ÀÚ¿¬¾îó¸® ½Ã½ºÅÛ

 

ÀÚ¿¬¾îÀÇ ÀÌ·ÐÀûÀÎ ¿¬±¸¸¸ÀÌ ¾Æ´Ï¶ó, ÀÌ°ÍÀ» ÀÀ¿ëÇÏ¿© ½ÇÁ¦·Î »ç¿ëÇÏ´Â ÀÚ¿¬¾îó¸® ½Ã½ºÅÛÀ» ¸¸µé·Á´Â ½Ãµµµµ È°¹ßÇÏ´Ù. ÇöÀç±îÁö °³¹ßµÈ ÀÚ¿¬¾î ½Ã½ºÅÛÀº,

µîÀ¸·Î ºÐ·ùµÈ´Ù.

ÀÚ¿¬¾îó¸® ½Ã½ºÅÛÀÇ Ãʱ⠼º°øÀÇ ¿¹·Î¼­´Â Winograd ÀÇ SHRDLU°¡ À¯¸íÇÏ´Ù. ÀÌ°ÍÀº È­¸é»óÀÇ ³ª¹«½×±â¸¦ ÀÚ¿¬¾îÀÎ ¿µ¾î·Î Á¶ÀÛÇÑ´Ù. ±× ÈÄ Á¤½ÅºÐ¼®À» ÇàÇÏ´Â ELIZA, ¿©Çà°èȹÀÇ ÁúÀÇÀÀ´ä ½Ã½ºÅÛÀÎ GUS, ¿µ¹®Çؼ® ½Ã½ºÅÛÀÎ MARGIE, ´ëÈ­Çؼ® ½Ã½ºÅÛÀÎ TOPLE µî ¿©·¯ °¡Áö ÀÚ¿¬¾îó¸® ½Ã½ºÅÛÀÌ °³¹ßµÇ¾ú´Ù.

ÀÚ¿¬¾î·Î µ¥ÀÌÅͺ£À̽º¸¦ Á¶ÀÛÇÏ´Â ½Ã½ºÅÛÀÇ °³¹ßµµ È°¹ßÈ÷ ÁøÇàµÇ°í ÀÖ´Ù. SRI°¡ °³¹ßÇÑ ´ë±Ô¸ð ºÐ»êµ¥ÀÌÅͺ£À̽º Á¶È¸ ½Ã½ºÅÛÀÎ LADDER, °ü°è¸ðµ¨(relational model)ÀÇ Ã¢½ÃÀÚÀÎ ÄÚµå(Codd)°¡ °³¹ßÇÑ RENDEZVOUS, ÁöÁúÇп¡ °üÇÑ ÁúÀÇÀÀ´ä ½Ã½ºÅÛ LUNAR, IBMÀÌ °³¹ßÇÑ ÁúÀÇÀÀ´ä ½Ã½ºÅÛ REQUEST, Ç×°øÆí¿¡ °üÇÑ ´ë±Ô¸ð °ü°èÇü(relational) µ¥ÀÌÅͺ£À̽º¸¦ Ãë±ÞÇÏ´Â ÁúÀÇ ÀÀ´ä ½Ã½ºÅÛ PLANE µîÀÌ ´ëÇ¥ÀûÀÌ´Ù.

