Intro to AI - Clarkson Universityjsearlem/cs451/fa13/lectures/24.NLU.pdf · language A large amount of human knowledge is assumed Language is pattern based: phonemes are components

11-20-2013

Natural Language Understanding

Read: AIMA Chapters 22 & 23

HW#8, due Monday, 11/25

What are some of the most impressive

technologies in futuristic Science Fiction?

e.g. consider Star Trek

One is the “Universal Translator”

Science behind Watson

Three key capabilities

Natural Language Understanding

Hypothesis Generation

Evidence-based Learning

NLP is a discipline that aims to build computer systems that will be able to analyze, understand and generate human speech.

NLP subareas of research are:

Speech Recognition (speech analysis),

Speech Synthesis (speech generation), and

Natural Language Understanding (NLU).

Putting meaning to the words

Input might be speech or could be typed in

Holy grail of Artificial Intelligence problems

Georgetown University:

“The spirit is willing but the flesh is weak.”

English to Russian

Russian to English

“The vodka is good but the meat has spoiled.”

Consider the following conversation between

Mary and Tom:

Tom: “Who do you like tonight, Boston or LA?”

Mary: “Lakers. You?”

Tom: “Come on Mary, LA can’t handle Bird.”

Mary: “I’ve got a five that says Magic will shut him down.”

Problem: English sentences are incomplete descriptions of the info they are intended to convey.

I called Linda to ask her to the movies. She said she’d love to go.

but…

speakers can be vague or precise; can leave out

details that the hearer is expected to know

Problem: The same expression means different things in different contexes.

Where’s the water?

but…

can communicate about an infinite world with a

finite number of symbols

Problem: New words, expressions and meanings evolve.

I’ll fax it to you.

In the 1600s, St. Paul’s cathedral was said to be “amusing, awful and artificial.”

“Selfie” named by Oxford dictionaries as word of the year 2013.

but…

languages can evolve as experiences change

Problem: There are a lot of ways of saying the same thing.

Mary was born on March 27th.

Mary’s birthday is March 27th.

but…

when you know a lot, facts imply each other

Speech recognition is the process of converting spoken language to written text or some similar form.

Speech synthesis is the process of converting the text into spoken language.

Natural Language Understanding (NLU) is a process of analysis of recognized words and transforming them into data meaningful to computer.

Other words, NLU is a computer based system that “understands” human language.

NLU is used in combination with speech recognition.

■Three major issues involved in understanding language

A large amount of human knowledge is assumed

Language is pattern based: phonemes are components of words and words make phrases and sentences.

Language acts are the product of agents, either human or computer

■Terry Winograd’s SHRDLU(Winograd 1972)

Early AI programs made progress by restricting the focus to microworld

SHRDLU could respond to English queries What is sitting on the red block? What shape is the blue block on the table? Place the green pyramid on the red brick.

Language is a complicated phenomenon, involving processes as varied as the recognition of sounds or printed letters, syntactic parsing, high-level semantic inferences, and even the communication of emotional content through rhythm and inflection.

To manage this complexity, linguists have defined different levels of analysis for natural language.

NLP Pipeline

Phonetic/Phonological Analysis

Morphological analysis

OCR/Tokenization

Syntactic analysis

Semantic Interpretation

Discourse Processing

speech text

Phonology

Syntax

Semantics

Pragmatics &

World Knowledge

Prosody: dealing with inflection, stress, pitch, timing

Phonology: examining sounds combined to form language, important for speech recognition and generation

Morphology: concerned with morphemes making up words. These include rules governing the formation of words. Important in determining the role of a word in a sentence in most languages in the world.

Morphological anomaly: “The computer eated an apple.”

Syntax: dealing with rules for combining words into legal phrases and sentences

Syntactic anomaly:

“The computer ate apple.”

“An the ate apple computer.”

Semantics: considers meaning of words, phrases,

and sentences also ways in which meaning is conveyed in natural language

Semantic anomaly: “The computer ate an apple.”

Pragmatics: dealing with ways in which language is used and its effects on the listener

“Do you know the time?”

Pragmatic anomaly: “Next year, all taxes will disappear.”

World knowledge: includes knowledge of physical world, is essential to understand the full meaning of a text

“The pen is in the box.” versus

“The box is in the pen.”

Lazy Contented Cats Sleep Peacefully

Sleep Furiously Colorless Green Ideas

Squad helps dog bite victim. Helicopter powered by human flies. I ate spaghetti with meatballs.

