AVENUE MILE

1
AVENUE / MILE
Authors Faculty/Staff Jamie Carbonell, Ralf
Brown, Alon Lavie, Lori Levin, Bob Frederking,
Rodolfo Vega. Students Ariadna Font-Llitjos,
Christian Monson, Erik Peterson, Katharina
Probst, Alison Alvarez
The Elicitation Tool
Our Goal AVENUE is a project dedicated to
building machine translation systems for low
resource languages. Our task is to amass
language resources and use them to automatically
infer language transfer rules. These rules are
then used by our run-time system to translate
previously unseen source language text into the
target language. The purpose of MILE is to gather
resources from bilingual speakers of our target
language using an elicitation and alignment tool.
These speakers may not have any linguistic
training. Using this we will amass a small
typologically diverse corpus of aligned sentences
that can be used by our learning system to
automatically infer features in our target
language. These features can then be used to
derive transfer rules, morphological information
or used in corpus navigation to control which
sentences are given to our language consultant.
The elicitation tool provides a simple interface
for bilingual informants with no linguistic
training and limited computer skills to translate
and word-align a corpus in some source language.
The tool can use a static corpus or can be used
with Feature Detection and Corpus Navigation to
dynamically choose sentences based upon
linguistic features already identified in the
target language. The output of the elicitation
tool is a text file containing triplets of
eliciting sentence, elicited sentence, and
alignment. We are also planning XML output so
that the output can be imported into other
linguistic annotation tools. The elicitation
tool can produce bilingual glossaries based on
the aligned corpus. It also has a simple
"auto-align" option to add alignments for
unambiguous word pairs in the same file.
Mapuche women learning how to manage a business
from a teacher of the Fundación Chol-Chol,
Temuco, Chile.
Feature Structure (SUBJ GIRL, 3rd, singular,
human, definite) (VS SEE, Activity, past,
perfect) (OBJ BOOK, 3rd, singular, indefinite)
Feature Structure (SUBJ GIRL, 3rd, singular,
human, indefinite) (VS SEE, Activity, past,
perfect) (OBJ BOOK, 3rd, singular, indefinite)
Feature Structure (SUBJ PRO, 1st, singular,
human) (VS SEE, Activity, past, perfect) (OBJ
BOOK, 3rd, singular, definite)
Feature Structure (SUBJ PRO, 1st, singular,
human) (VS SEE, Activity, past, perfect) (OBJ
BOOK, 3rd, singular, indefinite)
The Elicitation Process
Family or type-specific decision graphs
General Purpose Decision Graphs
I see a book. I see the book. You see a book. You
see the book. She sees the book. She sees a
book. He sees a book. He sees the book.
Test 1 check lowest score Test 2 sample to see
if X feature can be
Test 1 object definiteness Test 2 objects
modified by possessives
Corpus Navigator
Sentence Selector
Elicitation Corpus
Decision Graphs
I see a book. I see the book. You see a book. You
see the book. She sees the book. She sees a
book. He sees a book. He sees the book.
I saw the book ?, ? ?? ?
Feature Detection Archive
Relevant Part of Elicitation Corpus
Archive of Sentences with Alignments
Fact Record 1
Fact Record 2
Fact Record n
Key Human Program Data
Chinese Definiteness of direct object marked by
change of word order
Human Informant
Figure Components of MILE
Working with computer tools in Temuco, Chile
A meeting at the Institute for Indigenous
Studies, Universidad de la Frontera, Temuco, Chile
Quechua Speakers in Peru.