1970³â ÀÌÈÄ¿¡´Â ÀÚ¿¬¾îó¸® ½Ã½ºÅÛÀÌ »óÇ°È­µÇ±â ½ÃÀÛÇß´Ù. INTELLECT´Â ÀÚ¿¬¾î¸¦ Ãë±ÞÇÒ ¼ö ÀÖ´Â ÃÖÃÊÀÇ »óÇ°À¸·Î, ¾ÆƼÇÇ¼È ÀÎÅÚ¸®Àü½º»ç(Artificial Intelligence Inc.)°¡ ¹ß¸ÅÇÑ »óÇ°ÀÇ µ¥ÀÌÅͺ£À̽º¿ë ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽ºÀÌ´Ù. ÇÁ¸® ¾î¼Ò½Ã¿¡ÀÌÆ®»ç(Free Associate Inc.)ÀÇ THEMISµµ ÀÚ¿¬¾î µ¥ÀÌÅͺ£À̽º ÀÎÅÍÆäÀ̽º·Î, »õ·Î¿î ´Ü¾î¸¦ ½Ã½ºÅÛ¿¡ Ãß°¡ÇÒ ¼öµµ ÀÖ´Ù.  SAVVY´Â PC¿ë ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º, STRAIGHT TALK´Â ¿öµåÇÁ·Î¼¼¼­¿ë ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽ºÀÌ´Ù. NATURAL LINK´Â ¸Þ´º Áß¿¡¼­ ±¸(Ï£)¸¦ ¼±ÅÃÇÏ¿© º¹ÀâÇÑ Áú¹®À» ¸¸µå´Â ½Ã½ºÅÛÀÌ´Ù. ´Ù¸¥ ½Ã½ºÅÛó·³ Áú¹®À» ÇÏÁö ¾Ê´Â´Ù´Â °ÍÀº ¾Æ´Ï´Ù. ÃÖÃÊ¿¡ Q&A¶ó´Â Á¾ÇÕ ¼ÒÇÁÆ®¿þ¾î ÆÐÅ°Áö´Â ¿öµåÇÁ·Î¼¼¼­, µ¥ÀÌÅͺ£À̽º¿¡ ´ëÇÑ ¸í·É¾î¸¦ ÀÚ¿¬¾îÀÎ ¿µ¾î·Î ÇÒ ¼ö ÀÖµµ·Ï ±¸¼ºÇÏ¿© ¼º°øÇÑ ½Ã½ºÅÛÀÌ´Ù.

Çѱ¹¿¡¼­´Â Çѱ¹ÀüÀÚÅë½Å¿¬±¸¼Ò(ETRI; Electronic & Telecommunication Research Institute)°¡ ÁÖµµÇÏ´Â Á¤ºÎ ±¹Ã¥¿¬±¸°³¹ß°úÁ¦ÀÎ, Áö´ÉÇü ÄÄÇ»ÅÍ¿¡¼­ ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º¿¡ °üÇÑ ¿¬±¸¸¦ Çѱ¹°úÇбâ¼ú¿ø ÀΰøÁö´É¼¾ÅÍ¿Í °øµ¿À¸·Î ÇÑ ¹Ù ÀÖ´Ù.
 ½Ç¿ëÈ­µÈ ½Ã½ºÅÛÀ¸·Î¼­´Â, Çѱ¹°úÇбâ¼ú¿ø Àü»êÇаú Ãֱ⼱ ±³¼öÆÀ¿¡¼­ °³¹ßÇÑ Çѱ¹¾î ÅؽºÆ®ÀÇ ÀÚµ¿»öÀΰËÃâ±â KAIS(Korean Automatic Indexing System)¿Í, ÀÚµ¿Ã¶ÀÚ°Ë»ç ¹× ±³Á¤½Ã½ºÅÛ(Hspe 11)ÀÌ ÀÚ¿¬¾îó¸® ½Ã½ºÅÛÀÇ ½Ç¿ëÈ­·Î¼­´Â È¿½ÃÀÌ´Ù.
 KAIS´Â 1991³â¿¡, Hspe 11´Â 1989³â¿¡ ½ÃÁ¦Ç°ÀÌ ¿Ï¼ºµÇ¾ú´Ù. ÀÌ ¿Ü¿¡µµ ±Ý¼º¼ÒÇÁÆ®¿þ¾îÀÇ Ã¶ÀÚ°Ë»ç ½Ã½ºÅÛÀÌ »ó¿ëÈ­µÇ´Â µî, 1990³â ÇϹݱâºÎÅÍ ÀÌ¿¡ ´ëÇÑ ¿¬±¸°¡ ¹ø¼ºÇϱ⠽ÃÀÛÇÏ¿´´Ù.

