MARINA SANTINI PROGRAM: COMPUTATIONAL LINGUISTICS AND LANGUAGE TECHNOLOGY DEPT OF LINGUISTICS AND PHILOLOGY UPPSALA UNIVERSITY, SWEDEN 14 NOV 2013 Semantic Analysis in Language Technology Lecture 2: From Semantics to Semantic-Oriented Applications Course Website : http://stp.lingfil.uu.se/~santinim/sais/sais_fall2013.htm
25
Embed
Lecture 2: From Semantics To Semantic-Oriented Applications
From the "Natural Language Processing" LinkedIn group:
John Kontos, Professor of Artificial Intelligence
I wonder whether translating into formal logic is nothing more than transliteration which simply isolates the part of the text that can be reasoned upon using the simple inference mechanism of formal logic. The real problem I think lies with the part of text that CANNOT be translated one the one hand and the one that changes its meaning due to civilization advances. My own proposal is to leave NL text alone and try building inference mechanisms for the UNTRANSLATED text depending on the task requirements. All the best John"
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MARINA SANTINI
P R O G R A M : C O M P U TAT I O N A L L I N G U I S T I C S A N D L A N G U A G E T E C H N O L O G Y
D E P T O F L I N G U I S T I C S A N D P H I L O L O GY
UPPSALA UNIVERSITY, SWEDEN
14 NOV 2013
Semantic Analysis in Language Technology
Lecture 2: From Semantics to Semantic-Oriented Applications
The present: Computational Semantics & Semantic-Oriented Applications
The future: Actionable Intelligence
Lecture 2: From Semantics to Applications
3
Aristotelian Logic
The fundamental assumption behind the theory is that propositions are composed of two terms – hence the name "two-term theory" or "term logic" – and that the reasoning process is in turn built from propositions: Aristotle distinguishes singular terms such as Socrates and general terms such
as Greeks. Aristotle further distinguishes (a) terms that could be the subject of predication, and (b) terms that could be predicated of others by the use of the copula ("is a").
A proposition consists of two terms, in which one term (the "predicate") is "affirmed" or "denied" of the other (the "subject"), and which is capable of truth or falsity. Socrates is a man Socrates is not immortal
The syllogism is an inference in which one proposition (the "conclusion") follows of necessity from two others (the "premises"). Socrates is a man, all men are mortal, therefore Socrates is mortal = new knowledge (inferential knowledge)
Lecture 2: From Semantics to Applications
4
Syllogistic fallacies
People often make mistakes when reasoning syllogistically and mathematically with natural language:
• A=B• B=C• A=C
some cats (A) are black things (B), some black things (B) are televisions (C), it does not follow from the parameters that some cats (A) are
televisions (C). Existential fallacy (use of quantifiers)
The existential fallacy, or existential instantiation, is a formal fallacy: "Everyone in the room is pretty and smart". It does not imply that there is a pretty, smart person in the room, because it does not state that there is a person in the room.
In 1879 Frege published his Begriffsschrift (Concept Script). This introduced a calculus, a method of representing statements by the use of quantifiers and variables.
The two fundamental kinds of quantification in predicate logic are universal quantification and existential quantification. The traditional symbol for the universal quantifier "all" is "∀", an inverted letter "A", and for the existential quantifier "exists" is "∃", a rotated letter "E".
A predicate takes an entity or entities in the domain of discourse as input and outputs either True or False.
Consider the two sentences "Socrates is a philosopher" and "Plato is a philosopher".
In propositional logic, these sentences are viewed as being unrelated and are denoted, for example, by p and q. However, the predicate "is a philosopher" occurs in both sentences which have a common structure of "a is a philosopher". The variable a is instantiated as "Socrates" in first sentence and is instantiated as "Plato" in the second sentence
"There exist a such that a is a philosopher" .
Predicates can be also compared. Ex "if a is a philosopher, then a is a scholar". This formula is a conditional statement with "a is a philosopher" as hypothesis and "a is a scholar" as conclusion.
The truth of this formula depends on which object is denoted by a, and on the interpretations of the predicates "is a philosopher" and "is a scholar".
Variables can be quantified over. "For every a, if a is a philosopher, then a is a scholar". The universal quantifier "for every" in this sentence expresses the idea that the claim "if a is a philosopher, then a is a scholar" holds for all choices of a.
In linguistics, formal semantics seeks to understand linguistic meaning by constructing precise mathematical models of the principles that speakers use to define relations between expressions in a natural language and the world which supports meaningful discourse.
The mathematical tools used are the confluence of formal logic and formal language theory, especially lambda calculus.
Linguists rarely employed formal semantics until Richard Montague showed how English (or any natural language) could be treated like a formal language. His contribution to linguistic semantics, which is now known as Montague grammar, was the basis for further developments.
