Course overview Introduction to summarization Lecture 1

advertisement
Course overview
Introduction to summarization
Lecture 1

Instructor: Ani Nenkova
–
–

505 Levine, nenkova@seas.upenn.edu
Office hours: Tuesdays 3:15—4:15 or by
appointment
TA: Annie Louis
–
lannie@seas.upenn.edu
Textbook

No required text
–

Recommended
–

Slides/lecture notes and handouts will be given in class
Speech and Language Processing (second edition, 2007,
Prentice-Hall), by Daniel Jurafsky and James Martin
Also see
–
–
Christopher Manning and Hinrich Schutze, “Foundations of
statistical natural language processing”
Advances in Automatic Text Summarization
Edited by Inderjeet Mani and Mark T. Maybury
Grading

5 homeworks (65%)
–
–



One will be a literature overview assignment
One will be at the end of the semester, instead of
a final
You are encouraged to form teams for the
homework (programming) assignments, but
all write-ups should be individual
Midterm (20%)
Class participation (15%)
–
“Submit” 5 questions each week
Late submission policy

5 late days for the semester
–

Can be used for any assignment with no penalty
Late submissions after “late days” have been
used up will not be graded
What you will learn

A lot about summarization and natural
language techniques used in summarization

Tools and resources
–
Part of speech and named entity taggers, parsers,
Wordnet, WEKA

Problem formalization/distributions
–
–

Distributions: Zipfian, Binomial, Multinomial
Graph representations
System comparisons
–
Statistical significance and statistical tests

Reading scientific articles
–
–

Part of the assigned readings
Useful skill, regardless of your future job plans
Improving writing skills
–
–
Immensely useful, regardless of your future job plans
The literature overview assignment will focus on this, but in
other assignments the way you describe your work will also
be evaluated
What is summarization?
Columbia Newsblaster

The academic version
What is the input?

News, or clusters of news
–





a single article or several articles on a related
topic
Email and email thread
Scientific articles
Health information: patients and doctors
Meeting summarization
Video
What is the output




Keywords
Highlight information in the input
Chunks or speech directly from the input or
paraphrase and aggregate the input in novel
ways
Modality: text, speech, video, graphics
Ideal stages of summarization

Analysis
–

Transformation
–

Input representation and understanding
Selecting important content
Realization
–
Generating novel text corresponding to the gist of the input
Most current systems

Use shallow analysis methods
–

Rather than full understanding
Work by sentence selection
–
Identify important sentences and piece them
together to form a summary
Data-driven approaches

Relying on features of the input documents
that can be easily computes from statistical
analysis

Word statistics
Cue phrases
Section headers
Sentence position



Knowledge-based systems

Use more sophisticated natural language
processing

Discourse information
–

Use external lexical resources
–

Resolve anaphora, text structure
Wordnet, adjective polarity lists, opinion
Using machine learning
What are summaries useful for?

Relevance judgments
–
–


Does this document contain information I am
interested in?
Is this document worth reading?
Save time
Reduce the need to consult the full document
Multi-document summarization

Very useful for presenting and organizing
search results
–
–
Many results are very similar, and grouping
closely related documents helps cover more
event facets
Summarizing similarities and differences between
documents
Scientific article summarization

Not only what the article is about, but also
how it relates to work it cites

Determine which approaches are criticized
and which are supported
–
Automatic genre specific summaries are more
useful than original paper abstracts
Other uses

Document indexing for information retrieval

Automatic essay grading, topic identification
module

Data-driven summarization
Frequency as indicator of importance

The topic of a document will be repeated
many times

In multi-document summarization, important
content is repeated in different sources
Greedy frequency method

Compute word probability from input

Compute sentence weight as function of
word probability

Pick best sentence
How to deal with redundancy?
Author JK Rowling has won her legal battle in a
New York court to get an unofficial Harry Potter
encyclopaedia banned from publication.
A U.S. federal judge in Manhattan has sided with
author J.K. Rowling and ruled against the
publication of a Harry Potter encyclopedia created
by a fan of the book series.
–
Shallow techniques not likely to work well
Global optimization for content
selection

What is the best summary? vs What is the
best sentence?

Form all summaries and choose the best
–
What is the problem with this approach?
Sentence clustering for theme
identification
1. PAL was devastated by a pilots' strike in June and
by the region's currency crisis.
2. In June, PAL was embroiled in a crippling three-week
pilots' strike.
3. Tan wants to retain the 200 pilots because they
stood by him when the majority of PAL's pilots
staged a devastating strike in June.

Cluster sentences from the input into similar
themes

Choose one sentence to represent a theme

Consider bigger themes as more important
Using graph representations

Nodes
–
–

Sentences
Discourse entities
Arcs
–
–
Between similar sentences
Between related entities
Using machine learning


Ask people to select sentences
Use these as training examples for machine
learning
–
–

Each sentence is represented as a number of
features
Based on the features distinguish sentences that
are appropriate for a summary and sentences that
are not
Run on new inputs
Information ordering

In what order to present the selected
sentences?
–

An article with permuted sentences will not be
easy to understand
Very important for multi-document
summarization
–
Sentences coming from different documents
Automatic summary edits

Some expressions might not be appropriate
in the new context
–
References:
–
he
– Putin
– Russian Prime Minister Vladimir Putin
–
Discourse connectives


However, moreover, subsequently
Requires more sophisticated NLP techniques
Before
Pinochet was placed under arrest in London Friday by
British police acting on a warrant issued by a Spanish
judge. Pinochet has immunity from prosecution in
Chile as a senator-for-life under a new constitution that
his government crafted. Pinochet was detained in the
London clinic while recovering from back surgery.
After
Gen. Augusto Pinochet, the former Chilean dictator,
was placed under arrest in London Friday by British
police acting on a warrant issued by a Spanish
judge. Pinochet has immunity from prosecution in
Chile as a senator-for-life under a new constitution
that his government crafted. Pinochet was detained
in the London clinic while recovering from back
surgery.
Before
Turkey has been trying to form a new government
since a coalition government led by Yilmaz collapsed
last month over allegations that he rigged the sale of
a bank. Ecevit refused even to consult with the
leader of the Virtue Party during his efforts to form a
government. Ecevit must now try to build a
government. Demirel consulted Turkey's party
leaders immediately after Ecevit gave up.
After
Turkey has been trying to form a new government
since a coalition government led by Prime Minister
Mesut Yilmaz collapsed last month over allegations
that he rigged the sale of a bank. Premier-designate
Bulent Ecevit refused even to consult with the leader
of the Virtue Party during his efforts to form a
government. Ecevit must now try to build a
government. President Suleyman Demirel consulted
Turkey's party leaders immediately after Ecevit gave
up.
Download