ÀÚ¿¬¾î(¿µ¹®) ó¸® ½Ã½ºÅÛ

ºÐ   ¾ß

½Ã½ºÅÛ¸í

°³          ¹ß

±â             ´É

ÁúÀÇ ÀÀ´ä
½Ã½ºÅÛ

SHRDLU

MIT

³ª¹«½×±âÀÇ QA½Ã½ºÅÛ

GUS

Á¦·Ô½º»ç

¿©Çà°èȹÀÇ QA½Ã½ºÅÛ

ELIZA

ditto

Á¤½ÅºÐ¼®ÀÇ QA½Ã½ºÅÛ

SCHOLOR

MIT

Áö¸®ÇнÀ¿ë CAI

¹®Á¦ÇØ°á
½Ã½ºÅÛ

STUDENT

MIT

»ê¼ö ¹®Á¦ÇØ°á ½Ã½ºÅÛ

Newton

MIT

¹°¸® ¹®Á¦ÇØ°á ½Ã½ºÅÛ

Isaoc

Åػ罺´ë

Áö¸® ¹®Á¦ÇØ°á ½Ã½ºÅÛ

¹®ÀåÇؼ®
½Ã½ºÅÛ

MARGIE

½ºÅÄÆÛµå´ë

¿µ¹®Çؼ® ½Ã½ºÅÛ

TOPLE

MIT

´ëÈ­ÀÌÇØ ½Ã½ºÅÛ

LINGOL

MIT

¿µ¹®Çؼ® ½Ã½ºÅÛ

µ¥ÀÌÅͺ£À̽º

°Ë»ö

½Ã½ºÅÛ

LADDER

SRI

´ë±Ô¸ð ºÐ»ê µ¥ÀÌÅͺ£À̽º Á¶È¸ ½Ã½ºÅÛ

RENDEZOUS

CODD

°ü°è µ¥ÀÌÅͺ£À̽º Á¶ÀÛ ½Ã½ºÅÛ

LUNAR

BBN

ÁöÁúÇп¡ °üÇÑ ÁúÀÇÀÀ´ä ½Ã½ºÅÛ

REQUEST

IBM

µ¥ÀÌÅͺ£À̽º °Ë»ö ½Ã½ºÅÛ

PLANS

Àϸ®³ëÀÌ´ë

Ç×°øÆí¿¡ °üÇÑ ÁúÀÇÀÀ´ä ½Ã½ºÅÛ

»ó¿ë

½Ã½ºÅÛ

INTELLECT

AI

main flameÀÇ µ¥ÀÌÅͺ£À̽º¿ë ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º

SAVVY

¿¡Äí½ºÄ«¸®¹Ù
Å×Å©³î·ÎÁö

°³ÀÎÄÄÇ»ÅÍ¿ë ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º

STRAIGHT TALK

µñŸÆù

¿öµåÇÁ·Î¼¼¼­¿ÍÀÇ ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º

THEMIS

ÇÁ¸® ¾î¼Ò½Ã¿¡ÀÌÆ®

VAX-Orode¿ë ÀÚ¿¬¾î µ¥ÀÌÅͺ£À̽º ÀÎÅÍÆäÀ̽º

SUPER-NATURAL

¸¶ÀÌÅ©·Îµ¥ÀÌÅÍ

Á¤º¸Ã³¸®½Ã½ºÅÛ°úÀÇ ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º

NATURAL LINK

AI

¸Þ´º¿¡ ÀÇÇÑ ÀÚ¿¬¾î DBMS ÀÎÅÍÆäÀ̽º

PEARL

ÄÚ±×´ÏƼºê
½Ã½ºÅÛ

Áö½Äº£À̽º¿ÍÀÇ ÀÚ¿¬¾î ÀÎÅÍÆäÀ̽º

 

 

ÀÚ¿¬¾î½Ã½ºÅÛÀÇ ±¸¼º

 

¿ª»çÀûÀÎ Natural Language Understanding System

General Syntactic Processor (GSP)   SAD-SAM   BASEBALL   SIR   

STUDENT   ELIZA   LUNAR      MARGIE   SAM   PAM   LIFER

General Syntactic Processor (GSP)

A versatile system for the parsing and generation of strings in NL. GSP can directly emulate several other syntactic processors, including Woods' ATN grammar. It is not in itself an approach to language processing, but a system in which various approaches can be described. The basic idea is a chart that represents both the grammar and the input sentence as a modified tree (modified by making it binary and interchanging nodes and arcs). Chart manipulation mechanisms operate on the grammar itself.

