic of conversation, the background of the speakers, the languages, and the cultures.

14 This is not a statement of what should be, but of what is.

15 Humor has always made use of disambiguation strategies by manipulating the tone of voice and timing of pauses in order to lead the audience to interpret an utterance incorrectly A famous female comedienne would say, “I like to shop ” followed by a pause that was long enough for the audience to conclude that they had interpreted the sentence correctly, then she would continue, “lift and ”

11

Chapter 2 Relevance, Significance, and Brief Review of the Literature

Discovery consists of looking at the same thing as everyone else and thinking something different.

Albert Szent-Gyögyi

Introduction

Creating a language model immediately brings out proponents of opposing camps;some very great names throughout history have had radically different opinions on thenature of language, how it is acquired, and what the brain’s involvement is in this pro-cess Since the computer model developed in this research to translate the figurativelanguage of tropes is based on how the brain acquires language, relevant literature fromneurology, linguistics, and child language acquisition is included as needed to supply therationale for the resulting choices that will be made in the acceptance, rejection, ormodification of components from previous designs of computer models for naturallanguage

The first attempts to create a computer model for natural language date back to thelate 1940s, but assumptions made about the nature of language led to failure in theseinitial attempts Two very significant missing pieces have been the crucial role of catego-rization and equally crucial role of metaphor in language acquisition by the brain Includ-ing these missing pieces will allow the computer model to better handle the metaphorical

language of tropes

1212/4 “Since the

Early Attempts at Machine Translation of Natural Language

Research into the machine translation of natural language can trace its origins back to

1947, to correspondence between Warren Weaver of the Rockefeller Foundation andNorbert Weiner (Hutchins, 1998) Based on the successes of wartime code-breaking andthe advances in information theory, Weaver wrote a 1949 memorandum suggestingvarious proposals for research Universities in the U.S responded, and by 1954 the firstpublic demonstration of a computer used for translation occurred (Hutchins, 1998) In acollaboration between Georgetown University and IBM, an IBM 701 machine, pro-grammed with six grammar rules and a vocabulary of 250 words, translated severalsentences from Russian to English Leon Dostert, the head of the project, believed at thetime that a specialized computer for language translation was three to five years off(Plumb, 1954) When this had not materialized by 1966, the U.S Government’s Auto-matic Language Processing Advisory Committee (ALPAC), concluded that “there is noimmediate or predictable prospect of useful machine translation,” as machine translationwas twice as expensive as human translation, slower, and less accurate (Hutchins, 1998).Yehoshua Bar Hillel, who held the first full-time post in machine translation at Mas-sachusetts Institute of Technology (MIT) in 1951, and who organized the first conference

on machine translation (Hutchins, 1995), made a pronouncement that might seem ous now, that the machines must be able to process meaning in order to translate language(Nirenburg, 1997) While Bar Hillel felt that meaning should be based on logic, he feltthat designing a logical system for translation was not an obtainable goal (Nirenburg,

obvi-1997):

1313

Expert human translators use their background knowledge, mostly subconsciously, in order to resolve syntactical and semantical ambi- guities which machines will have either to leave unresolved or re- solve by some “mechanical” rule which will every so often result in

a wrong translation (Bar Hillel, 1964).

Let us be satisfied with a machine output which will every so often be neither unique nor smooth, which every so often will present the post- editor with a multiplicity of renderings among which he will have to take his choice, or with a text which, if it is unique, will not be gram- matical Let the machine provide the post-editor with all pos- sible help, present him with as many possible renderings as he can digest without becoming confused by the embarrass de richesse (Bar Hillel, 1964).

This was, in fact, what many machine translation projects were forced to do

Bar Hillel collaborated with philosopher and semanticist Rudolf Carnap,16 and guist Noam Chomsky (Nirenburg, 1997; Wikipedia, 2005) Chomsky had joined the staff

lin-of MIT in 1955, a few years after Bar Hillel, and was involved with the machine tion project for two years (Nirenburg, 1997; Wikipedia, 2005) One of the fateful conclu-sions Chomsky had reached was that the meaning of a sentence was dependent to asignificant degree on its grammatical analysis (Chomsky, 1957; Huck and Goldsmith,1995); Chomsky also came to believe that the brain had a Language Acquisition Device(LAD), that the brain already possessed an innate grammar, and that this grammar repre-

transla-sented the deep structure17 of language (Huck & Goldsmith, 1995) The work of Broca,Wernicke, Carnap and Chomsky had set the stage for the direction machine translationwas going to go for the next several decades.18

_

16 Carnap was a leading figure in logical positivism (the logical analysis of scientific knowledge) He also

felt that language consists of a system of formation rules that may not at any point make reference to

semantics.