Lecture 2: From Semantics to Applications
11
Translating Natural Language to Formal Language by:
Lamba calculus: is a formal system in mathematical
logic and computer science for expressing computation based on function abstraction and application via variable binding and substitution.
(Cf also J&M: 593)
Prolog: Prolog is a general purpose logic
programming language associated with artificial ntelligence and computational linguistics.
Prolog has its roots in first-order logic, a formal logic, and unlike many other programming languages, Prolog is declarative: the program logic is expressed in terms of relations, represented as facts and rules.
Top-down rule-based systems
Lecture 2: From Semantics to Applications
12
Syntax-based compositionality of meaning
Lecture 2: From Semantics to Applications
13
Stumbling block: meaning is not always compositional…
Lecture 2: From Semantics to Applications
14
Multi-Word Expressions
MWEs (a.k.a multiword units or MUs) are lexical units encompassing a wide range of linguistic phenomena, such as: idioms (e.g. kick the bucket = to die), collocations (e.g. cream tea = a small meal eaten in
Britain, with small cakes and tea), regular compounds (cosmetic surgery), graphically unstable compounds (e.g. self-contained <>
self contained <> selfcontained - all graphical variants have huge number of hits in Google),
light verbs (e.g. do a revision vs. revise), lexical bundles (e.g. in my opinion), etc.
Lecture 2: From Semantics to Applications
15
Stumbling Block: Ambiguity
Lexical ambiguity: ex polisemy Ex: bank
Referential ambiguity: ex anaphoric ambiguity… it was funded by a tycoon
Scopal ambiguity: I can’t find a piece of paper
(a particular piece of paper or any piece of paper? Existential or universal quantifier "∀”or "∃“?)
Lecture 2: From Semantics to Applications
16
Computational Semantics (wikipedia)
Computational semantics is the study of how to automate the process of constructing and reasoning with meaning representations of natural language expressions. It consequently plays an important role in natural language processing and computational linguistics.
Some traditional topics of interest are: construction of meaning representations, semantic underspecification, anaphora resolution, presupposition projection, quantifier scope resolution.
Methods employed usually draw from formal semantics or statistical semantics. Computational semantics has points of contact with the areas of lexical semantics (word sense disambiguation and semantic role labeling), discourse semantics, knowledge representation and automated reasoning …
Lecture 2: From Semantics to Applications
17
What is Semantics? ---- What is LT?
Students’ intuition about semantics:
1. Meaning of language (words, phrases, etc.)
2. Break down complex meaning into simpler blocks of meaning
3. Content understanding4. Disambiguation5. Understanding a phrase6. Understanding the
meaning of phrases depending on different contexts
7. Meaning and connotation
Language technology is often called human language technology (HLT) or natural language processing (NLP) and consists of computational linguistics (or CL) and speech technology as its core but includes also many application oriented aspects of them. Language technology is closely connected to computer science and general linguistics. (wikipedia)
Must add:•Statistics•Machine learning
Lecture 2: From Semantics to Applications
18
What shall we keep from the past?
Computational semantics must be….
Lecture 2: From Semantics to Applications
19
Computational semantics must address open issues:
Ambiguity Overcome compositionality Etc.
Lecture 2: From Semantics to Applications
20
Our definition of semantics for LT must include:
Continuity with the past approaches Must be computationally tractable1. Meaning of language
(words, phrases, etc.)2. Break down complex
meaning into simpler blocks of meaning
3. Content understanding4. Disambiguation5. Understanding a phrase6. Understanding the meaning
of phrases depending on different contexts
7. Meaning and connotation
More advanced than past systems:
Must address ambiguity Must address non
compositional meaningAbove all, must tackle new media.In less than 50 years, new media (internet, web, social networks) have completely scrambled ”traditional” semantics and human communication by creating :
•New meanings (sentiment, opionion, etc)•New language (unconvetional texts and syntax and many sublanguages, like tweets, FB posts, etc.)•Big amounts of wild data
Semantics for Language Technolgy must now take also these aspects into account.
Lecture 2: From Semantics to Applications
21
In conclusion
More than creating a ”understanding system”, currently the stress in how to automatically extract meaningful and actionable information depending on specifc tasks….
Lecture 2: From Semantics to Applications
22
Visual Insight into big data around us…
Big Data Video: http://youtu.be/qqfeUUjAIyQ (2:21 min)
Sentiment Analysis’s purpose: detect and extract emotions, attitudes, opnions from text… People behaviour and choices (politics, products, reactions) are driven by sentiment rather than ”sensibility”
(Sense and Sensibility by J. Austin well describe these two opposite behaviours)
A basic ML algorithm underlying many (but not all) applications detecting sentiments: Daniel Jurafsky, Coursera, NLP – Stanford University (video, 13 min)