SAD-SAM

[Lindsay, 1963] Syntactic Appraiser and Diagrammer -- Semantic Analyzing Machine. Programmed by Robert Lindsay in 1963 at CMU. It used an basic English vocabulary (1,700 words) and followed a context-free grammar. It parsed input from left to right, built derivation trees, and passed them to SAM, which extracted the semantically relevant information to build family trees and find answers to questions.

BASEBALL

[Bert Green, 1963] An information retrieval program with a large database of facts about all American League games over a given year. It accepted input questions from the user, limited to one clause with no logical connectives.

SIR

[Bertram Raphael, 1968]

Semantic Information Retrieval system, it was a prototype "understanding" machine, since it could accumulate facts and then make deductions about them in order to answer questions.

STUDENT

[Daniel Bobrow, 1968] STUDENT was a pattern-matching natural language program written by Bobrow as his doctoral thesis work at MIT. It could solve high-school level algebra story problems.

LUNAR

[William Woods, 1973] LUNAR answered questions about the rock samples brought back from the moon using two databases -- the chemical analyzes and the literature references. Specifically, it helped geologists access, compare, and evaluate chemical analysis data on moon rocks and soil composition obtained from the Apollo-11 mission. It operated by translating a question entered in English into an expression in a formal query language. The translation was done with an ATN parser coupled with a rule-driven semantic interpretation procedure.

MARGIE

[Schank, 1973] Meaning Analysis, Response Generation, and Inference on English system, developed at Stanford in 1973. Provided an intuitive model of the process of natural language understanding; see Conceptual Parsing above.

MARGIE consisted of three components. The first, the conceptual analyzer, converted English sentences into a CD representation using production-like rules called "requests." The middle phase was an inference system that accepted CD propositions and deduced facts from it, given the current system memory. Inference knowledge was represented in a semantic net. (So, for example, if told John hit Mary the system might infer Mary might get hurt.)

The final component was a text-generation module that took internal CD representations and converted them into English-like output.

SAM

[Schank] Script Applier Mechanism. Makes use of frame-like data structures called scripts, which represent stereotyped sequences of events, to understand simple stories. Prototype frames make it possible to use expectations about the usual properties of known concepts and about what typically happens in a variety of familiar situations to help understand sentences about those objects and situations.

PAM

[Wilensky, 1978] Plan Applier Mechanism. Understands stories by determining the goals that are to be achieved in the story and attempting to match the actions of the story with the methods that it knows will achieve the goals.

Plans are the means by which goals are accomplished, and understanding plan-based stories involves discerning the goals of the actor and the methods by which the actor chooses to fulfill those goals. In a plan-based story, the understander must discern the goals of the main actor and the actions that accomplish those goals.

PAM tries to determine the main goal and the D-goals that will satisfy the goal. It analyzes input conceptualizations for their potential realization of one of the plan boxes that are called by one of the determined D-goals. PAM utilizes two kinds of knowledge structures in understanding goals: named plans and themes (such as LOVE, which contain background information upon which predictions can be based that individuals will have certain goals)..

The distinction between plan-based and script-based stories is simple: in a script-based story, parts or all of the story correspond to one or more scripts available to the story understander; in a plan-based story, the understander must discern the goals of the main actor and the actions that accomplish those goals.

LIFER

[Hendrix, 1977] Built at SRI, it is an off-the-shelf system for building "natural language front-ends" for applications in any domain. It has a set of interactive functions for specifying a language and parser.