17 Chomsky divided language into a surface structure and a deep structure.

18 Appendix A gives more history of computers and natural language.

14

Is There a Language Acquisition Device?

Chomsky’s belief in a Language Acquisition Device (LAD) was formed from two significant hypotheses of his: that there is a Universal Grammar innate to the brain, and

that syntactic analysis must be done without reference to semantics These two

hypoth-eses are called respectively the innateness hypothesis and the autonomy hypothesis.

Unfortunately, research in neurology and language acquisition was not supportingChomsky’s belief in a grammatical LAD (Aboitiz & Garcıa, 1997) If grammar was trulyinnate, and the foundation on which the rest of language is built, it should be the first

aspect of language to show itself; instead, grammar turned out to be one of the last

aspects of language to be learned and used, not occurring until the third year of language

development As Claparéde states in the preface to Piaget’s The Language and Thought of the Child, in looking at thought and language in the child, we have incorrectly applied the

“mold and pattern of the adult mind” (Piaget, 1926)

Not all linguists agreed with Chomsky; there were others, such as George Lakoff,whose beliefs were in sharp contrast (Huck & Goldsmith, 1995) Lakoff, who was origi-

nator of the term generative semantics, received his undergraduate degrees in

Mathemat-ics and English Literature from MIT in 1962, and a Ph.D in LinguistMathemat-ics from IndianaUniversity in 1965 At Berkeley in 1975, Lakoff organized a Linguistics Institute funded

by a NSF grant: Rosch gave her first lecture on basic level categories; Talmy gave hisfirst lecture on spatial relations; Fillmore gave his first lecture on frame semantics; andKay and MacDaniel presented their work on the neurobiology of color categorization.Twenty years after Chomsky had set the stage for the syntactic direction in machine15

15

translation, Lakoff was setting the stage for a semantic underpinning for language sentation.

repre-Rather than viewing grammar being innate, Lakoff sees the ability to conceptualizeand form cognitive models as being that innate, biological basis on which language isbuilt (Lakoff, 1987) These cognitive models, how the child thinks about things, are used

in forming the categories that tropes are based on

The Development of Tropes

Vico hypothesized a historical sequence for the development of the four tropes: frommetaphor to metonymy to synecdoche to irony, and Hayden White compared the develop-ment of the tropes to Piaget’s stages of cognitive development (White, 1978; Chandler,2001) Piaget himself was interested in connections between historical systems and

cognitive development, as was Carl Jung (Gelernter, 1994); the idea of ontogeny pitulates phylogeny may have eventually fallen by the wayside in biology, but it contin-

reca-ued as a useful model for understanding of linguistic development (Gelernter, 1994).Even though Chandler (2001) saw the comparison of the tropes to Piaget’s levels ofdevelopment as a “speculative analogy,” combining Vico’s sequence of tropes and

White’s idea of comparing tropes to Piaget’s stages of cognitive development does in fact

give an accurate order and timing of these tropes during language development (Table 1)

As shown in column 3 of Table 1, language acquisition first goes through a perceptualconceptualization and lexicalization step, then a thematic categorization, a metaphoric

categorization, a mereologic categorization, and then finally a taxonomic categorization.16

16

It is only after these initial stages, and three years into the acquisition of language, thatgrammar begins to become part of the language equation The remaining three tropesthen complete the basic acquisition of language These tropes are so pervasive in lan-guage they can be used to identify the level of language acquisition more accurately than

a simple measure of vocabulary size19 (Vygotsky, 1934; Lantolf, 2005)

Piagetian Stage of Age in Child Starts Acquiring Primary Brain Lobe

Sensorimotor 0 — 2 Language Stimuli Left cerebellum

Perceptual Conceptualization Motor cortex and temporal lobes

Lexicalization Right anterior temporal lobe

Thematic Categorization Right posterior temporal lobe

Metaphoric Categorization Right posterior temporal lobe

Pre-operational 2 — 6 Mereologic Categorization Left posterior temporal lobe

Taxonomic Categorization Left posterior temporal lobe

Grammar Left anterior temporal lobe

Metonymy Right posterior temporal lobe

Concrete Operations 6 — 12 Synecdoche Left posterior temporal lobe

Formal Operations 12 — 18 Irony Right anterior temporal lobe

Table 1 The Development of Tropes

What research in neurology is showing (column 4 of Table 1) is that initial languageacquisition primarily occurs in the motor cortex, the right anterior temporal lobe and theright posterior temporal lobe during the first couple of years, followed by the left poste-rior temporal lobe and the left anterior temporal lobe in subsequent years.|

_

19 When the last trope irony is acquired, normal language acquisition has been successful.

17

When language learning does not occur in the child, as happens with autism,20 then wehave additional information about how the brain learns and processes language.

