U.S. patent application number 11/762816 was filed with the patent office on 2008-02-07 for acronym extraction system and method of identifying acronyms and extracting corresponding expansions from text.
This patent application is currently assigned to ITT Manufacturing Enterprises, Inc.. Invention is credited to Kalyan M. Gupta.
Application Number | 20080033714 11/762816 |
Document ID | / |
Family ID | 38178852 |
Filed Date | 2008-02-07 |
United States Patent
Application |
20080033714 |
Kind Code |
A1 |
Gupta; Kalyan M. |
February 7, 2008 |
Acronym Extraction System and Method of Identifying Acronyms and
Extracting Corresponding Expansions from Text
Abstract
An acronym expansion system of the present invention receives
electronic documents and extracts acronyms and their corresponding
expansions. A part-of-speech tagger decomposes text into string
tokens or words and tags them with their part-of-speech, while an
acronym identifier determines whether a word is a potential acronym
based on various conditions. An expansion identifier retrieves
lists of words preceding and following a potential acronym to
search for the expansion. The resulting word lists are examined
sequentially to identify and retrieve an expansion for the
potential acronym. An expansion extractor receives the potential
acronym and a processed word list to retrieve the expansion of the
potential acronym from that list. The extractor may utilize
information from prior search iterations, and verifies an extracted
expansion against a set of rules to remove spurious expansions.
Inventors: |
Gupta; Kalyan M.;
(Alexandria, VA) |
Correspondence
Address: |
EDELL, SHAPIRO & FINNAN, LLC
1901 RESEARCH BOULEVARD
SUITE 400
ROCKVILLE
MD
20850
US
|
Assignee: |
ITT Manufacturing Enterprises,
Inc.
Wilmington
DE
|
Family ID: |
38178852 |
Appl. No.: |
11/762816 |
Filed: |
June 14, 2007 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
10212914 |
Aug 7, 2002 |
7236923 |
|
|
11762816 |
Jun 14, 2007 |
|
|
|
Current U.S.
Class: |
704/9 |
Current CPC
Class: |
G06F 40/284
20200101 |
Class at
Publication: |
704/009 |
International
Class: |
G06F 17/27 20060101
G06F017/27 |
Claims
1-95. (canceled)
96. A system for identifying abbreviated terms within text each
representing a corresponding phrase of at least one term and
extracting expansions of said abbreviated terms from said text in
the form of said corresponding phrases comprising: a computer
system to receive said text and identify abbreviated terms and
corresponding expansions therein, said computer system including:
an identification module to examine said text to identify at least
one abbreviated term residing therein; an expansion retrieval
module to retrieve at least one portion of said text for an
identified abbreviated term, wherein each retrieved text portion is
located within said text proximate said identified abbreviated
term; and an expansion extraction module to produce a plurality of
sets of terms from a corresponding retrieved text portion with each
set including a member term that includes an initial portion
containing an initial portion of said identified abbreviated term,
to compare said identified abbreviated term with said produced sets
to extract an expansion for said abbreviated term from one of said
sets, and to verify said extracted expansion to produce a valid
expansion for said identified abbreviated term.
97. The system of claim 96, wherein said at least one abbreviated
term includes an acronym.
98. The system of claim 96 further including: a classification
module to determine a grammatical classification of each term
within said text to enable identification of an abbreviated term
and a corresponding valid expansion.
99. The system of claim 98, wherein said identification module
includes: a text attribute module to examine a term within said
text and determine compliance of text term attributes with
particular conditions in order to identify that text term as an
abbreviated term, wherein said conditions relate to at least one
of: said grammatical classification of a text term; a length of a
text term being within a particular range; the presence of a
predetermined delimiter within a text term; an amount of
capitalization within a text term; an amount of numerals within a
text term; and a text term being a predetermined term excluded from
consideration as an abbreviated term.
100. The system of claim 96 further including: a decomposition
module to selectively decompose terms within at least one retrieved
text portion for said identified abbreviated term into individual
sub-terms to facilitate identification of an expansion within that
retrieved text portion.
101. The system of claim 96, wherein said expansion extraction
module includes: a set module to identify terms within each
retrieved text portion that include an initial portion containing a
first character of said identified abbreviated term and to produce
a corresponding set for each identified term including said
identified term and following terms within said retrieved text
portion.
102. The system of claim 101, wherein said expansion extraction
module further includes: a search module to compare portions of
said identified abbreviated term to terms within a search window to
identify a corresponding expansion term for each abbreviated term
portion, wherein said search window includes a predetermined number
of terms from a corresponding set and is modified in response to
identification of an expansion term to include terms within that
set following said identified expansion term.
103. The system of claim 102, wherein said search module includes:
an expansion term module to compare an abbreviated term portion to
said terms within said search window and identify said
corresponding expansion term for that abbreviated term portion in
response to a search window term including that abbreviated term
portion as a starting portion or within an initial portion of that
search window term.
104. The system of claim 102, wherein said search module includes:
a backtrack module to identify a corresponding expansion term for a
current abbreviated term portion utilizing at least one previous
abbreviated term portion and at least one term from a previous
search window in response to a failure to identify an expansion
term for said current abbreviated term portion.
105. The system of claim 104, wherein said backtrack module
includes: a first level module to compare a combination of the
current and previous portions of said identified abbreviated term
to terms within a previous search window to identify an expansion
term for said current abbreviated term portion.
106. The system of claim 105, wherein said backtrack module further
includes: a second level module to compare a combination of the
current and a plurality of previous portions of said identified
abbreviated term to a combination of terms including terms within a
current search window and a plurality of previous identified
expansion terms to identify an expansion term for said current
abbreviated term portion in response to a failure by said first
level module to identify that expansion term.
107. The system of claim 96, wherein said expansion extraction
module includes: a verification module to verify said extracted
expansion against a set of verification rules that indicate a valid
expansion in response to attributes of said extracted expansion
satisfying particular conditions.
108. The system of claim 107, wherein said conditions relate to at
least one of: said extracted expansion including said identified
abbreviated term, a predetermined delimiter or a predetermined
expansion invalidating term; numerals within said identified
abbreviated term corresponding to numerals within said extracted
expansion; a first term of said extracted expansion including a
predetermined expansion invalidating term; a grammatical
classification of a last term of said extracted expansion; the
presence of particular terms within said extracted expansion being
skipped during identification of said extracted expansion; and the
presence of at least two terms within said extracted expansion
containing an abbreviated term portion within an intermediate
portion of those terms.
109. A method of identifying abbreviated terms within text each
representing a corresponding phrase of at least one term and
extracting expansions of said abbreviated terms from said text in
the form of said corresponding phrases comprising: (a) examining
said text to identify at least one abbreviated term residing
therein; (b) retrieving at least one portion of said text for an
identified abbreviated term, wherein each retrieved text portion is
located within said text proximate said identified abbreviated
term; and (c) producing a valid expansion for said identified
abbreviated term by producing a plurality of sets of terms from a
corresponding retrieved text portion with each set including a
member term that includes an initial portion containing an initial
portion of said identified abbreviated term, comparing said
identified abbreviated term with said produced sets to extract an
expansion for said abbreviated term from one of said sets, and
verifying said extracted expansion.
110. The method of claim 109, wherein said at least one abbreviated
term includes an acronym.
111. The method of claim 109, wherein step (a) further includes:
(a.1) determining a grammatical classification of each term within
said text to enable identification of an abbreviated term and a
corresponding valid expansion.
112. The method of claim 111, wherein step (a.1) further includes:
(a.1.1) examining a term within said text and determining
compliance of text term attributes with particular conditions in
order to identify that text term as an abbreviated term, wherein
said conditions relate to at least one of: said grammatical
classification of a text term; a length of a text term being within
a particular range; the presence of a predetermined delimiter
within a text term; an amount of capitalization within a text term;
an amount of numerals within a text term; and a text term being a
predetermined term excluded from consideration as an abbreviated
term.
113. The method of claim 109, wherein step (b) further includes:
(b.1) selectively decomposing terms within at least one retrieved
text portion for said identified abbreviated term into individual
sub-terms to facilitate identification of an expansion within that
retrieved text portion.
114. The method of claim 109, wherein step (c) further includes:
(c.1) identifying terms within each retrieved text portion that
include an initial portion containing a first character of said
identified abbreviated term and producing a corresponding set for
each identified term including said identified term and following
terms within said retrieved text portion.
115. The method of claim 114, wherein step (c) further includes:
(c.2) comparing portions of said identified abbreviated term to
terms within a search window to identify a corresponding expansion
term for each abbreviated term portion, wherein said search window
includes a predetermined number of terms from a corresponding set
and is modified in response to identification of an expansion term
to include terms within that set following said identified
expansion term.
116. The method of claim 115, wherein step (c.2) further includes:
(c.2.1) comparing an abbreviated term portion to said terms within
said search window and identifying said corresponding expansion
term for that abbreviated term portion in response to a search
window term including that abbreviated term portion as a starting
portion or within an initial portion of that search window
term.
117. The method of claim 115, wherein step (c.2) further includes:
(c.2.1) identifying a corresponding expansion term for a current
abbreviated term portion utilizing at least one previous
abbreviated term portion and at least one term from a previous
search window in response to a failure to identify an expansion
term for said current abbreviated term portion.
