from syntax to semantics how to get from form to meaning in two different ways
TRANSCRIPT
From Syntax to From Syntax to SemanticsSemantics
How to get from Form to Meaning How to get from Form to Meaning in Two different waysin Two different ways
What is meaning?What is meaning?
Connection (grounding) in something Connection (grounding) in something outside itselfoutside itself
Mental concept (ideas)Mental concept (ideas) Objects and events in the world Objects and events in the world
(true/false)(true/false) Some combination of the aboveSome combination of the above Ultimately – the success of the program in Ultimately – the success of the program in
which it is embeddedwhich it is embedded
Principle of CompositionalityPrinciple of Compositionality
The meaning of the whole is derived from The meaning of the whole is derived from the meaning of the parts and the manner the meaning of the parts and the manner of their combinationof their combination
{John, kiss, Sally} {John, kiss, Sally} John kissed Sally.John kissed Sally. Sally kissed John.Sally kissed John.
Semantics -- For our purposesSemantics -- For our purposes
Formal representational language that Formal representational language that represents the “manner of combination”represents the “manner of combination”
Lexicon that connects lexical items with Lexicon that connects lexical items with some externally grounded object, the some externally grounded object, the “meaning of the parts”“meaning of the parts”
Two approachesTwo approaches
Logical Logical Language of formal logicLanguage of formal logic Model (set) theoretic groundingModel (set) theoretic grounding
InterlingualInterlingual Specially-developed InterLingual (IL) Specially-developed InterLingual (IL)
RepresentationRepresentation Ontology to represent word meaningOntology to represent word meaning
To some extent complementaryTo some extent complementary
Logical approachLogical approach
Predicate calculus and model theory PLUSPredicate calculus and model theory PLUS Extra stuff to handle some of the complexities of Extra stuff to handle some of the complexities of
natural language, such asnatural language, such as (Scope) Every man loves a woman.(Scope) Every man loves a woman. (Generics) Dogs have four legs.(Generics) Dogs have four legs. (Specificity) John wants to marry a Norwegian.(Specificity) John wants to marry a Norwegian. (Intension) What if all bald men are tall?(Intension) What if all bald men are tall? (Roles) The temperature is ninety and rising.(Roles) The temperature is ninety and rising.
Logical approach – Logical approach – λ calculusλ calculus
Key idea: semantic construction parallels Key idea: semantic construction parallels syntactic constructionsyntactic construction
John = John = john’john’ sleep = sleep = sleep’sleep’ John is sleeping = John is sleeping = sleep’(john’)sleep’(john’) sleep = sleep = λx[sleep’(x)]λx[sleep’(x)] John is sleeping = John is sleeping = λx[sleep’(x)](john’)λx[sleep’(x)](john’) Lambda conversion = Lambda conversion = sleep’(john’)sleep’(john’)
Logical approach – possible Logical approach – possible worldsworlds
Instead of one model – many modelsInstead of one model – many models Each model is a “possible world” – one is Each model is a “possible world” – one is
designated as “real”designated as “real” Temporal logicTemporal logic Modal logicModal logic Intensional logicIntensional logic
IL approachIL approach
Developed in the context of Machine TranslationDeveloped in the context of Machine Translation Interested in word sense disambiguationInterested in word sense disambiguation
The pig is in the pen.The pig is in the pen. The ink is in the pen.The ink is in the pen.
Non-literal language: metonymy/metaphorNon-literal language: metonymy/metaphor ““The White House reported today that …”The White House reported today that …” ““The business opened its doors in 1928.”The business opened its doors in 1928.”