The following sections will trace the brain’s acquisition of language during thoseinitial years, and supply the rationale for the design of the proposed natural languagemodel The model is intended to represent the foundations of language that must beacquired before a grammar/lexicon model can be applied In doing so, it will supply thenecessary non-grammatical foundations for the interpretation of tropes as well

The descriptions of the order of linguistic acquisition in the various parts of the brainhave been simplified The areas of the brain not being described at each level are notsitting dormant while one area acquires language; however, the descriptions do showwhere the greatest involvement is at that point of language development.21

Language Stimuli

From birth to age six months the language mechanism in the brain primarily involvesthe cerebellum, the motor cortex, and the right anterior temporal lobe Since the leftcerebellum handles input for the right cerebrum,the language input for right anteriortemporal lobe comes from the left cerebellum (Figure 4)

_

20 Autism is a neurological disorder that affects language and communication Half of all autistic children

do not acquire any language Irony forms the dividing line between the highest level of language

develop-ment that can occur in autism and normal language acquisition Autistic children who acquire language never make it as far as this last milestone See Appendix B for more details.

21 WADA tests showed that 95-98% of right-handed people are left brain dominant for the grammatical aspects of speech, as well as 69% of left handers 18% of the left-handers are right brain dominant for these aspects of speech, with the remaining 13% having a bilateral dominance (The WADA test uses sodium amytal , injected into either the right or left carotid artery, to put one hemisphere of the brain to sleep The language in the other hemisphere can then be assessed in isolation) (Caplan, 1998).

18

Anterior Temporal Lobe

Language Stimuli

Motor Cortex Motor Cortex

Figure 4 Language Stimuli

The cerebellum is involved in the preprocessing of sensory data, the integration ofvisual, auditory, vestibular and somatosensory input, and the acquisition and maintenance

of classical conditioning.22 Input to the cerebellar cortex is via mossy fibres and climbingfibres (Figure 5) Mossy fibres (via the parallel fibres) connect the pontine nuclei to the

Inferior Olive Climbing FibreMossy Fibres

Pontine nuclei

Purkinje cell

Output from the cerebellum Parallel Fibres

Conditioned Stimuli (CS)

Input to the cerebellum

Unconditioned Stimuli (UC)

Figure 5 The Cerebellum _

22 Classical conditioning is explained in Appendix C

19

Purkinje cells of the cerebellum and provide a graded response, with many mossy fibresneeded to cause one of the cerebellum’s Purkinje cells to fire The pontine nuclei carriesthe conditioned stimuli (CS) information Climbing fibres, originating at the inferiorolive, provide all-or-nothing firing, with one fibre firing a single Purkinje cell Theinferior olive carries the unconditioned stimuli (US) information (Figure 5).

Output from the cerebellum is controlled by the Purkinje cells which inhibit the firing

of the deep nuclei in the midbrain.23 In autism, there is a 41% loss of Purkinje cells andthe inferior olive is smaller than normal (Courchesne, 1988)

Lesions in the cerebellum are known to disrupt classical conditioning (Schmajuk,1997) Eyeblink response is considered a hallmark of cerebellar function; it is a classi-cally conditioned response, and when there is damage to the cerebellum it is abnormal.Because it signals damage, the presence of abnormal eye blink response is consideredevidence of cerebellar damage in autism (Belmonte & Carper, 1998) This cerebellardamage in autism prevents classically conditioned language learning from occurringnormally

Perceptual Conceptualization and Lexicalization

In the first three months of life infants are already capable of recognizing familiarvoices (PSHC, 2004) and of attending closely to the sound of an unfamiliar voice (Bo-wen, 2004) At three to six months the infant enjoys music and rhythm (Bowen, 2004)

_

23 The midbrain is explained in Appendix D

20

21and is capable of responding to changes in tone of voice (PSHC, 2004) These are func-tions of the right anterior temporal lobe (Figure 6).

Tonal Rhythm/

Lexicalization

Language Stimuli

Functional Definitions Functional

Definitions

Figure 6 Tonal Rhythm and Lexicalization

Basic-level sensory-motor conceptualization also develops using the general shapeand motor interaction to form the mental image of an object (Lakoff, 1987) The proper-ties that define the object for the child are not inherent to the object but are in the interac-tions the child has with the object (Lakoff, 1987); so for the young child, the object we

assign with a label of chair can be defined as something that is sat upon24 instead ofsolely in terms of having four legs, a seat and a back When initial definitions are formedthis way, the motor cortex is triggered (Figure 4) Basic-level perceptual categories will

be discussed in more depth in that section, and the motor cortex will be discussed in moredepth in the action verbs section

_

24 This also means that a chair that is never sat upon, perhaps because it is broken, might not be categorized

as a chair by the young child Likewise, a front stoop that is sat upon might become categorized as a chair

by the child.