118. The method of claim 117, wherein step (c.2.1) further
includes: (c.2.1.1) comparing a combination of the current and
previous portions of said identified abbreviated term to terms
within a previous search window to identify an expansion term for
said current abbreviated term portion.
119. The method of claim 118, wherein step (c.2.1) further
includes: (c.2.1.2) comparing a combination of the current and a
plurality of previous portions of said identified abbreviated term
to a combination of terms including terms within a current search
window and a plurality of previous identified expansion terms to
identify an expansion term for said current abbreviated term
portion in response to a failure to identify that expansion term
within step (c.2.1.1).
120. The method of claim 109, wherein step (c) further includes:
(c.1) verifying said extracted expansion against a set of
verification rules that indicate a valid expansion in response to
attributes of said extracted expansion satisfying particular
conditions.
121. The method of claim 120, wherein said conditions relate to at
least one of: said extracted expansion including said identified
abbreviated term, a predetermined delimiter or a predetermined
expansion invalidating term; numerals within said identified
abbreviated term corresponding to numerals within said extracted
expansion; a first term of said extracted expansion including a
predetermined expansion invalidating term; a grammatical
classification of a last term of said extracted expansion; the
presence of particular terms within said extracted expansion being
skipped during identification of said extracted expansion; and the
presence of at least two terms within said extracted expansion
containing an abbreviated term portion within an intermediate
portion of those terms.
122. A program product apparatus including a computer readable
medium with computer program logic recorded thereon for identifying
abbreviated terms within text each representing a corresponding
phrase of at least one term and extracting expansions of said
abbreviated terms from said text in the form of said corresponding
phrases, said program product apparatus comprising: an
identification module to examine said text to identify at least one
abbreviated term residing therein; an expansion retrieval module to
retrieve at least one portion of said text for an identified
abbreviated term, wherein each retrieved text portion is located
within said text proximate said identified abbreviated term; and an
expansion extraction module produce a plurality of sets of terms
from a corresponding retrieved text portion with each set including
a member term that includes an initial portion containing an
initial portion of said identified abbreviated term, to compare
said identified abbreviated term with said produced sets to extract
an expansion for said abbreviated term from one of said sets, and
to verify said extracted expansion to produce a valid expansion for
said identified abbreviated term.
123. The apparatus of claim 122, wherein said at least one
abbreviated term includes an acronym.
124. The apparatus of claim 122 further including: a classification
module to determine a grammatical classification of each term
within said text to enable identification of an abbreviated term
and a corresponding valid expansion.
125. The apparatus of claim 122, wherein said identification module
includes: a text attribute module to examine a term within said
text and determine compliance of text term attributes with
particular conditions in order to identify that text term as an
abbreviated term, wherein said conditions relate to at least one
of: said grammatical classification of a text term; a length of a
text term being within a particular range; the presence of a
predetermined delimiter within a text term; an amount of
capitalization within a text term; an amount of numerals within a
text term; and a text term being a predetermined term excluded from
consideration as an abbreviated term.
126. The apparatus of claim 122 further including: a decomposition
module to selectively decompose terms within at least one retrieved
text portion for said identified abbreviated term into individual
sub-terms to facilitate identification of an expansion within that
retrieved text portion.
127. The apparatus of claim 122, wherein said expansion extraction
module includes: a set module to identify terms within each
retrieved text portion that include an initial portion containing a
first character of said identified abbreviated term and to produce
a corresponding set for each identified term including said
identified term and following terms within said retrieved text
portion.
128. The apparatus of claim 127, wherein said expansion extraction
module further includes: a search module to compare portions of
said identified abbreviated term to terms within a search window to
identify a corresponding expansion term for each abbreviated term
portion, wherein said search window includes a predetermined number
of terms from a corresponding set and is modified in response to
identification of an expansion term to include terms within that
set following said identified expansion term.
129. The apparatus of claim 128, wherein said search module
includes: an expansion term module to compare an abbreviated term
portion to said terms within said search window and identify said
corresponding expansion term for that abbreviated term portion in
response to a search window term including that abbreviated term
portion as a starting portion or within an initial portion of that
search window term.
130. The apparatus of claim 128, wherein said search module
includes: a backtrack module to identify a corresponding expansion
term for a current abbreviated term portion utilizing at least one
previous abbreviated term portion and at least one term from a
previous search window in response to a failure to identify an
expansion term for said current abbreviated term portion.
131. The apparatus of claim 130, wherein said backtrack module
includes: a first level module to compare a combination of the
current and previous portions of said identified abbreviated term
to terms within a previous search window to identify an expansion
term for said current abbreviated term portion.
132. The apparatus of claim 131, wherein said backtrack module
further includes: a second level module to compare a combination of
the current and a plurality of previous portions of said identified
abbreviated term to a combination of terms including terms within a
current search window and a plurality of previous identified
expansion terms to identify an expansion term for said current
abbreviated term portion in response to a failure by said first
level module to identify that expansion term.
133. The apparatus of claim 122, wherein said expansion extraction
module includes: a verification module to verify said extracted
expansion against a set of verification rules that indicate a valid
expansion in response to attributes of said extracted expansion
satisfying particular conditions.
134. The apparatus of claim 133, wherein said conditions relate to
at least one of: said extracted expansion including said identified
abbreviated term, a predetermined delimiter or a predetermined
expansion invalidating term; numerals within said identified
abbreviated term corresponding to numerals within said extracted
expansion; a first term of said extracted expansion including a
predetermined expansion invalidating term; a grammatical
classification of a last term of said extracted expansion; the
presence of particular terms within said extracted expansion being
skipped during identification of said extracted expansion; and the
presence of at least two terms within said extracted expansion
containing an abbreviated term portion within an intermediate
portion of those terms.
Description
BACKGROUND OF THE INVENTION
[0001] 1. Technical Field
[0002] The present invention pertains to text analysis and
processing systems. In particular, the present invention pertains
to a system that identifies acronyms and extracts the appropriate
acronym expansion from text.
[0003] 2. Discussion of Related Art
[0004] An acronym is a word that is formed from the initial letter
or letters of each component of a compound term (e.g., NATO, RADAR,
SNAFU, etc.), while an abbreviation is a shortened form of a
written word or phrase that is used or substituted for the whole
word (e.g., "amt" is an abbreviation for amount). Acronyms and
abbreviations tend to overlap and are frequently used in daily
verbal discourse, in written documents and in electronic documents
and web pages on the Internet. In certain communities (e.g.,
military, engineering, medicine, etc.), numerous acronyms are
employed constantly. For example, a page of a military document
commonly includes in excess of ten acronyms.
[0005] Acronyms may present challenges to readers in several
manners. In particular, individuals unfamiliar with a certain
acronym tend to have difficulty understanding the acronym and using
the acronym in vocabulary. For example, commonly known acronyms,
such as "LASER" and "CDROM", are widely understood, while
infrequently used or subject specific acronyms may be difficult for
readers to understand (e.g., "AABFS" for Amphibious Assault Bulk
Fuel System). Further, individuals preparing and/or compiling
information for customers (e.g., librarians, technical writers,
etc.) are aware of acronyms and typically provide convenient
manners to search and access an acronym expansion. Systems that
provide these types of services in the digital and electronic media
are commonly referred to as "digital libraries" and "document
databases". In order to be effective, a digital library should
recognize acronyms and the corresponding expansion during a search.
This process may be performed manually; however, preparing acronym
lists with corresponding expansions in this fashion becomes
prohibitive due to the effort required and is prone to errors.
[0006] The related art has attempted to overcome these problems by
providing various systems for acronym expansion. For example, the
AcronymFinder system enables access to a manually compiled list of
acronyms on the Internet. This system receives manual submissions
of acronyms and corresponding expansions to update the list. The
compiled list (e.g., in excess of 150,000 acronyms) is available
for embedding in applications.
[0007] The Acronym Finding Program (AFP) is an early acronym
extraction system designed primarily for an optical character
recognition (OCR) environment. This system utilizes a few simple
heuristics for acronym identification and expansion. For an example
of this type of system, reference is made to Taghva et al.,
"Recognizing Acronyms and Definitions", Proceedings of the Fourth
International Conference on Document Analysis and Recognition, pp.
191-198, 1999, Los Alamitos, Calif.: IEEE Computer Society, the
disclosure of which is incorporated herein by reference in its
entirety.
[0008] A further system, TLA, is derived from ATP and uses five
heuristics. This system produced a performance of 68% recall and
91% precision on a set of computer science technical reports. For
an example of this type of system and performance, reference is
made to Yeates, "Automatic Extraction of Acronyms from Test",
Proceedings of the Third New Zealand Computer Science Research
Student's Conference, pp. 117-124, 1999, Hamilton, New Zealand, the
disclosure of which is incorporated herein by reference in its
entirety.
[0009] Another acronym extraction system employs text compression
algorithms. This system uses zero-order compression models as a
manner to extract acronym expansions, where the model parameter
settings are learned using an encoded training set. For an example
of this type of system, reference is made to Yeates et al., "Using
Compression to Identify Acronyms in Text", Proceedings of the IEEE
Data Compression Conference, pp. 582-589, 2000, Los Alamitos,
Calif.: IEEE Computer Society, the disclosure of which is
incorporated herein by reference in its entirety.
