ACL4 Machine Translation Exam 2001
Andy Way
Dublin City University
Attempt five questions. All questions carry equal marks.
Q 1.
Critically evaluate the following comments from Arnold et al,
1994:7:
- MT is a waste of time because you will never make a machine that can
translate Shakespeare.
- Generally, the quality of translation you can get from
an MT system is very low. This makes them
useless in practice.
- MT threatens the jobs of translators.
Q 2.
Study the following quotation:
`I'm sure you can have somebody write some kind of gobbledygook in
Multinational Customized English that some kind of drone can read and
say, ``Yeah, this sounds okay to me!''' (Hofstadter, 2000).
Is this a fair reflection of the current state of controlled language
applications? How do you think that such techniques can help in the
areas of technical authoring and the design of MT systems?
Q 3.
Why might you opt for a transfer-based system over an interlingual
system? Substantiate your opinions with insights from translational
ambiguities, e.g. river --> riviere, fleuve, or wall --> Wand, Mauer.
Q 4.
(a) Give one example of each of the following types of difficult
translation problem:
- Headswitching;
- Relation-Changing;
- Lexical Gap.
(b) Draw source and target dependency structures for
each sentence.
(c) Write a translation rule (using a notation of your choice)
capable of mapping the source dependency structure into the target
dependency structure for each sentence pair.
Q 5.
Most generation systems include the notions of:
- predetermined choice;
- structure preservation.
What are the advantages and disadvantages of such approaches? Give
examples.
Q 6.
What are the benefits and drawbacks of using Terminology
Management systems? Describe how the use of such tools has changed
the nature of the translator's task over the years.
Q 7.
(a) Describe the three main stages to Example-Based MT systems.
(b) What are the advantages of EBMT over rule-based systems?
(c) Explain the main difference between EBMT systems and the Translation
Memory approach. Give a simple example as an illustration.
Q 8.
(a) Any statistical approach to MT requires the availability of large,
good-quality, representative bilingual corpora. Why?
(b) Describe some of the problems involved in deriving aligned
corpora, both at the sentential and word level. How might these be
overcome using some simple heuristics?
Andy Way, 22nd March 2002.