21

At five to six months the infant starts babbling, imitating the tonal aspects of guage with inflection, a rising and falling of pitch and rhythm which makes it sound liketrue speech (PSHC, 2004).

lan-The tonal processing of utterances occurs before the processing individual words(Piaget, 1926) As a result, the child will respond to phrases such as “pat-a-cake” and

“wave bye-bye” (Bowen, 2004) Language at this point is being treated idiomatically bythe child, and the lack of syntax is not allowing for any substitutions For instance, the

child’s understanding of the phrase bye-bye daddy does not mean the child is capable of making a substitution and understanding the phrase bye-bye mommy Gardner (1983)

makes an important point: “Just because a child’s output of language starts as individualwords, and progresses to phrases, does not mean that the input of language is processingthe same way.25 After all, the child has no concept of words when he starts, and languageoutput is almost a year into the language learning process.”

The brain’s analysis of an utterance as possibly consisting of smaller units of meaning

is called lexicalization The process of lexicalization is accomplished by the brain through two means: sentence subtraction and ostensive definition Sentence subtraction occurs

when two almost identical utterances are compared by the brain, and the piece that differsbecomes a separate semantic unit (Piaget, 1926) These semantic units are still not always

as small as a word This sentence subtraction will also eventually lead to grammar

_

25 In this pre-word/pre-grammar stage, the goal of the brain is not to parse the sentence into individual words, but to identify the patterns that are occurring:

<Bill> <wants> <animal crackers>

<fruit flies> <like> <a banana>

The sentence may even remain unparsed, functioning as an idiom that cannot be divided without losing meaning: <time flies like an arrow>

12/19 Sentence added

23Ostensive definition occurs when an object is pointed to and an utterance consisting

of only a label is supplied (Markman, 1989) When young children hear these labels they

are predisposed to assume that the label refers to the whole object rather than to itsproperties (Markman, 1989) In autism, pointing is not understood (Frith, 1989) andostensive definitions may not occur Because ostensive definition aids in sentence sub-traction, the separation of utterances into words may also not occur.26

From approximately six months old to a year, and if the tonal recognition and ization have been working correctly, the child starts to respond to his name and to look atcommon objects when the names for these objects are spoken (PSHC, 2004) Betweenone and two years the spoken vocabulary increases to approximately 300 words (PSHC,2004), and during the second year the vocabulary increases to almost 1000 words(floridaspeech, 2002) There is no grammar present yet as there needs to be a criticalmass of words in the vocabulary before the language mechanism starts involving thegrammar areas of the brain.|

lexical-If there is early damage to the right anterior temporal lobe lexicalization will notdevelop correctly In a French study by Lalande, lexicalization problems were seen inilliterate adults, where they ran together words that are separate in the written language,and also divided up words that are single words in the written language (Piaget, 1926)

of the tone of voice, the autistic child does not “get” the ostensive definition and may not acquire language.

12/4 p 23: I don’t

understand the

“there-fore” here: “In

au-tism, pointing is not

understood (Frith,

1989), and ostensive

definitions, and

therefore the

separa-tion of utterances into

words, may not

occur.” Am I just not

Pragmatic problems in conversational turn-taking can also occur with damage to the rightanterior temporal lobe, as turn-taking requires attention to pauses in conversation, andinterpreting the meaning of those pauses Speech becomes excessive and rambling as aresult.|

In autism, if language does develop, lexicalization often does not occur at all, and the

entire sentence will remain undivided, with no attention paid to the ostensive definitions

or to pauses indicating separate semantic chunks Echolalia results, where the autisticchild uses entire sentences verbatim.27 Some autistic children have used chunks evenlarger than the sentence, sometimes paragraphs, or even short stories, repeating themverbatim in their attempt to communicate (Heffner, 2000)

In the adult, the right anterior temporal lobe specializes for the prosodic elements ofspeech: pitch, rhythm (duration), and stress (frequency, intensity, and timing) (Hooper,2003) Prosody performs a chunking function with speech, distinguishing compound

words from noun phrases (redcoat vs red coat, backward vs back ward , greenhouse vs green house) (Morgan, 2003), distinguishing some nouns and verbs (REcord vs.

reCORD),28 and declarative sentences from interrogative sentences (Hooper, 2003)

When the right anterior temporal lobe develops normally but then is damaged in an

adult, aprosodia, the inability to detect or use affect in speech, occurs; the emotional

content of speech is lost Speech has a flat affect and a monotonous intonation; stress on

words is indicated with amplitude changes rather than with pitch and duration changes

_

27 Seventy five percent of the autistic children who do acquire language are echolalic.

28 These words and not ambiguous when spoken so the chunking function is not performing any

disambigu-ation Examples requiring disambiguation such as the box is brown vs box the clothing are disambiguated

by another area of the brain.