Martin Kay
Stanford University
and the University of the Saarland
Martin Kay Translation—Introduction 1
Open Questions
Large Structures
Information Structure
Mental states in a restricted domain
Discourse Structure
Directions
The restaurant script (Schank)
The Xerox experience
Martin Kay Translation—Introduction 2
Hyphenation
p-resentation C
pr-esentation C
pre-sentation C
pres-entation E
prese-ntation C
pres-en-ta-tion presen-tation E
present-ation C
presenta-tion E
presentat-ion C
presentati-on C
presentatio-n C
Martin Kay Translation—Introduction 3
Hyphenation
Examples
Counterexamples
Pure Counterexamples
Strings containng at least
one pure counterexample
Strings containng no pure
counterexamples
Martin Kay Translation—Introduction 4
Hyphenation
• Very expensive because full set of
constraints is reimposed after every
character
Martin Kay Translation—Introduction 5
Words
Martin Kay Translation—Introduction 6
Inflexions
mensa
mensam
table mensae
mensarum
mensis
darselo
• Align stems
mostrártela
• Use Lemmatized Text
estudiándola
Martin Kay Translation—Introduction 7
Idioms & Collocations
without a trace
without a shadow of a doubt
in the final analysis
all things considered
conventional weapons
short story
ear, nose, and throat specialist
a little bit at a time
ignition key
spare tire
Martin Kay Translation—Introduction 8
Conjuncts
Over and above
By and large
time and again
once and again
large and small
big and little
bread and butter
boys and girls
men and women
ladies and gentlemen
Martin Kay Translation—Introduction 9
German Compounds
• Lebensversicherungsgesellschaftsangestellter
• Zweihundertsiebenundfünfzig
Languages that do not mark
word boundaries in writing
Martin Kay Translation—Introduction 10
A Suffix
trie
$
p i $
p
p i $
s p
s i
s s i p p i $
i
i s s i s s i p p i $
m
s i p p i $
s
s i p
p i $
s s i p p i $
i s
p p $
p i
$ i $
p
i $ Search Time: O(p)
$
Space: O(t2)
Martin Kay Translation—Introduction 11
A Compressed Suffix
trie (digital search tree)
0 1 2 3 4 5 6 7 8 9 10 11 12
m i s s i s s i p p i $
Space: O(t)
Martin Kay Translation—Introduction 12
Other Applications?
• Searching
• Language Modeling
Martin Kay Translation—Introduction 13
Syntax
Optimality
Optimality and statistics
Nonsyntactic syntax
Robustness
Martin Kay Translation—Introduction 14
Optimality
• A program is nondeterministic if information
required to answer a question is not on hand
when the question arises.
• Questions can be reduced to yes-no
questions.
• If a cost is associated with answering a
question, "no" can be replaced by an
expensive "yes"
Martin Kay Translation—Introduction 15
Hans hat dem Kind das Wasser gegeben
?Hans hat das Wasser dem Kind gegeben
Hans hat Maria Wein gegeben
*Hans hat Wein Maria gegeben
Martin Kay Translation—Introduction 16
Optimality Charges
Place-value notation
No carries
Indirect &
Spelling Errors
Direct Objects
Martin Kay Translation—Introduction 17
Where the grammar fails
• Noun-noun compounds
• Strings of adjectives
• Prepositional phrases
• Chinese?
Martin Kay Translation—Introduction 18
How May Trees?
1 n
2n 1 2n 1
60 00
50 00
Catalan
numbers
40 00
Structures
30 00
20 00
10 00
n3
0
1 2 3 4 5 6 7 8 9 10
Martin Kay
Words
Translation—Introduction 19
Proposal
Use human input but do no more human work
than would be necessary in any case
Martin Kay Translation—Introduction 20
Reflective Editing
Translation is nondeterministic—it implements a choice tree
We can identify an outcome and, in particular, the preferred
one, by giving the answers to the questions on the path
that leads to it.
If any of those questions arise in a subsequent translation
into another language, they should presumably be answered
the same way.
Martin Kay Translation—Introduction 21
Strategy
Produce many translations
Display one of them—the best one.
The editor changes it into …
A version that the system had already
foreseen, but not chosen as the preferred
version.
We know what choices the system would
have had to make to reach that version.
We will make those choices when
translating into the next language.
Martin Kay Translation—Introduction 22
Reflective Editing
First
Translations
Choose
one
Original
Martin Kay Translation—Introduction 23
Reflective Editing
First
Translations
Original Choose
one
Second
Translations
Martin Kay Translation—Introduction 24
There are three windows in the room
Il y a trois fenêtres dans la salle.
Il y a trois guichets dans la salle.
Es gibt drei Fenster in dem Zimmer.
Es gibt drei Schalter in dem Zimmer.
fenêtre ~ Fenster
guichet ~ Schalter
Martin Kay Translation—Introduction 25
It is cold
Il/elle est froid(e)
Il/elle a froid
Il fait froid
Er/sie/es ist kalt
Ihm/ihr ist kalt
Es ist kalt
[faire] froid/chaud ... ~ Es [sein] kalt/warm ...
X [avoir] froid/chaud ... ~ [Dat] [sein] kalt/warm
...
Martin Kay Translation—Introduction 26
Wir haben noch zwei
We still have two.
We have two more.
Il nous en reste deux.
Nous en avons encore deux
still ~ [rester]
more ~ encore
Martin Kay Translation—Introduction 27
Multilingual Grammar
Martin Kay Translation—Introduction 28
It turns out that
• Translation quality is not improved by
considering:
—Morphology
—Grammaticality of phrases
—Word senses
—The first and last words of the sentence
—The original text
Martin Kay Translation—Introduction 29
Martin Kay
Stanford University
and the University of the Saarland
Martin Kay Translation—Introduction 30