PA153 Natural Language Processing
02 - Semantics I (lexical meaning and its representation)
Karel Pala, Zuzana Nevěřilová
NLP Centre, Fl MUNI, Brno
Sep 25, 2019
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 1/23
Q Lexical Meaning
Q Meaning in Context
Q Lexical Meanings in NLP Semantic Classes
Q Conclusion, Take Home
Karel Pala, Zuzana Nevěřilová
PA153 Natural Language Processing
Lexical Meaning
(cs: lexikální význam): meaning of a word in isolation [Oxford Dictionaries, 2013]
• regardless of meaning of the sentence the word is part of
9 regardless of grammar categories
other types of meaning: grammatical meaning, word meaning, sentence meaning
• buy - bought
• image - picture
9 The old professor runs to catch the bus. The cheetah runs to catch the prey.
02 - Semantics I 3
PA153 Natural Language Processing Lexical Meaning
Lexical Meaning
Lexical Meaning
(cs: lexikální význam): meaning of a word in isolation [Oxford Dictionaries, 2013]
• regardless of meaning of the sentence the word is part of
• regardless of grammar categories
other types of meaning: grammatical meaning, word meaning, sentence meaning
• buy - bought
• image - picture
• The old professor runs to catch the bus. The cheetah runs to catch the prey.
buy and bought - the same lexical meaning, different grammatical meaning image and picture - different words with the same lexical (and grammatical) meaning
to run - the same meaning, different activity
Lexical Form and Lexical Unit
Lexical Unit (cs: lexikální jednotka, LU) [Ziková, 2003]: o represented by a lexical form
• asociated with a particular lexical meaning
• has grammatical properties (e.g. transitive verb)
o can have pragmatic properties (e.g. / each time references to some other person)
o LU with the same meaning but different form pause synonyms (e.g. beautiful, lovely)
o LU with the same form but different meaning pause homonyms (e.g. bark)
02 - Semantics I
Where is information about lexical meaning
Dictionary/lexicon/lexical database - storage of lexical units Dictionaries:
• general, Language for General Purpose (LGP) also defining/explanatory (with definitions)
• bilingual (contains translations of LUs)
• etymological 9 encyclopedic
• reverse
• rhyming
• single-field (contains domain terminology)
• historical
• ...
for NLP, machine readable dictionaries are used
□
Karel Pala, Zuzana Nevěřilová	PA153 Natural Language Processing		02 - Semantics 1 5/23
Anatomy of a dictionary entry
"| determine-definition and synonyms 7
Q®OQrs _
2 VERB Promroatio" |   .'dťl3:{r}miri/  I Wtaid Farrre~| ^
8
Contribute to our Open Dictionary
[transitive] [often passive] to control what something will be Our prices are determined by the market.
genetically/cjlturally/biologically determined: Sne claims thatmc£ human behaviour is socially determined.
Wl Synonyms and related words
To limit or control something or someone:
draw a line in the sand, limit, control.
Explore Thesaurus
1: headword 2: definition
3: grammatical category 4-5: pronunciation (sound & IPA)
6
7
8
inflection frequency numbered senses
[ inTran s it i VErrRAN s it iVE] to officially decide something determine whethertwhy^who: it is for the court to determine whether she is guiity.
El Synonyms and related words To make a decision: decide, determine, arrive at...
Explore Thesaurus
https://www.macmillandictionary.com/ learn/dictionary-entry.html
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 6/23
00 OJ
(J) 1—I o
OJ
PA153 Natural Language Processing
Anatomy of a dictionary entry
Anatomy of a dictionary entry
1 determine *fiiii
o©oo.
**7
6
Cuplo-e Th?;auiu!
1: headword 2: definition
3: grammatical category
4-5: pronunciation (sound
& IPA)
6: inflection
7: frequency
8: numbered senses
Czech Remark on Czech Dictionaries: SSC SSJC: V SSJC není žádná odvozená forma, ale v SSC je. Pro jiná slova je v SSC mnohem více odvozených forem: květ, květen, květena, květák, květenství, květina, květináč, květinářka, květinářství
Collocation as a Dictionary Entry
A language user has available ...  a large number of semi-preconstructed phrases that constitute single choices.