[0010] Yet another acronym extraction system exploits duality of
patterns and relations. The system is seeded with extraction
patterns for acronym expansion relations. After an initial set of
extractions has been obtained, the extracted instances are utilized
to learn new patterns and the process repeats until convergence.
For examples of this type of system, reference is made to U.S. Pat.
No. 6,385,629 (Sundaresan et al.) and to Yi et al., "Mining the Web
for Acronyms Using the Duality of Patterns and Relations",
Proceedings of the ACM CIKM '99 Second Workshop on Web Information
and Data Management", pp. 48-52, Kansas City, Mo., the disclosures
of which are incorporated herein by reference in their
entireties.
[0011] Still another system for acronym expansion is a heuristic
extractor and server commonly referred to as "Acrophile". This
system includes three versions of varying capability that use
acronym identification and expansion extraction rules. For an
example of this type of system, reference is made to Larkey et al,
"Acrophile: An Automated Acronym Extractor and Server", Proceedings
of the ACM Digital Libraries Conference, pp. 205-214, 2000, the
disclosure of which is incorporated herein by reference in its
entirety.
[0012] The related art systems described above suffer from several
disadvantages. In particular, the AcronymFinder system is highly
inefficient due to the list being generated by manual submissions.
Further, the list is typically generic and static and may not suit
or be tailored to various needs of particular organizations.
Although the above-described systems extract acronyms and
corresponding expansions, the results produced by these systems
have limited accuracy. This tends to frustrate readers since the
systems may omit acronyms within text or provide incorrect
expansions for the acronyms, thereby requiring the reader to
perform an additional task of ascertaining the correct expansion in
another manner (e.g., manually). Thus, there exists a need in the
art for a system that processes electronic text and documents and
produces acronyms and corresponding expansions with a high degree
of accuracy.
OBJECTS AND SUMMARY OF THE INVENTION
[0013] Accordingly, it is an object of the present invention to
identify acronyms within text and determine corresponding
expansions with a high degree of accuracy.
[0014] It is another object of the present invention to employ a
bi-directional candidate search technique to select terms near an
acronym in order to enhance identification of acronym
expansions.
[0015] Yet another object of the present invention is to combine
search iterations for acronym expansion with prior iterations in
response to a search iteration failure to enhance identification of
acronym expansions within text.
[0016] Still another object of the present invention is to apply
rules to identified acronym expansions to verify the validity of
those expansions and enhance accuracy.
[0017] The aforesaid objects may be achieved individually and/or in
combination, and it is not intended that the present invention be
construed as requiring two or more of the objects to be combined
unless expressly required by the claims attached hereto.
[0018] According to the present invention, a computer system
receives electronic ASCII or plain text documents and extracts
acronyms and their corresponding expansions. In particular, the
system includes a part-of-speech tagger, an acronym candidate
identifier, an expansion candidate identifier, a word splitter and
an expansion extractor. The part-of-speech tagger decomposes text
into string tokens or words and tags them with their part-of-speech
(e.g., noun, verb, adjective, etc.). Each tagged word or token is
passed to the acronym candidate identifier that determines whether
the word is a potential acronym based on various conditions (e.g.,
whether the word is a noun, the size of the word, amount of
capitalization, etc.). If a word is identified as a potential
acronym, the expansion candidate identifier retrieves respective
lists of words preceding and following the potential acronym within
the text to search for the expansion corresponding to the potential
acronym.
[0019] The resulting word lists are processed sequentially (e.g.,
one at a time) as described below to identify and retrieve an
expansion for a potential acronym. When a valid expansion is
identified within a list, processing of the current potential
acronym ceases (e.g., remaining lists are not searched) and the
next potential acronym within the text is identified and processed.
In particular, a resulting word list is passed to the word splitter
that splits words residing in the word list to produce a search
word list (e.g., including the word list and split words). For
example, the word "Hypertext" may be decomposed into the terms
"Hyper" and "text". The potential acronym and search word list are
passed to the expansion extractor to retrieve the expansion of the
potential acronym from the search word list. The expansion
extractor includes an expansion preprocessor, an expansion searcher
and an expansion filter. The expansion preprocessor initially
prepares potential acronyms for searching by expanding potential
acronyms including numerals (e.g., "W3C" is expanded to
"WWWC").
[0020] The preprocessor further generates search candidates (e.g.,
subsets of the search word list) from the search word list in order
to search for the acronym expansion. The search candidates are
processed sequentially to identify and retrieve an acronym
expansion. In particular, the expansion searcher looks for
characters in the potential acronym to match with the beginning or
middle characters in the words within a search window, typically
including a portion of a current search candidate. Basically, the
expansion searcher sequentially examines the potential acronym
characters and locates corresponding words in the search window.
When a search within the current search window for a potential
acronym character or segment is successful, the current segment,
corresponding word and search window are saved and the search
window is updated for searching further acronym characters. If a
search within the search window for a potential acronym character
or segment fails (e.g., no corresponding word resides within the
search window), the expansion searcher performs backtracking.
Specifically, the system maintains a stack of potential acronym
segments and search windows. When a search fails, the expansion
searcher appends the current acronym segment with prior acronym
segments and modifies the search window with prior search window
portions to perform a search. For example, the acronym "TACON" and
initial search window "Tactical Control" enable identification of
"Tactical" for "T". When searching for acronym segment "A" in a
modified search window of "Control", the search fails. The system
backtracks by appending segment "A" to prior segment "T" (e.g.,
"TA") and searching the modified search window of "Tactical
Control" to identify "Tactical" for "TA".
[0021] When an expansion is identified within a search candidate,
the expansion is extracted (e.g., produced from the identified
words corresponding to the acronym segments) and is verified
against a set of rules by the expansion filter to remove spurious
expansions. A valid expansion and corresponding acronym are
produced, and the next acronym is retrieved for processing as
described above. The above process is repeated for each tagged word
to identify acronyms within the text and retrieve corresponding
expansions.
[0022] The above and still further objects, features and advantages
of the present invention will become apparent upon consideration of
the following detailed description of specific embodiments thereof,
particularly when taken in conjunction with the accompanying
drawings wherein like reference numerals in the various figures are
utilized to designate like components.
BRIEF DESCRIPTION OF THE DRAWINGS
[0023] FIG. 1 is a perspective view of an exemplary computer system
employed by the present invention.
[0024] FIG. 2 is a block diagram of an acronym expansion system
according to the present invention.
[0025] FIG. 3 is a procedural flow chart illustrating the manner in
which the acronym expansion system identifies acronym candidates
according to the present invention.
[0026] FIG. 4 is a procedural flow chart illustrating the manner in
which the acronym expansion system retrieves text portions serving
as candidates for expanding the identified acronym candidates
according to the present invention.
[0027] FIG. 5 is a procedural flow chart illustrating the manner in
which the acronym expansion system decomposes words within an
expansion candidate into individual terms to produce an expansion
list according to the present invention.
[0028] FIG. 6 is a procedural flow chart illustrating the manner in
which the acronym expansion system identifies and retrieves an
acronym expansion from the expansion list according to the present
invention.
[0029] FIG. 7 is a procedural flow chart illustrating the manner in
which the acronym expansion system searches for an acronym
expansion within a subset of the expansion list according to the
present invention.
[0030] FIG. 8 is a procedural flow chart illustrating the manner in
which the acronym expansion system compares an acronym candidate to
words within a search window to identify an acronym expansion
according to the present invention.
[0031] FIG. 9 is a procedural flow chart illustrating the manner in
which the acronym expansion system backtracks or combines prior
search iterations to identify an acronym expansion according to the
present invention.
[0032] FIG. 10 is a procedural flow chart illustrating the manner
in which the acronym expansion system verifies an identified
acronym expansion according to the present invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0033] The acronym expansion system or tool of the present
invention basically receives ASCII or plain text documents and
extracts acronyms (e.g., including phrasal abbreviations, such as
"SYSAD" for System Administration) and corresponding expansions.
The acronym expansion tool may process text contained in documents
of other formats (e.g., .pdf, HTML, etc.); however, these documents
are typically converted to plain text format for processing. The
acronym expansion tool is preferably implemented by a computer
system as illustrated, by way of example only, in FIG. 1.
Specifically, the computer system is typically implemented by a
conventional personal or other suitable computer system or
workstation preferably equipped with a display or monitor 2, a base
4 (e.g., including the processor, memories and/or internal or
external communications devices (e.g., modem, network cards,
etc.)), a keyboard 6 and optional mouse 8 or other input device.
The computer system includes software (e.g., operating system,
acronym tool software, etc.) and appropriate components (e.g.,
processor, disk storage or hard drive, etc.) including sufficient
processing and storage capabilities to effectively execute that
software. The acronym software is preferably implemented in the
Java computing language, while the computer system may include any
platform that supports Java (e.g., Unix, Windows, Macintosh, etc.).
However, the acronym tool software may be implemented in any
desired computing language with the computer system utilizing any
of the major platforms (e.g., Linux, Macintosh, Unix, OS2, Windows,
etc.). By way of example, the computer system may be implemented by
a personal computer (PC) with a one gigahertz processor and
employing a Windows 2000 platform. The computer system, under
software control, basically implements the acronym expansion tool
of the present invention for identifying and extracting acronyms
and corresponding expansions from text. The computer system
preferably operates as a stand-alone tool. However, the present
invention may be in the form of an embeddable system (e.g., within
another system, as an embeddable software component within other
software, etc.) or the computer system may function as a server in
communication with client or user systems to process client
requests. The acronym tool software may be available on a recorded
medium (e.g., magnetic, optical, floppy, DVD, CD, etc.) or in the
form of a carrier wave or signal for downloading from a source via
a communication medium (e.g., bulletin board, network, WAN, LAN,
Intranet, Internet, etc.).
