Download INF 5820 Language technological applications

INF 5820
LANGUAGE TECHNOLOGICAL
APPLICATIONS
Fall 2012
Today

1. Hour: Course overview




General
Machine translation
Spoken Dialogue Systems
2. Hour: Starting Spoken Dialogue Systems
INF5820


http://www.uio.no/studier/emner/matnat/ifi/INF58
20/index-eng.xml
Builds on
 INF4820
Algorithms for Artificial Intelligence and
Natural Language Processing,
 might be taken in parallel

Alternates with
 INF5830
Natural Language Processing
Two applications
Spoken Dialogue Systems


First part of semester
Pierre Lison (plison)
Machine translation


Second part
Jan Tore Lønning (jtl)
Language Technology Group (LTG), 7. floor
Classes

Fridays 10.15-12
 Lectures
 OJD

2458 Postscript
Mondays 12.15-14
 Group/lectures
 OJD
2458 Postscript
Obligatory assignments

3 obligatory assignments
 Dialogue
1: 26 September
 Dialogue 2: 17 October
 Translation: final,15 November

PhDs:
 Paper
presentation
Exam


Written exam
18 December at 0900
Spoken Dialogue Systems
Pierre Lison
Machine Translation
Jan Tore Lønning
Machine Translation

Active research field since 1949,
 In
the 1950s MT was not only the most important
NLP/computational linguistics field, it was the only one
 IBM 1954 press release


Interest, results and funding have varied over time
Today:
 Fully-automatic
text-translation: Systran, Google
 Speech-translation: commercial
 Aid for professional translators: trados
Two types of approaches to NLP
Rule-based

Empirical
Build a declarative
model using

 Linguistics

 Logic


Algorithms
How does it fit data?
Start with naturally
occurring text
What information can
we get?
 Statistics/Machine
learning

Use this to reproduce
the examples
Applied to MT
Rule-based

Which linguistic
information should be
included,



syntax?
semantics?
Approaches




Direct translation
Syntax-based transfer
Semantic-based transfer
..
Empirical


Example-based
translation
Statistical machine
translation (SMT)
 Word-based
 Phrase-based
 Syntactic
Figure 25.8
What we will study
1.
2.
3.
MT overview
MT evaluation
Statistical MT,

4.
5.
The main part
Rule-based MT with semantic transfer
Hybrid methods
Literature



J&M, ch. 25
Koehn, in particular, Part II Core methods: ch. 4-8
A few papers
Recommended prior knowledge




INF4820 - Algorithms for artificial intelligence and
natural language processing
Some knowledge in statistics is an advantage
(Wrong text on English page)
In particular:
 Probability
theory
 N-grams
 Hidden
Markov Models
 Dynamic Programming
Assignments




Building an SMT system with open-source software
Experiment with different settings and parameters
Evaluation
One obligatory assignment with several parts