Inferencing for translation mismatchesInferencing for translation mismatches
IL approach IL approach
An Ontology, a language-independent An Ontology, a language-independent classification of objects, event, relationsclassification of objects, event, relations
A Semantic Lexicon, which connects A Semantic Lexicon, which connects lexical items to nodes (concepts) in the lexical items to nodes (concepts) in the ontologyontology
An analyzer that constructs IL An analyzer that constructs IL representations and selects (an?) representations and selects (an?) appropriate oneappropriate one
IL approach – OntologyIL approach – Ontology
A classification tree in which mother node A classification tree in which mother node contains all below it, and daughter nodes are contains all below it, and daughter nodes are distinct (is-a links)distinct (is-a links)
Complications: expandable to a lattice, with non-Complications: expandable to a lattice, with non-exclusive daughter nodesexclusive daughter nodes
Inheritable features and relations (now looks Inheritable features and relations (now looks more like a dictionary)more like a dictionary)
““Instances” can hang from bottom nodes Instances” can hang from bottom nodes (providing grounding)(providing grounding)
Semantic lexiconSemantic lexicon
Provides a syntactic context for the Provides a syntactic context for the appearance of the lexical itemappearance of the lexical item
Provides a mapping for the lexical item to Provides a mapping for the lexical item to a node in the ontologya node in the ontology
Or more complex associationsOr more complex associations Also providing connections from syntactic Also providing connections from syntactic
context to semantic roles context to semantic roles And constraints on these rolesAnd constraints on these roles
Deriving basic semantic dependency (a toy example)
Input: John makes tools
Syntactic Analysis:cat verbtense presentsubject
root johncat noun-proper
object root toolcat nounnumberplural
Relevant parts of the (appropriate sense of the lexical entry for make)
make-v1syn-struc
root makecat vsubj root $var1
cat nobject root $var2
cat nsem-struc
manufacturing-activityagent ^$var1theme ^$var2
Relevant Extract from the Specification of the Ontological Concept Used to Describe the Appropriate Meaning of make:
manufacturing-activity...
agent humantheme artifact
…
John-n1syn-struc
root johncat noun-proper
sem-struchuman
name johngender male
tool-n1syn-struc
root toolcat n
sem-structool
Relevant parts of the (appropriate senses of the)lexicon entries for John and tool
The basic semantic dependency component of the TMR forJohn makes tools is as follows:
…
manufacturing-activity-7
agent human-3theme set-1
element toolcardinality > 1
…
try-v3syn-struc
root trycat vsubj root $var1
cat nxcomp root $var2
cat vform OR infinitive gerund
sem-strucset-1 element-type refsem-1
cardinality >=1refsem-1 sem event
agent ^$var1effect refsem-2
modalitymodality-type epiteucticmodality-scope refsem-2modality-value < 1
refsem-2 value ^$var2sem event
Constructing an IL Constructing an IL representationrepresentation
For each syntactic analysisFor each syntactic analysis Access all semantic mappings and Access all semantic mappings and
contexts for each lexical itemcontexts for each lexical item Create all possible semantic Create all possible semantic
representationsrepresentations Test them for coherency of structure and Test them for coherency of structure and
contentcontent
REQUEST-INFO-130 THEME DEVELOP-2601.PURPOSE DEVELOP-2601.REASON TEXT-POINTER why INSTANCE-OF REQUEST-INFO
DEVELOP-2601THEME SET-2555AGENT NATION-97PHASE CONTINUOUS
TIME FIND-ANCHOR-TIME INSTANCE-OF DEVELOP
TEXT-POINTER developing
NATION-97HAS-NAME Iraq
INSTANCE-OF NATIONTEXT-POINTER Iraq
SET-2555 ELEMENT-TYPE WEAPONCARDINALITY > 1
INSTRUMENT-OF KILL-1864 THEME-OF DEVELOP-2601 INSTANCE-OF WEAPON
TEXT-POINTER weapons
KILL-1864 THEME SET-2556 INSTRUMENT SET-2555 INSTANCE-OF KILL
TEXT-POINTER destruction
SET-2556 THEME-OF KILL-1225 ELEMENT-TYPE HUMAN
CARDINALITY > 100 INSTANCE-OF HUMAN
TEXT-POINTER mass
“Why is Iraq developing weapons of mass destruction?”
Concluding questionConcluding question
Is all this really necessary?Is all this really necessary? Do we need it to do – Machine Do we need it to do – Machine
Translation, IR, IE, Q/A, summarization?Translation, IR, IE, Q/A, summarization? Can we “ground” the symbols of language Can we “ground” the symbols of language
without a special representation of the without a special representation of the “meaning”?“meaning”?
Word sense disambiguationWord sense disambiguation
Constraint checking – making sure the Constraint checking – making sure the constraints imposed on context are metconstraints imposed on context are met
Graph traversal – is-a links are Graph traversal – is-a links are inexpensiveinexpensive
Other links are more expensiveOther links are more expensive The “cheapest” structure is the most The “cheapest” structure is the most
coherentcoherent Hunter-gatherer processingHunter-gatherer processing