[0034] The acronym expansion tool of the present invention is
illustrated in FIG. 2. Initially, the tool includes various
software modules or units (e.g., routines, procedures, etc.) to
control a computer system to perform acronym processing tasks as
described above. However, the modules or units may alternatively be
implemented by any individual one or combination of circuitry,
software modules and/or hardware modules to perform the functions
described below. Specifically, the acronym expansion tool includes
a part-of-speech tagger module 10, an acronym candidate identifier
module 12, an expansion candidate identifier module 14, a word
splitter module 16 and an expansion extractor module 18.
Part-of-speech tagger module 10 receives plain text (e.g., "PLAIN
TEXT" as indicated in FIG. 2) and decomposes the entire text into
string tokens or words. The part-of-speech tagger module examines
each word and tags that word with a particular grammatical
classification or part of speech (e.g., noun, verb, adjective,
etc.). The part-of-speech tagger module preferably employs the
conventional Brill Part-of-Speech tagger. For an example of this
type of tagger, reference is made to Brill, "A Corpus-based
Approach to Language Learning", Doctoral Dissertation: University
of Pennsylvania, 1993, the disclosure of which is incorporated
herein by reference in its entirety. However, any conventional or
custom part-of-speech tagger for the English language may be
employed.
[0035] Acronym identifier module 12 receives the tagged words
(e.g., "STRING_TOKENS" as indicated in FIG. 2) from the
part-of-speech tagger module to identify potential acronyms. For
example, the word "U3C" is a potential acronym, whereas the word
"nation" is not a potential acronym. The acronym identifier module
examines various characteristics of each word to identify potential
acronyms as illustrated in FIG. 3. Initially, the acronym
identifier module examines various characteristics of a word to
identify a potential acronym. By way of example only, these
characteristics include: the speech classification of the word; the
quantity or number of characters within the word; the presence of
certain delimiters within the word; the amount of capitalization
within the word; the proportion of numerals within the word; and
the presence of the word within a user configurable list.
Specifically, the acronym identifier module receives or retrieves a
tagged word at step 30, and examines the type of speech of that
word based on the corresponding tag at step 32. If the word is
determined to be a noun (e.g., thereby satisfying one of the
conditions for identification as a potential acronym), module 12
continues to examine the word for compliance with remaining
criteria. Otherwise, the module accesses the next tagged word at
step 30 for processing in response to determining the presence of
additional tagged words at step 46.
[0036] When the word is identified as a noun, module 12 determines
whether or not the length of the word (e.g., number of characters)
is between two and fourteen (e.g., 2.ltoreq.length.ltoreq.14) at
step 34. If the length is within this range (e.g., thereby
satisfying another condition for identification as a potential
acronym), module 12 continues to examine the word for compliance
with remaining criteria. Otherwise, the module accesses the next
tagged word at step 30 for processing in response to determining
the presence of additional tagged words at step 46 as described
above.
[0037] When the length of the word is within the acceptable range
(e.g., 2.ltoreq.length.ltoreq.14), module 12 determines the
presence of delimiters (e.g., "[ ]", "{ }", punctuations, etc.)
within the word in accordance with a user configurable delimiter
list (e.g., "ACRONYM_DELIMITER_LIST" as indicated in FIG. 2) at
step 36. Basically, module 12 compares each character within the
word to the delimiter list in order to identify delimiters within
the word. If delimiters are absent from the word (e.g., thereby
satisfying another condition for identification as a potential
acronym), module 12 continues to examine the word for compliance
with remaining criteria. Otherwise, the module accesses the next
tagged word at step 30 for processing in response to determining
the presence of additional tagged words at step 46 as described
above.
[0038] When delimiters are absent from the word, module 12
determines the amount of capitalization within the word at step 38.
This is typically accomplished by determining the quantity or
number of capital letters within the word and dividing by the word
length (e.g., Capitalization=(Number of Capital Letters)/(Word
Length)). If this ratio is greater than 0.6 or, in other words,
greater than sixty percent of the word is capitalized (e.g.,
thereby satisfying another condition for identification as a
potential acronym), module 12 continues to examine the word for
compliance with the remaining criteria. Otherwise, the module
accesses the next tagged word at step 30 for processing in response
to determining the presence of additional tagged words at step 46
as described above.
[0039] When the amount of capitalization is acceptable, module 12
determines the proportion of numerals within the word at step 40.
This is typically accomplished by determining the quantity or
number of numerals within the word and dividing by the word length
(e.g., Numeral Proportion=(Number of Numerals)/(Word Length)). If
this ratio is less than 0.5 or, in other words, less than fifty
percent of the characters within the word are numerals (e.g.,
thereby satisfying another condition for identification as a
potential acronym), module 12 continues to examine the word for
compliance with remaining criteria. Otherwise, the module accesses
the next tagged word at step 30 for processing in response to
determining the presence of additional tagged words at step 46 as
described above.
[0040] When the numeral proportion is acceptable, module 12
determines the presence of the word within a user-configurable
exclusion list (e.g., "ACRONYM_EXCLUSION_LIST" as indicated in FIG.
2) at step 42. This list basically identifies particular words to
be excluded from consideration as an acronym and enhances accuracy.
Basically, module 12 searches the exclusion list for the word. If
the word is absent from the list, the word at this point has
satisfied each of the criteria for identification as a potential
acronym, and module 12 identifies the word as an acronym candidate
at step 44 and a search for a corresponding expansion of the
identified acronym candidate commences as described below.
Otherwise, the module accesses the next tagged word at step 30 for
processing in response to determining the presence of additional
tagged words at step 46 as described above. When each tagged word
has been processed as determined at step 46, processing of the
document or plain text by the acronym expansion tool terminates.
The various conditions and thresholds described above may be
modified in any manner to adjust the sensitivity of module 12 with
respect to identifying potential acronyms.
[0041] When acronym identifier module 12 identifies a tagged word
as an acronym candidate, expansion candidate identifier module 14
(FIG. 2) receives the acronym candidate (e.g., "ACRONYM CANDIDATE"
as indicated in FIG. 2) from module 12 and selects expansion
candidates or terms within the text in the neighborhood of the
identified acronym as illustrated in FIG. 4. In other words, module
14 retrieves individual sets of tagged words within the text that
precede (e.g., are prior to) and follow (e.g., are subsequent) the
identified acronym, respectively. These sets or expansion
candidates each basically include a sequence of words in which an
expansion for an acronym may be searched. Specifically, module 14
receives an acronym candidate and text tagged with the appropriate
speech classification (e.g., "CONTEXT_STRING" as indicated in FIG.
2), and determines the length or number of characters within the
acronym candidate at step 50. The received information may further
include positional information (e.g., a pointer) to indicate the
location of an acronym candidate within the text. If the number of
acronym characters is less than or equal to three (e.g., number of
characters.ltoreq.3) as determined at step 51, module 14 sets the
number of words to retrieve from the tagged text to be twice the
acronym candidate length (e.g., number of words to retrieve=2 *
acronym candidate length) at step 52. Otherwise (e.g., when the
acronym candidate length is greater than three), the number of
words to retrieve is set to five more than the acronym length
(e.g., number of words to retrieve=acronym candidate length+5) at
step 53. Module 14 subsequently retrieves from the tagged text at
step 54 a set of words prior to or preceding the acronym candidate
and a set of words following or subsequent the acronym candidate
with each set including the number of words determined as described
above (e.g., twice the acronym length for acronym candidates having
a length less than or equal to three, or five more than the acronym
candidate length for acronyms having lengths greater than three).
The retrieval may utilize the positional information to rapidly
locate the words for each set within the text. The smaller sets for
shorter acronyms enable avoidance of selecting spurious expansions,
while the length of sets for larger acronyms is determined from
effective results indicated through simulations. However, the set
lengths and corresponding conditions may be modified in any manner
to adjust sensitivity of expansion candidate selection.
[0042] Once the word sets or expansion candidates are retrieved,
the expansion candidates are sequentially processed to determine
the presence of an expansion for the acronym candidate. In
particular, an initial expansion candidate (e.g., set of words
preceding the acronym) is processed and searched for an expansion
at step 55 in the manner described below. Module 14 basically
places the current expansion candidate in a list (e.g.,
"EXPANSION_CANDIDATE_WORD_LIST" as indicated in FIG. 2) for
processing by word splitter module 16 described below to commence
expansion searching. If an expansion is identified within that
expansion candidate as determined at step 56, module 14 informs the
acronym identifier module at step 58 to locate and facilitate
processing of the next potential acronym within the text in the
manner described above (e.g., generally returning to step 46 of
FIG. 3). When the expansion is absent from the initial expansion
candidate, module 14 determines the presence of additional
expansion candidates at step 57. If additional expansion candidates
are present, the next expansion candidate (e.g., the set of words
following the acronym) is processed and searched for an expansion
at step 55 in the manner described below. When an expansion is
identified within that expansion candidate, the acronym identifier
module is informed at step 58 to locate and facilitate processing
of the next acronym candidate as described above. The process is
basically repeated until each expansion candidate has been
processed or until an expansion is identified. If no expansion
candidates contain an expansion, processing of the current
potential acronym ceases and the acronym identifier module is
informed at step 58 to locate and facilitate processing of the next
potential acronym as described above.
