Answering complex, list and context questions with LCC's Question-Answering Server Page: 3
7 p.View a full description of this paper.
Extracted Text
The following text was automatically extracted from the image on this page using optical character recognition software:
complex answer. When the most relevant paragraphs
are retrieved, the answers are processed.
When answers to complex, factual questions are ex-
tracted, their validity is granted by semantic unifica-
tions with the question. If the question was asked in
context, the unifications of previous questions and an-
swers are also used to grant the validity of the an-
swer of the current question. When unifications are
not possible, several expansions that use the gloss def-
initions of the WordNet concepts are considered. The
question is ruled not to have an answer when none of
the expansions generate unifications. The processing
of list answers is performed differently because LCC's
QASTM extracts for each question all N best candi-
date answers, where N is the quantifier scalar. Addi-
tional answers are sought only if we could not find all
N answers and if variations of the keywords defining
the same answer type are possible.
Processing questions from the main
task
Two main trends have characterized the main task in
TREC-10. First, the percentage of questions that ask
for definitions of concepts, e.g. "What are capers?" or
"What is an antigen?" represented 25% of the ques-
tions from the main task, an increase from a mere 9%
in TREC-9 and 1% in TREC-8 respectively. The defi-
nition questions normaly require an increase in the so-
phistication of the question-answering system. Second,
in general, the questions had an increased level of diffi-
culty. Questions like What is the esophagus used for?"
or "Why is the sun yellow?" are difficult to process
because the answer relies on expert knowledge, from
medicine in the former example, and from physics in
the latter one. Nevertheless, if a lexical dictionary that
explains the definitions of concepts is available, some
supporting knowledge can be mined. For example, by
inspecting WordNet (Miller 1995), in the case of esoph-
agus we can find that it is "the passage between the
pharynx and the stomach". Moreover, WordNet en-
codes several relations, like meronymy, showing that
the esophagus is part of the digestive tube or gastroin-
testinal tract. The glossed definition of the digestive
tube shows that one of its function is the digestion.
The information mined from WordNet guides sev-
eral processes of bridging inference between the ques-
tion and the expected answer. First the definition
of the concept defined by the WordNet synonym set
{ esophagus, gorge, gullet} indicates its usage as a
passage between two other body parts: the pharynx
and the stomach. Thus the query " esophagus AND
pharynx AND stomach" retrieves all paragraphs con-
taining relevant connections between the three con-cepts, including other possible functions of the esoph-
agus. When the query does not retrieve relevant
paragraphs, new queries combining esophagus and its
holonyms (i.e. gastrointestinal tract) or functions of
the holonyms (i.e. digestion) retrieve the paragraphs
that may contain the answer. To extract the correct
answer, the question and the answer need to be seman-
tically unified.
Q912: What is epilepsy?
Q1273: What is an annuity?
Q1022: What is Wimbledon?
Q1152: What is Mardi Gras?
Q1160: What is dianetics?
Q1280: What is Muscular Distrophy?
Table 1: Examples of definition questions.
The difficulty stands in resolving the level of pre-
cision required by the unification. Currently, LCC's
QASTM considers an acceptable unification when (a)
a textual relation can be established between the ele-
ments of the query matched in the answer (e.g. esoph-
agus and gastrointestinal tract); and (b) the textual
relation is either a syntactic dependency generated by
a parser, a reference relation or it is induced by match-
ing against a predefined pattern. For example, the
pattern "X, particularly Y" accounts for such a rela-
tion, granting the validity of the answer "the upper gas-
trointestinal tract, particularly the esophagus ". How-
ever, we are aware that such patterns generate multiple
false positive results, degrading the performance of the
question-answering system.
Definition Phrase to be Candidate answer
pattern defined (QP) phrase (AP)
<AP> such What is developmental
as < QP> autism? disorders
such as autism
<AP> (also What is manic-dipressive
called < QP>) bipolar illness (also called
disorder? bipolar disorder )
< QP> is What is caffeine is an
an <AP> caffeine? alkaloidTable 2: Identifying candidate
matching.answers with pattern
Predefined patterns are also important for process-
ing definition questions, similar to those listed in Ta-
ble 1. Table 2 lists several patterns and their compo-
nents: the question phrase (QP) that requires a defi-
nition, and the candidate answer phrase (AP) provid-
ing the definition. To process definition questions in
a more robust manner, the search space is enlarged,
allowing the substitution of the phrase to be defined
QP with the immediate hypernym of its head. Table 3
Upcoming Pages
Here’s what’s next.
Search Inside
This paper can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Paper.
Harabagiu, Sanda M.; Moldovan, Dan I.; Paşca, Marius. 1974-; Surdeanu, Mihai; Mihalcea, Rada, 1974-; Gîrju, Corina R. et al. Answering complex, list and context questions with LCC's Question-Answering Server, paper, November 2001; [Gaithersburg, Maryland]. (https://digital.library.unt.edu/ark:/67531/metadc83297/m1/3/?rotate=90: accessed April 19, 2024), University of North Texas Libraries, UNT Digital Library, https://digital.library.unt.edu; crediting UNT College of Engineering.