(Sinclair 1991: 110)
New York, ad hoc, foreign language, second language, to save time
special collocation dictionaries: Oxford Collocations Dictionary, Macmillan Collocations Dictionary
In NLP, the term multiword expresion (MWE) is used when the notion of MWE is useful:
• phrasemes/idioms (imagine translating them)
• the meaning of the components is different/unclear (MWEs as words with spaces)
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 7/23
Dictionary Definitions
• extensional (denotative) definitions
► demonstrative definitions (by pointing)
► definition by enumeration (e.g. Baltic states are: Estonia, Latvia, and Lithuania)
► definition by subclass (e.g. "flower" means rose, lily, daisy, and the like)
• intensional (connotative) definitions
► synonymous definition (e.g. "physician" means "doctor")
► etymological definition (e.g. the word "capital" comes from Latin word "caput" meaning "head")
► operational definition (e.g. "brain activity" happens iff an electroencephalograph shows oscillations)
► definition by genus and differentia
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 8/23
Definition by Genus and Differentia
a triangle: A plane figure that has 3 straight bounding sides.
a quadrilateral: A plane figure that has 4 straight bounding sides. hyperonymy
Example: check a Wikipedia definition
□
Karel Pala, Zuzana Nevěřilová
PA153 Natural Language Processing
02 - Semantics I 9/23
PA153 Natural Language Processing
Definition by Genus and Differentia
Definition by Genus and Differentia
a triangle: A plane figure that has 3 straight bounding sides.
a quadrilateral: A plane figure that has 4 straight bounding sides. hyperonymy
Example: check a Wikipedia definition
dictionary entries assume at least some knowledge of the language (e.g. English GPL is around 2000 words)
for NLP dictionaries for humans are not fully suitable
Meaning in Context
Lexical meaning is not always enough (in fact, it is not enough most of the time)
=4> know the context
Word Sense Disambiguation (cs: lexikální desambiguace) function: (1/1/, c) —> s
• w G W - set of words
• c G C - set of contexts
• s G S - set of meanings
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 10/23
Word Sense Disambiguation
All algorithms rely on a lexical database with discrete meanings Different dictionaries have different granularity of meanings. Meanings are not fully discrete.
(interaction •valium
\
laddict
irelapse
)abuse #acldiction •rehab^ /
►store
\
locator
\
idealer
treatment
►therapy
(cartel
►center
\
trafficking
►testing
Karel Pala, Zuzana Nevěřilová
PA153 Natural Language Processing
02 - Semantics I
11 / 23
Word Sense Disambiguation is dead
long live the ...
Word Sense Discrimination
riviŕrc
irrigation
match
[Véronis, 2004]
□
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 12/23
Componential analysis (komponentová analýza)
= description of meaning using a (small) set of semantic features that are either present, or not present, or irrelevant.