[0043] Word splitter module 16 (FIG. 2) receives an expansion
candidate (e.g., a word set) from the expansion candidate
identifier module and utilizes a data resource to split or
decompose words within the expansion candidate to facilitate
expansion searching as illustrated in FIG. 5. The word splitter
module basically splits words with hyphens, slashes and common
prefixes, and employs a user-defined prefix list (e.g., "PREFIX
LIST" as indicated in FIG. 2) to identify prefixes and decompose
words into individual terms. For example, the word "Hypertext" may
be decomposed into the terms "Hyper" and "Text", while the word
"Unexploded" may be decomposed into the terms "Un" and
"exploded".
[0044] The word splitter module searches each word in an expansion
candidate for a prefix or delimiter within respective prefix and
delimiter lists. If a prefix is identified, the module decomposes
the word into individual terms that include the prefix and
remaining word portion, respectively. When a delimiter is present,
the module decomposes the word into individual terms that include
the word portions preceding and following the delimiter,
respectively. The terms are added to the expansion candidate.
Specifically, word splitter module 16 receives an expansion
candidate at step 60 and further retrieves a word from that
expansion candidate at step 62. Module 16 determines the presence
of a delimiter within the word at step 64. This is typically
accomplished by comparing each character of the word to the
delimiter list. If a delimiter is present, the word is decomposed
at step 66 into individual terms containing the word portion
preceding and following the delimiter, respectively. These terms
are added as words to the expansion candidate at step 72 with the
resulting expansion candidate stored in an expansion list (e.g.,
"EXPANSION_CANDIDATE_WORD _LIST_SPLIT" as indicated in FIG. 2). The
expansion list is updated in response to each word decomposition
within the expansion candidate for further processing as described
below.
[0045] When a delimiter is absent from the word as determined at
step 64, the word splitter module determines the presence of a
prefix within the word at step 68. This is typically accomplished
by comparing the word to the prefix list. If the word includes a
prefix, the word is decomposed into individual terms at step 70
containing the prefix and remaining word portion, respectively.
These terms are added as words to the expansion candidate at step
72 as described above.
[0046] After terms have been added to the expansion candidate at
step 72 or if the word does not include a prefix as determined at
step 68, the word splitter module retrieves and processes
subsequent or following words in the expansion candidate in the
manner described above until each candidate word has been processed
as determined at step 74. Thus, the resulting expansion candidate
is stored in the expansion list and includes the words initially
within the expansion candidate (e.g., words retrieved from the text
in the neighborhood of an identified acronym) and the decomposed
terms of those words.
[0047] Expansion extractor module 18 (FIG. 2) receives an acronym
candidate and the resulting expansion list and determines the
presence of an expansion for the acronym candidate within that
list. The expansion extractor module includes an expansion
preprocessor module 20, an expansion searcher module 22 and an
expansion filter module 24. The preprocessor module prepares the
acronym candidate and expansion list for searching, while the
expansion searcher module searches for characters in the prepared
acronym candidate that match the initial or intermediate portions
of words contained in a search window (e.g., including a portion of
the prepared expansion list). When the search is successful, an
expansion is extracted and prepared for output with the
corresponding acronym. The expansion filter module verifies the
extracted expansion in accordance with filtering rules to remove
expansions considered to be spurious.
[0048] The manner in which the expansion extractor module
determines the presence of an acronym expansion within an expansion
list is illustrated in FIG. 6. Initially, expansion preprocessor
module 20 (FIG. 2) preprocesses the acronym candidate at step 80.
This typically includes expanding acronym candidates that contain
numerals by replacing each combination of a numeral (e.g., numerals
less than five) and preceding character (e.g., alphabetic) within
an acronym candidate with a string including the preceding
character repeated a number of instances corresponding to the
numeral. For example, the candidate "W3C" is expanded to "WWWC",
while the candidate "C2PC" is expanded to "CCPC". The preprocessor
module may further modify an acronym candidate from plural to
singular form by removing the plural suffix (e.g., the suffix "s").
By way of example, the acronym candidate "FLOPs" may be modified to
"FLOP". The resulting acronym or acronym search candidate is
utilized by module 18 to search for and identify an expansion as
described below.
[0049] Once the acronym candidate is preprocessed, expansion
preprocessor module 20 basically creates subsets of the expansion
list at step 82. This is typically accomplished by locating each
word within the expansion list including the same first letter as
the acronym search candidate or including the first letter of the
acronym search candidate within the initial thirty percent of that
word. Each expansion list subset includes a corresponding located
word and the words following the located word within the expansion
list (e.g., words within the expansion list preceding the located
word are excluded from the expansion list subset). For example, the
acronym search candidate "ATO" and corresponding expansion list
"The, Air, Tasking, Order, was, issued" enables identification of
the words "Air" and "Tasking" since the letter "A" is the first
letter of "Air" and within the initial portion of "Tasking". This
example produces expansion list subsets "Air, Tasking, Order, was,
issued" and "Tasking, Order, was, issued" with each subset
containing a corresponding identified word and subsequent or
following words within the expansion list. If no expansion subsets
are generated (e.g., words are absent from the expansion list
corresponding to the first acronym search candidate letter) as
determined at step 83, module 18 determines the presence of
additional expansion candidates at step 102. In other words, the
system excludes the current expansion candidate from consideration
when no words within the expansion list derived from that candidate
correspond to the acronym search candidate first letter. If
additional expansion candidates exist (e.g., the set of words
following the acronym candidate), module 18 informs the expansion
candidate identifier module at step 106 to commence processing of
the next expansion candidate in the manner described above (e.g.,
generally returning to step 57 of FIG. 4). Otherwise, module 18
informs the acronym identifier module at step 108 to locate and
facilitate processing of the next potential acronym within the text
in the manner described above (e.g., generally returning to step 46
of FIG. 3).
[0050] When expansion list subsets are generated, the acronym
search candidate and expansion list subsets are utilized to
identify the acronym expansion. Basically, the expansion searcher
module applies a bi-directional search pattern to match segments
within the acronym search candidate with words in the respective
expansion list subsets. The expansion searcher module searches for
words in a search window (e.g., a portion of an expansion list
subset) that contains the acronym segment. The expansion searcher
module examines each character in the acronym search candidate
(e.g.,"ACRONYM_SEARCH_CANDIDATE" as indicated in FIG. 2)
sequentially, locates corresponding words in the search window and
updates the window to exclude the located word. When the search
fails, the expansion searcher module employs backtracking, where
prior acronym segments are combined with a current segment and the
search window is modified to determine a match. This feature
enhances extraction performance. Once matches are identified for
all acronym segments, the expansion searcher module prepares the
resulting expansion by combining the located words for the acronym
segments.
[0051] In particular, the expansion searcher module retrieves an
expansion list subset (e.g., "EXPANSION_LIST_SUBSET" as indicated
in FIG. 2) at step 84 and searches for an expansion within the
expansion list subset as described below at step 86. If an
expansion is found as determined at step 88, the expansion is
verified at step 96 as described below. When an expansion is not
found, module 18 determines the presence of additional expansion
list subsets at step 92. If additional expansion list subsets are
present for an expansion candidate, the next subset is retrieved
for processing at step 84 as described above. When each expansion
list subset for an expansion candidate has been processed without
identification of an expansion, module 18 determines the presence
of additional expansion candidates at step 102 as described above.
If additional expansion candidates exist (e.g., the set of words
following the acronym candidate), module 18 informs the expansion
candidate identifier module at step 106 to commence processing of
the next expansion candidate as described above. Otherwise, module
18 informs the acronym identifier module at step 108 to locate and
facilitate processing of the next potential acronym within the text
as described above. The process is basically repeated until an
expansion is identified within an expansion list subset as
determined at step 88 or until each expansion candidate (and
corresponding expansion list subsets) is processed as determined at
steps 92 and 102 as described above.
[0052] The manner in which the expansion searcher module searches
for an expansion within an expansion list subset is illustrated in
FIG. 7. Initially, the expansion searcher module searches for a
valid expansion corresponding to an acronym search candidate by
repeatedly cycling through the acronym search candidate to generate
acronym segments and locating those segments within words of a
search window (e.g., including a portion of the expansion list
subset). This process is repeated until each segment is located
within a search window or until the expansion list subset is
exhausted. A failure to locate an acronym segment within a search
window invokes backtracking or utilization of prior segments and
modified search windows to locate a match. Backtracking is utilized
since a greedy search technique is employed. When an acronym
segment is located in a search window word, the search window is
modified to exclude the located word (e.g., shifted to start at the
word immediately following the located word).
[0053] Specifically, the expansion searcher module retrieves a
character from the acronym search candidate at step 110 to form an
acronym segment. The expansion searcher module determines if the
segment is a delimiter (e.g., "-", "=", "/", etc.) at step 112.