… with salad. … with abandon. … with a fork … with a friend.

Ambiguity can be lexical, syntactic, semantic, or referential

S

NP VP

V NP PP

Art N PP

John

saw

a

with a telescope

in a park boy

S

NP VP

V NP

PP Art N PP

John

saw

a with a telescope in a park boy

S

NP VP

V NP

PP

Art N PP

John

saw

a

with a telescope

in a park boy

John saw a boy in a park

with a telescope.

S

NP VP

V NP PP

Art N PP

John

saw

a

with a telescope

in a park boy

S

NP VP

V NP

PP Art N PP

John

saw

a with a dog in a park boy

S

NP VP

V NP

PP

Art N PP

John

saw

a

with a statue

in a park boy

Identify all noun phrases that refer to the same entity

John Simon, Chief Financial Officer of Prime Corp.

since 1986, saw his pay jump 20%, to $1.3 million,

as the 37-year-old also became the financial-

services company’s president...

Best results: F-measure of 70.4 (MUC-6) and 63.4 (MUC-7) [Ng & Cardie, 2002]

Advances in software and hardware create NLP needs for information retrieval (web), machine translation, spelling and grammar checking, speech recognition and synthesis.

Stochastic and symbolic methods combine for real world applications.

Speech Processing

A Voice Interface

Some Applications

■Information Retrieval: Web search (uni- or multi-lingual)

■Query Answering/ Dialogue, e.g.,

■Report Generation: English/French weather report

■Foreign Language Training: Spanish/Arabic tutorial systems

for military linguists

■Machine Translation : on Yahoo

Chat-80

Babelfish

Speech

Synthesizer

Speech

Recognizer

Natural

Language

Generator

“I would like to fly to

Seattle tomorrow.”

“When would you

like to leave?”

Natural

Language

Understanding

Dialog

Manager

Domain

Knowledge

What is speech? Vibrations of vocal cords creates sound “ahh” Mouth, throat, tongue, lips shape sound

English speech 40 phonemes; 24 consonants, 16 vowels

Sounds transmit “language”

Speech does not equal written language

"I told him to go back where he came from, but he wouldn't listen."

Tell which person it is (voice print)

Could also be important for monitoring meetings, determining speaker

Primarily identifying words

Improving all the time

Commercial systems:

IBM ViaVoice, Dragon Dictate, ...

Speaker dependent/independent Parametric patterns are sensitive to speaker With training (dependent) can get better

Vocabulary Some have 50,000+ words

Isolated word vs. continuous speech Continuous: where words stop & begin Typically a pattern match, no context used

Did you vs. Didja

Java Speech SDK FreeTTS 1.1.1

http://freetts.sourceforge.net/docs/index.php

IBM JavaBeans for speech Visual/Real Basic speech SDK OS capabilities (speech recognition and

synthesis built in to OS) (TextEdit) VoiceXML

tool automate the construction of NLP systems

avoid the need for large linguistic knowledge bases

portability move to new domain quickly

reduce the need for expertise in computational linguistics

robustness handle ungrammatical or unexpected text

missing domain knowledge

Statistical methods have transformed the field of NLP

Very good performance on increasing numbers/types of problems in NLP

Thus far, the most successful statistical and ML algorithms are supervised learning algorithms

Require large amounts of training data that has been annotated with the “correct” answers

Corpus annotation bottleneck

Japanese, Chinese, Thai, ...: no spaces between words

Combining simple statistics from unsegmented Japanese

newswire yields results rivaling grammar-based approaches.

[Ando & Lee 2000, 2003]

Translating from one language to another is challenging even to human translators.

e.g. signs translated into English by a person:

Utmost of chicken with smashed pot. (restaurant in Greece)

Nervous meatballs (restaurant in Bulgaria)

The nuns harbor all diseases and have no respect for religion. (Swiss nunnery hospital)

All the water has been passed by the manager. (German hotel)

Morphological analysis

Syntactic analysis

Semantic Interpretation

Interlingua

input analysis generation

Morphological synthesis

Syntactic realization

Lexical selection

output

Doesn’t work well enough yet

ACL 2013 8th Workshop on Statistical Machine Translation

MT Summit 2013

Machine Translation without the Translation, Chronicle of Higher Education

NLPCS 2013: 10th International Workshop on Natural Language Processing and Cognitive Science

Workshop on Natural Language Processing and Automated Reasoning