• man = +HUMAN +ADULT +MALE
• woman = +HUMAN +ADULT -MALE
• boy = +HUMAN -ADULT +MALE
• toddler = +HUMAN -ADULT ±MALE [Katz and Fodor, 1963] a [Bierwisch, 1971]
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 13/23
Componential analysis I
Later related with semanti
Substantives
Relational Substantives
Determiners
Quantifiers
Evaluators
Descriptors
Mental predicates
Speech
Actions, Events, Movement Existence, Possession Life and Death Time
Space
Logical Concepts Intensifier, Augmentor Similarity
c primes and natural semantic metalanguage:
I, YOU, SOMEONE, PEOPLE, SOMETHING/THING, BOD\ KIND, PART
THIS, THE SAME, OTHER ELSE ANOTHER
ONE, TWO, SOME, ALL, MUCH/MANY, LITTLE/FEW
GOOD, BAD
BIG, SMALL
THINK, KNOW, WANT, DON'T WANT, FEEL, SEE, HEAR SAY, WORDS, TRUE DO, HAPPEN, MOVE
BE (SOMEWHERE), THERE IS, BE (SOMEONE/SOMETHI LIVE, DIE
WHEN/TIME, NOW, BEFORE, AFTER, A LONG TIME, I-FOR SOME TIME, MOMENT
WHERE/PLACE,  HERE, ABOVE,  BELOW,  FAR, NEAR,
TOUCH (CONTACT)
NOT, MAYBE, CAN, BECAUSE, IF
VERY, MORE
LIKE/AS/WAY
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 14/23
Semantic Classes
= group words that share a semantic feature
van - truck - motor vehicle - self-propelled vehicle - wheeled vehicle -vehicle - transport - instrumentation - artifact - whole - object - physical entity - entity
taxonomy, hierarchy, tree structure
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 15/23
The Tree of Porphyry
Supreme genus: Differentiae: Subordinate genera: Differentiae: Subordinate genera: Differentiae: Proximate genera: Differentiae: Species:
Substance
immaterial Body Spirit
sensitive insensitive
Animal Plant
irrational Human Beast
InttividitatM
Socrates     Plato     Aristotle etc.
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 16/23
Semantic Networks, Inference
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 17/23
Semantic Networks
WordNet (Princeton WordNet, PWN) - lexical network
• originally a psychology project (G. A. Miller, od r. 1985)
• usable by humans and computers (NLP) [Fellbaum, 1998]
• basic unit: synonymical set (synset), cs: synonymická řada
• relations between synsets:
► hyperonymy/hyponymy: truck - van
► holonymy/meronymy (part of, member of): car - brake
► troponymy: whisper - speak
► near-antonym: day - night
► derivation: wide - width
• POS: nouns, adjectives, verbs, adverbs 9 each POS organized in a different way
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 18/23
Word Net
size: PWN (117K synsets)
follow-up projects: EuroWordNet (en, nl, it, es, de, fr, es, et)
• ILI - InterLingual Index
• Top Ontology (63 kategorif)
o Base Concepts
BalkaNet: bg, es, ro, gr, sr, tr Global WordNet Association (GWA) Czech W.: 28K synsets
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 19/23
Ontologies
Lexical network - lexical knowledge Ontology - knowledge
Ontology = explicit specification of shared conceptualization
• domain ontologies
• general o. SUMO/MILO (Suggested Upper Merged Ontology, Mid-Level Ontology)
• common sense o. ConceptNet
Karel Pala, Zuzana Nevěřilová
PA153 Natural Language Processing
02 - Semantics I
Conclusion, Take Home
lexical meaning description:
• human friendly: dictionaries
NLP friendly: semantic primes, semantic networks, ontologies
sense disambiguation
human friendly: numbered senses in dictionaries
• NLP friendly: measures, vectors ...
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 21/23
References
Bierwisch, M. (1971).
On classifying semantic features.
In M. Bierwisch, K. E. H., editor, Progress in Linguistics, pages 27-50 Mouton.
Fellbaum, C. (1998).
WordNet: An Electronic Lexical Database (Language, Speech, and Communication ). The MIT Press. Published: Hardcover.
Katz, J. and Fodor, J. (1963). The structure of a semantic theory. Language, (39):170-210.
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 22/23
References
Oxford Dictionaries (2013).
lexical meaning. Oxford Dictionaries.
online.
http://oxforddictionaries.com/definition/english/ lexical-meaning (accessed October 03, 2013).
Véronis, J. (2004).
Hyperlex: Lexical cartography for information retrieval.
In Computer Speech and Language: Special Issue on Word Sense
Disambiguation, page 23.
Ziková, M. (2003).
Současný český jazyk: Tvoření slov.
online.
http://www.phi1.muni.cz/cest/lide/zikova/CJA009_l.rtf (accessed October 03, 2013).
Karel Pala, Zuzana Nevěřilová		PA153 Natural Language Processing	02 - Semantics 1 23/23