This is typically accomplished by comparing the acronym segment to
a delimiter list similar to the manner described above. When the
segment contains a delimiter, the next character in the acronym
search candidate is retrieved for processing at step 110 as
described above in accordance with the presence of additional
acronym search candidate characters as determined at step 130.
[0054] If the segment does not contain a delimiter, the acronym
segment is examined for the presence of an ampersand (e.g.,
"&") at step 114. When the segment contains an ampersand, the
expansion list subset is examined for the presence of the
corresponding word "AND" at step 116. If the word "AND" is present
as determined at step 118, various variables or parameters are
updated at step 128 in order to process the next acronym search
candidate character. Basically, this includes storing the current
acronym segment (e.g., "&"), identified expansion word (e.g.,
"AND") and search window within respective data structures (e.g.,
stacks) and modifying the search window to remove the identified
expansion word. The search window is typically modified to begin
with the word in the expansion list subset that immediately follows
the identified expansion word. Thus, the search window is basically
moved forward within the subset to begin at the word following the
identified expansion word. Once the parameters are updated, the
next character in the acronym search candidate is retrieved for
processing at step 110 as described above in accordance with the
presence of additional acronym search candidate characters as
determined at step 130. When the word "AND" is not present within
the expansion list subset, the search with respect to the current
expansion list subset terminates (e.g., generally returning to step
88 of FIG. 6).
[0055] When the acronym segment is not an ampersand as determined
at step 114, the expansion search module determines the presence of
a word within a search window (e.g., including a portion of the
expansion list subset) that includes the acronym segment as an
initial portion or within an initial thirty percent of the word at
step 120. This process is further illustrated in FIG. 8. Initially,
the presence of an acronym segment within a search window of the
expansion list subset is determined. The search window size is
preferably three words and starts from the beginning of an
expansion list subset, but may be modified by a system
administrator. The search initially attempts to locate words
starting with the acronym segment. If this search fails, the search
is subsequently conducted to identify portions within a word that
contain the acronym segment. Specifically, a word within the search
window is retrieved at step 140. The search window preferably
includes the initial three words of the expansion list subset. The
retrieved word is examined to determine the presence of the acronym
segment as the starting portion of the word at step 142 (e.g.,
"Department" starts with segment "D" or "De"). If the word starts
with the acronym segment, expansion searcher module 22 indicates
the acronym segment is within the expansion list subset at step 154
and the search with respect to the current acronym segment
terminates (e.g., generally returning to step 122 of FIG. 7
described below). Otherwise, the word is compared to a list (e.g.,
"STOP-WORD LIST" as indicated in FIG. 2) of certain stop words
(e.g., "the", "and", "to", "he", "she", etc.) at step 144. These
words are basically those that are to be excluded from
consideration as an expansion. When the word is a stop word or
within the list, the next word is retrieved from the expansion list
subset at step 140 for processing in accordance with the presence
of additional words within the search window as determined at step
148.
[0056] If the word is not a stop word, the expansion searcher
module determines if a prior non-stop word has been skipped at step
146. When a prior non-stop word has been skipped, the initial
letter search is terminated in favor of the search for intermediate
word portions described below. If a prior non-stop word has not
been skipped, the next word is retrieved from the expansion list
subset at step 140 for processing as described above in accordance
with the presence of additional words within the search window as
determined at step 148. This process is basically repeated until a
word is found as determined at step 142 or until each word within
the search window has been processed as determined at step 148 as
described above.
[0057] When words within the search window do not contain the
acronym segment as the starting portion, the words are examined to
determine if the acronym segment resides within an initial thirty
percent of a word. In particular, a word is retrieved from the
search window at step 150. The retrieved word is examined to
determine the presence of the acronym segment within an initial
thirty percent of the word. This enables identification of
expansions where the acronym utilizes intermediate portions of a
word. For example, the acronym "UXO" for "Unexploded Ordinance"
contains the acronym segment "X" within the word "exploded". The
word "Unexploded" is split by the prefix "un" to the terms "un" and
"exploded" during processing as described above. However, these
terms are combined subsequent the search to produce an expansion as
described below. If the word contains the acronym segment,
expansion searcher module 22 indicates the acronym segment is
within the expansion list subset at step 154 and the search with
respect to the current acronym segment terminates as described
above. Otherwise, the next word is retrieved from the search window
at step 150 for processing as described above in accordance with
the presence of additional words within the search window as
determined at step 156. The process basically repeats until a word
is found as determined at step 152 or until each word in the search
window has been processed as determined at step 156 as described
above. When each word within the search window has been processed
and determined to lack the acronym segment, module 22 indicates the
acronym segment is lacking within the expansion list subset at step
158 and the search with respect to the current acronym segment
terminates as described above.
[0058] Referring back to FIG. 7, if the search for an acronym
segment is successful as determined at step 122, various parameters
are updated at step 128 in order to process the next acronym search
candidate character. Basically, this includes storing the current
acronym segment, identified expansion word and search window within
respective data structures (e.g., stacks) and modifying the search
window to remove the identified expansion word as described above.
The search window is typically modified to begin with the word in
the expansion list subset that immediately follows the identified
expansion word. Thus, the search window is basically moved forward
within the subset to begin at the word following the identified
word. Once the parameters are updated, the next acronym search
candidate character is retrieved for processing at step 110 as
described above in accordance with the presence of additional
acronym search candidate characters as determined at step 130.
[0059] When the expansion search is unsuccessful as determined at
step 122, the expansion searcher module performs backtracking at
step 124 to identify an expansion for the acronym segment. This
includes combining the current acronym segment and search window
with prior iterations to determine a match. Backtracking basically
enables identification of a single expansion term corresponding to
plural acronym characters (e.g., the acronym "OPORD" for "Operation
Order"). The backtracking process is further illustrated in FIG. 9.
Specifically, expansion searcher module 22 determines at step 160
if any prior searches for acronym segments were successful in order
to provide information to perform a backtrack. If prior information
is unavailable (e.g., no acronym segments stored), an unsuccessful
indication is provided at step 184 and the search with respect to
the current expansion list subset terminates (e.g., generally
returning to step 126 of FIG. 7 described below). Similarly, module
22 determines at step 162 if the current acronym segment is the
same as the first acronym segment stored. When this is the case, an
unsuccessful indication is provided at step 184 and the search
terminates with respect to the current expansion list subset as
described above to avoid spurious identifications. If there is
sufficient information and the current acronym segment is valid,
module 22 retrieves the prior search window at step 164. This is
basically accomplished by moving the search window back to begin at
the location within the expansion list subset utilized for the
prior search. Module 22 further retrieves the prior acronym segment
stored within the data structure (e.g., stack) and appends the
current segment to the retrieved segment at step 166, thereby
producing a combined segment. The expansion searcher module
searches the prior search window for the combined segment at step
168 in the manner described above for FIG. 8. If the search is
successful as determined at step 170, a successful indication is
provided at step 182 and a search with respect to the next acronym
segment commences (e.g., generally returning to step 126 of FIG. 7)
as described below.
[0060] Otherwise (e.g., the search is unsuccessful), a subsequent
level of backtracking is performed. In particular, expansion
searcher module 22 determines at step 172 if at least two prior
searches were successful (e.g., at least two segments are present
on the data structure) in order to provide sufficient information
to perform an additional backtrack. When sufficient information is
unavailable, an unsuccessful indication is provided at step 184 and
the search terminates with respect to the current expansion list
subset as described above. If the information is available, module
22 retrieves the two prior expansions and incorporates them into
the current search window at step 174 to provide a combined search
window. The expansion searcher module further retrieves the two
prior acronym segments and appends the current segment to those
segments at step 176, thereby forming a combined segment. The
expansion searcher module searches the combined search window for
the combined segment at step 178 in the manner described above for
FIG. 8. If the search is successful as determined at step 180, a
successful indication is provided at step 182 and the search with
respect to the next acronym segment commences as described below.
Otherwise, an unsuccessful indication is provided at step 184 and
the search with respect to the current expansion list subset
terminates as described above.
[0061] An example of the backtracking process is described with
respect to an acronym "OPORD" (e.g., corresponding to "Operation
Order") and an expansion list subset of "Operation, Order, was,
issued, to, the, unit". The initial acronym segment is "O", and a
search for that segment in a search window of "Operation, Order,
was" is successful in identifying "Operation" as an expansion word.
The acronym and expansion word are stored and the search window is
updated as described above. The next acronym character is retrieved
and the acronym segment is "P" with a search window of "Order, was,
issued". The search for this segment fails since no words contain
an initial portion with the acronym segment. The system performs
backtracking by combining the current and previous acronym segments
to obtain the combined segment "OP", while the search window is
updated to include the prior window (e.g., "Operation, Order,
was"). A search for the combined segment succeeds since "OP" is
located within "Operation". The process continues in this fashion
to identify "Order" as the expansion word for the segment "ORD".
Thus, an initial backtrack level combines current and prior acronym
segments and utilizes the prior search window, while a subsequent
level combines the current and two prior acronym segments and
utilizes the current search window modified to include the two
previous expansion words.
[0062] Referring back to FIG. 7, if the backtrack search is
successful as determined at step 126, expansion searcher module 22
updates parameters at step 128 and subsequently retrieves the next
acronym search candidate character for processing at step 110 in
accordance with the presence of additional characters as determined
at step 130 as described above. When the backtrack search fails,
the search terminates with respect to the current expansion list
subset as described above (e.g., generally returning to step 88 of
FIG. 6). The process continues until a search (with backtracking)
fails as determined at step 126 or until each acronym search
candidate character has been processed as determined at step 130 as
described above.
[0063] With reference to FIG. 6, once expansion list subsets have
been processed without identifying an expansion as determined at
step 92, expansion searcher module 22 determines the presence of
additional expansion candidates at step 102 as described above. If
additional expansion candidates exist (e.g., the set of words
following the acronym candidate), module 22 informs the expansion
candidate identifier module at step 106 to commence processing of
the next expansion candidate to search for an expansion as
described above. Otherwise, no expansion has been identified for
the current acronym candidate and module 22 informs the acronym
identifier module at step 108 to locate and facilitate processing
of the next potential acronym within the text as described
above.
[0064] Once an expansion is identified as determined at step 88,
the expansion (e.g., "VALID_EXPANSION_CANDIDATE" as indicated in
FIG. 2) is verified at step 96 by expansion filter module 24 (FIG.
2). The verification applies a set of rules to the identified
expansion to validate that expansion and to filter spurious
expansions. This enhances the accuracy of expansions identified by
the system. In particular, simulations indicated that approximately
eighteen percent of expansions were spurious. The rules applied for
the verification are derived based on properties of those spurious
expansions. The values for particular rules are measured during the
expansion search described above, while the rules and values are
applied during verification. The rules, by way of example, examine
various attributes of an expansion including: the presence of
objectionable words within the expansion; the numbers within the
expansion correspond to those within the acronym; presence of an
objectionable word as the first expansion word; presence of an
adverb as the last expansion word; presence of skipped stop and
non-stop words within the expansion; presence of two words within
the expansion containing an acronym segment in an intermediate
portion; presence of the acronym within the expansion; and presence
of delimiters within the expansion.
[0065] This verification process is further illustrated in FIG. 10.
Initially, expansion filter module 24 receives an acronym candidate
(e.g., original form prior to preprocessing) and corresponding
expansion at step 190. The expansion filter module examines the
expansion at step 192 for the presence of an objectionable word
(e.g., "his", "her", "with", etc.). This is typically accomplished
by comparing expansion words to an objectionable word list (e.g.,
"OBJECTIONABLE_WORD_LIST" as indicated in FIG. 2). If an
objectionable word is present, module 24 indicates an invalid
expansion at step 210.
[0066] When objectionable words are absent from the expansion
(e.g., thereby satisfying a condition for a valid expansion), the
expansion filter module determines if numbers, preferably greater
than four, within the acronym correspond with numbers within the
expansion (e.g., "RM-98" with expansion "Rim Pac '98") at step 194.
If numbers do not correspond, module 24 indicates an invalid
expansion at step 210 as described above.
[0067] When the numbers correspond (e.g., thereby satisfying
another condition for a valid expansion), the expansion filter
module determines if the first word in the expansion is an
objectionable first word (e.g., `And`, `To`, etc.) at step 196.
This is typically accomplished by comparing the first word to a
list of objectionable first words (e.g., "FIRST_WORD_OBJECTIONABLE
_LIST" as indicated in FIG. 2). This list and the objectionable
word list described above are preferably mutually exclusive. If the
first word of the expansion is objectionable, module 24 indicates
an invalid expansion at step 210 as described above.
[0068] When the first expansion word is not objectionable (e.g.,
thereby satisfying another condition of a valid expansion), the
expansion filter module determines if the last expansion word is a
verb or adverb at step 198. This is typically accomplished by
examining the part-of-speech tag associated with that word. If the
last expansion word is a verb or adverb, module 24 indicates an
invalid expansion at step 210 as described above.
[0069] When the last expansion word is not a verb or adverb (e.g.,
thereby satisfying another condition of a valid expansion), the
expansion filter module determines if a stop word and non-stop word
each skipped during the expansion search described above are
present next to each other within the expansion at step 200. If
these words are each present in the expansion, module 24 indicates
an invalid expansion at step 210 as described above.
[0070] When the expansion does not contain skipped stop and
non-stop words (e.g., thereby satisfying another condition for a
valid expansion), the expansion filter module determines at step
202 if the expansion includes two words each containing a
respective acronym segment within an intermediate portion. In other
words, module 24 determines the presence of two words identified
based on the intermediate portion search (e.g., within the initial
thirty percent of the word) described above. If these words are
present in the expansion, module 24 indicates an invalid expansion
at step 210 as described above.
[0071] When the expansion does not contain the intermediate portion
search words (e.g., thereby satisfying another condition for a
valid expansion), the expansion filter module determines the
presence of the acronym within the expansion at step 204. If the
acronym is present within the expansion, module 24 indicates an
invalid expansion at step 210 as described above.
[0072] When the acronym is not present within the expansion (e.g.,
thereby satisfying another condition for a valid expansion), the
expansion filter module determines the presence of delimiters
(e.g., ":", "/", "(", "{", etc.) within the expansion at step 206.
This is typically accomplished by comparing the expansion word
characters to a delimiter list (e.g., "EXPANSION_DELIMITER_LIST" as
indicated in FIG. 2), which may be user-configurable. If a
delimiter is not present (e.g., thereby indicating satisfaction of
each condition for a valid expansion), module 24 indicates a valid
expansion at step 208. Otherwise, an invalid expansion is indicated
at step 210 as described above.
[0073] Referring back to FIG. 6, if an expansion is valid as
determined at step 98, the expansion filter module merges terms
that were split (e.g., by hyphens, slashes, prefixes, etc.) prior
to the search and during the expansion candidate processing
described above. The acronym and resulting expansion are produced
at step 104. When an expansion is invalid, the expansion searcher
module determines the presence of additional expansion list subsets
at step 92 as described above. If additional expansion list subsets
are present, the next expansion list subset is retrieved for
processing at step 84 as described above. When each expansion
subset list has been processed, the expansion searcher module
determines the presence of additional expansion candidates at step
102 as described above. If additional expansion candidates exist
(e.g., the set of words following the acronym candidate), the
expansion searcher module informs the expansion candidate
identifier module at step 106 to commence processing of the next
expansion candidate to search for an expansion as described above.
The search process basically repeats until a valid expansion is
found as determined at step 98 or until each expansion list subset
within each expansion candidate is processed as determined at step
102 as described above. Once a valid expansion is produced at step
104, or each expansion candidate has been processed as determined
at step 102, the expansion searcher module informs the acronym
identifier module at step 108 to locate and facilitate processing
of the next potential acronym within the text as described
above.
[0074] The techniques and heuristics described above enhance system
performance. For example, simulations indicated that the present
invention incorrectly identifies expansions for three to five
acronyms within every one hundred acronyms appearing in text. In
contrast, the Acrophile system incorrectly identified expansions
for thirty to fifty-five acronyms that occur in text.
[0075] The present invention may be utilized for various
applications. For example, the present invention may be used to
enhance digital library management, document management, regular f
all text search engines and technical information portals. As a
stand-alone application, the present invention may assist document
authors and publishers (e.g., military doctrine publishers, etc.)
to compile a list of acronyms and corresponding expansions that is
complete and accurate.
[0076] In the form of an embedded application, the present
invention can improve indexing and search applications that operate
on documents and web pages. For example, a search engine, such as
Verity (available from Verity, Inc.), may utilize the present
invention to expand a search request with an acronym (e.g., "HDP")
to include the expansion (e.g., "High Density Polymer"), thereby
encompassing documents within the search that do not explicitly
include the expansion. Alternatively, the search engine may employ
the expansion and uncover documents containing the acronym. These
capabilities are especially desirable in technical, engineering,
medical and scientific disciplines. Thus, the present invention
offers technology that can serve a large market need for digital
library management and search.
[0077] It will be appreciated that the embodiments described above
and illustrated in the drawings represent only a few of the many
ways of implementing an acronym extraction system and method of
identifying acronyms and extracting corresponding expansions from
text.
[0078] The computer system of the present invention may be
implemented by any quantity of any personal or other type of
computer or processing system (e.g., IBM-compatible, Apple,
Macintosh, laptop, palm pilot, microprocessor, etc.). The computer
system may include any commercially available operating system
(e.g., Windows, OS/2, Unix, Linux, etc.), any commercially
available and/or custom software (e.g., communications software,
acronym tool software, etc.) and any types of input devices (e.g.,
keyboard, mouse, microphone, voice recognition, etc.). It is to be
understood that the software of the acronym expansion system may be
implemented in any desired computer language, and could be
developed by one of ordinary skill in the computer and/or
programming arts based on the functional description contained
herein and the flow charts illustrated in the drawings. Further,
any references herein of software performing various functions
generally refer to computer systems or processors performing those
functions under software control. The computer system may
alternatively be implemented by hardware or other processing
circuitry. The various functions of the acronym expansion tool may
be distributed in any manner among any quantity (e.g., one or more)
of hardware and/or software modules or units, computer or
processing systems or circuitry, where the computer or processing
systems may be disposed locally or remotely of each other and
communicate via any suitable communications medium (e.g., LAN, WAN,
Intranet, Internet, hardwire, modem connection, wireless, etc.).
The software and/or algorithms described above and illustrated in
the flow charts and diagrams may be modified in any manner that
accomplishes the functions described herein.
[0079] The acronym expansion tool may accommodate any quantity of
any type of document containing text in any desired format (e.g.,
ASCII, plain text, any word processor or other application format,
etc.). The acronym expansion tool may identify any types of
shortened forms of words or phrases (e.g., acronyms, abbreviations,
phrasal abbreviations, etc.) of any quantity, type or length in any
type of text or document and extract the corresponding expansion,
and accommodate documents or text of any language (e.g., languages
of Indo-European origin including English, etc.) with suitable
adaptations of the system language specific components and data
resources (e.g., part-of-speech tagger, objectionable word list,
stop word list, etc.). The acronym expansion tool may be
implemented as a separate stand-alone program or software module or
may be in the form of an embeddable system (e.g., within another
system, as an embeddable software component within other software,
etc.). The computer system may further function as a server in
communication with client or user systems via any communications
medium (e.g., network, WAN, LAN, Internet, Intranet, modem, etc.)
to process client requests. The acronym tool software may be
available on a recorded medium (e.g., magnetic or optical mediums,
magneto-optic mediums, floppy diskettes, CD-ROM, memory devices,
etc.) for use on stand-alone systems or systems connected by a
network or other communications medium, and/or may be downloaded
(e.g., in the form of carrier waves, packets, etc.) to systems via
a network or other communications medium.
[0080] The acronym tool may utilize any conventional or other
speech classifier or tagger for any desired language, and may
classify any portion of text in any manner and at any point during
processing (e.g., initially, during acronym identification, during
expansion verification, when a classification is utilized, etc.).
The speech information may be provided in any desired fashion
(e.g., any symbols or character strings to indicate the part of
speech, a data structure or record to store a word and
classification, etc.). The various resource lists (e.g., stop-word,
delimiter, prefix, objectionable word, etc.) and the lists
containing processed data (e.g., expansion candidates, expansion
lists and subsets, expansion candidate with split words, etc.) may
be implemented by any quantity of lists of any desired data
structure (e.g., file, array, linked-list, stack, queue, etc.) and
may include any desired words or information. The resource lists
may further be manipulable by a user or system administrator.
[0081] Identification of potential acronyms may be based on any
desired criteria. The potential acronym may be identified based on
compliance with all or any portion of the criteria. The thresholds
for the criteria may be set to any desired values (e.g., any amount
of capitalization or numeral proportion, any desired length, etc.)
to identify words as potential acronyms. The delimiter list may be
implemented by any quantity of lists and include any quantity of
any types of delimiters or other characters or symbols. The
exclusion list may be implemented by any quantity of lists and
include any quantity of any types of words, symbols, acronyms or
other information.
[0082] Any quantity of sets of any desired length may be retrieved
from the text at any desired location (e.g., preceding and/or
following the potential acronym, portions including the potential
acronym, etc.) with respect to a potential acronym to serve as an
expansion candidate. The positional information may include any
type of information (e.g., pointer, array index, etc.) to identify
the location of a word within the text. The acronym expansion tool
may search any quantity or portion of the expansion candidates to
identify an expansion. The tool may further search each of the
candidates irrespective of prior identification of an expansion,
and may remove any duplicate expansions.
[0083] The expansion candidate words may be spilt into any quantity
of terms based on any desired conditions (e.g., delimiters,
punctuation, prefixes, syllables, word length, etc.). The prefix
list may be implemented by any quantity of lists and include any
quantity of any portions of words, symbols or other characters.
Similarly, the delimiter list may be implemented by any quantity of
lists and include any quantity of any portions of words, symbols or
other characters. The words may be split in any fashion with the
terms including any desired portions of the split word. The words
may be split based on any desired lists containing any desired
information.
[0084] The acronym candidate and expansion candidate or list may be
processed in any manner for searching (e.g., numeral expansion,
plural or singular forms, subsets, removal of particular
characters, symbols or delimiters, etc.). The numeral expansion may
be performed for any types of numerals (e.g., Roman, etc.) within
any range and for any characters or symbols. The acronym candidate
may be modified in any fashion to convert from plural to singular
form (e.g., remove any type of plural suffix and/or replace with a
singular suffix, word replacement, utilize any desired lists for
suffix and/or word replacement, etc.). The acronym and expansion
candidates may alternatively be utilized for searching in their
original or any prior form (e.g., without preprocessing, without
word splitting, etc.).
[0085] The expansion candidates or lists may be utilized to derive
any quantity of subsets based on any desired conditions (e.g.,
starting with or including within any portion any letter or symbols
of the acronym candidate, etc.). The subsets may include any
quantity of any words (e.g., preceding, following and/or including
the identified word) within the expansion list. The acronym
expansion tool may search any quantity or portion of the subsets to
identify an expansion. The tool may further search each of the
subsets irrespective of prior identification of an expansion, and
may remove any duplicate expansions. The search window may be of
any desired length (e.g., to include any portion of the subset or
expansion candidate) and may be set by a user or system
administrator. The search window may start at any desired location
within an expansion subset or candidate and may be updated in any
desired fashion (e.g., to start at any location for a subsequent
search, adjust search window length or size, etc.).
[0086] The acronym expansion tool may accommodate any types of
symbols or delimiters within an acronym candidate (e.g.,
delimiters, ampersands, etc.), and may identify any types of
expansions for those symbols or delimiters (e.g., "AND" for the
symbol "&", etc.). The acronym expansion tool may search any
portions of search window words for an acronym segment or character
(e.g., starting letter, any initial portion, any portion within the
word, etc.) to identify an expansion word. The stop-word list may
be implemented by any quantity of lists containing any quantity of
words to be excluded from consideration as an expansion word. Any
quantity of stop or non-stop words may be skipped during a search
without indicating a failure to identify an expansion.
[0087] The acronym expansion tool may perform any quantity of
levels of backtracking provided sufficient information is
available. A backtrack may utilize any quantity or combinations of
any portions of prior acronym segments and any quantity or
combinations of any portions of prior search windows. The acronym
segment portions may be combined or arranged in any fashion to
conduct an expansion search, while the search windows may be
combined or arranged in any fashion to similarly conduct an
expansion search. The acronym segments, search windows and
identified expansion words may be saved in any fashion via any
suitable data structures (e.g., file, array, linked list, stack,
queue, etc.).
[0088] The rules for verification may include any quantity of any
type of rules, criteria or conditions to verify an identified
expansion. The identified expansion may be verified based on
compliance with all or any portion of the criteria. The thresholds
for the criteria may be set to any desired values (e.g., any
numbers correspond, any quantity of skipped stop and non-stop
words, any quantity of words containing the acronym segment in an
intermediate portion, last or other expansion word may be any type
of grammatical classification, any portion of the acronym resides
within the expansion, etc.) to validate any identified expansion.
The delimiter list may be implemented by any quantity of lists and
include any quantity of any types of delimiters or other characters
or symbols. The objectionable word list and objectionable first
word list may each be implemented by any quantity of lists and
include any quantity of any types of words, symbols, acronyms or
other information. These lists may be implemented by a common list
or be mutually exclusive. The acronym expansion tool may
alternatively produce an acronym and identified expansion without
the verification.
[0089] The acronym expansion tool may interact with a user and
receive input text in any desired fashion. For example, the tool
may include a user interface (e.g., Graphical User Interface, etc.)
to receive file information concerning a file containing the text.
The user may enter information via the computer system input device
(e.g., mouse, voice recognition, etc.), where the interface may
display directories to enable a user to select a file. When the
tool is embedded in another application or system, the tool may
receive the text or a file (e.g., containing the text) directly or
an indication of a file or other storage area where the text may be
found for processing. The acronyms and corresponding expansions may
be provided to the user, software or system in any desired form
(e.g., display, hardcopy, stored in a file or other data structure
in any desired format, etc.) and/or arranged in any manner or
format.
[0090] The acronym expansion tool may process and transfer
information between modules in any desired fashion. For example,
the potential acronyms may be processed sequentially or any
quantity or all of the potential acronyms may be processed at a
given time. Similarly, one or any other quantity of expansion
candidates or subsets may be processed at a given time. The tool
may process documents upon receipt of text or at any time in a
batch mode. In addition, the computer system may function as a
server to process client requests for acronym extraction within
supplied documents or text.
[0091] The present invention may be utilized for various
applications. For example, the present invention may be used to
enhance digital library management, document management, regular
full text search engines, technical information portals and voice
responsive and/or voice and/or speech synthesis systems. As a
stand-alone application, the present invention may assist document
authors and publishers (e.g., military doctrine publishers, etc.)
to compile a list of acronyms and corresponding expansions that is
complete and accurate. In the form of an embedded application, the
present invention can improve indexing and search applications that
operate on documents and web pages.
[0092] From the foregoing description, it will be appreciated that
the invention makes available a novel acronym extraction system and
method for identifying acronyms and extracting corresponding
extractions from text, wherein acronyms are identified within text
and corresponding expansions are extracted with enhanced
accuracy.
[0093] Having described preferred embodiments of a new and improved
acronym extraction system and method for identifying acronyms and
extracting corresponding extractions from text, it is believed that
other modifications, variations and changes will be suggested to
those skilled in the art in view of the teachings set forth herein.
It is therefore to be understood that all such variations,
modifications and changes are believed to fall within the scope of
the present invention as defined by the appended claims.
* * * * *