Network Working Group H. Murray
Request for Comments: 420 Computer Corporation of America
NIC 12764 04 January 1973
CCA ICCC WEATHER DEMO
THE WEATHER DEMO ORIGINALLY DEVELOPED FOR THE ICCC SHOW IS NOW
GENERALLY AVAILABLE. IT IS A VERY NICE NETWORK DEMO SINCE IT REALLY
CAN/DOES RUN ON TWO MACHINES.
THIS DEMO IS TWO INDEPENDENT PROGRAMS. TOGETHER THEY TRY TO ANSWER
ENGLISH QUESTIONS ABOUT THE WEATHER. FOR EXAMPLE, 'DID IT RAIN IN
BOSTON ON OCTOBER 7?'. THE NATURAL LANGUAGE PROGRAM IS WRITTEN IN
LISP. IT IS A MODIFICATION OF A SYSTEM DEVELOPED BY T. WINOGRAD AT
MIT. THE OTHER HALF IS A PRIMITIVE DATACOMPUTER. THE DATACOMPUTER
PROVIDES NETWORK DATAMANAGEMENT FACILITIES AND WILL INCORPORATE THE
TRILLION-BIT UNICON CURRENTLY BEING INSTALLED AT AMES.
THE TWO PROGRAMS COMMUNICATE VIA DATALANGUAGE USING THE NET.
'WEATHER' NOW USES HOST '37' AS THE LOCATION OF THE DATACOMPUTER.
THIS FIXES THE OLD 'CCA' NOT IN THE HOST TABLES BUG.
KEEP IN MIND THAT IS AN EVOLVING SYSTEM, WE PLAN TO IMPROVE 'WEATHER'
AND THE DATACOMPUTER THAT IT USES, SO THINGS THAT WORK ONE DAY MAY
GET BROKEN THE NEXT.
AVAILABILITY
'WEATHER' (THE LISP HALF) IS A REAL MACHINE USER - BOTH CPU AND CORE.
WE WOULD OBVIOUSLY PREFER THAT YOU RUN IT ON YOUR MACHINE. IF WE ARE
REALLY EMPTY (WHICH FREQUENTLY HAPPENS AT 3 AM) FEEL FREE TO RUN IT
ON OUR MACHINE. SINCE OUR SYSTEM HAS NO DRUM, AND WEATHER TAKES ALL
OF CORE, PLEASE CHECK WITH SOMEBODY AT CCA BEFORE RUNNING WEATHER IF
WE ARE DOING ANYTHING AT ALL. IT IS REALLY A MAJOR LOAD ON OUR
MACHINE.
SINCE THE DATACOMPUTER IS LESS OF A LOAD, THE SYSTEM WILL BE MORE
AVAILABLE IF YOU CAN RUN 'WEATHER' ON YOUR MACHINE. WE WILL TRY TO
RUN THE DATACOMPUTER DURING NON-PRIME TIME. IT CAN BE THOUGHT OF AS
THE SERVER HALF OF THIS COMBINATION. PRIME TIME HERE IS 1 TO 5 PM
EASTERN TIME, BUT THE MORNING IS FREQUENTLY QUITE BUSY TOO. PLEASE
DO NOT USE THE DATACOMPUTER DURING THIS TIME. IT WILL SAVE US THE
EFFORT OF TURNING IT OFF. CURRENTLY, ONLY ONE USER AT A TIME CAN RUN
- A LIMITATION OF THE DATACOMPUTER THAT WILL GO AWAY SOMEDAY.
Murray [Page 1]
RFC 420 CCA ICCC Weather Demo January 1973
IF YOU WANT TO RUN A DEMO (NOT JUST PLAY AROUND) WE WILL PROBABLY BE
ABLE TO COOPERATE. PLEASE CALL TO MAKE ARRANGEMENTS WELL IN ADVANCE.
WE STRONGLY RECOMMEND THAT YOU DO NOT TRY TO GIVE AN IMPORTANT DEMO
UNLESS YOU HAVE USED BOTH VERSIONS OF THE SYSTEMS THAT WILL BE USED
FOR THE DEMO. EVEN WE GET IN TROUBLE WHEN WE FORGET TO REHEARSE
THINGS A BIT.
WEDNESDAY MORNING IS SCHEDULED PREVENTIVE MAINTENANCE TIME. OUR
MACHINE WILL NOT BE UP BETWWEN 9 AND 12, BUT IT FREQUENTLY IS DOWN
PAST NOON.
MORE INFORMATION
FOR THE LATEST INFORMATION, LOOK AT *.TXT IN <WDB-DEMO> AT CCA. IN
PARTICULAR, MESSAGE.TXT SHOULD BE A QUICK SUMMARY OF ANY OTHER
INTERESTING TXT FILES. THE LATEST VERSION OF THIS FILE (IN RUNOFF
FORMAT) IS <HGM-MEMOS>ANNOUNCE.RNO.
THE LATEST VERSION OF THE LISP PROGRAM WILL BE IN <WDB-DEMO> AS
WEATHER.SAV. NORMAL FTP CAN BE USED TO COPY IT - IT IS 400+ PAGES.
IT TAKES 6 MIN TO TRANSFER IT TO OURSELVES (1.5 MIN OF CPU, 22KBAUD)
WHEN THE MACHINE IS VERY LIGHLY LOADED. BE SURE TO SET THE BYTE SIZE
TO 36 ('TENEX' COMMAND TO FTP) SINCE THIS IS A SAVE FILE.
PROBLEMS
IF YOU FIND ANY PROBLEMS WITH LISP HALF, PLEASE USE SNDMSG TO SEND
THINGS TO JEFF HILL - HE IS JMH@CCA. WE ARE INTERESTED IN EXAMPLES
WHICH LOOK LIKE THEY SHOULD WORK BUT DON'T. IF YOU HAVE ANY COMMENTS
ABOUT GENERAL OPERATIONS GLITCHES - LIKE THE DATACOMPUTER IS NOT UP -
BITCH TO ME - HGM@CCA. CCA IS HOST 31 DECIMAL, 37 OCTAL.
USING THE SYSTEM
AFTER STARTING WEATHER, THE SYSTEM WILL INITIALIZE ITSELF, TYPE OUT
SOME MESSAGES, AND ASK FOR THE JULIAN DATE (REALLY THE DAY NUMBER OF
THE YEAR OF 'TODAY'). 300 IS THE BEST ANSWER. THAT MAKES 'TODAY' BE
OCT 26, THE LAST DAY OF THE ICCC DEMO, AND ONE DAY PAST THE LATEST
DATA. IF YOU TYPE SOME OTHER NUMBER, 'YESTERDAY', AND 'LAST WEEK'
WON'T WORK AS WELL. THE SYSTEM PROMPTS WITH 'READY' WHEN IT IS READY
TO ACCEPT A SENTENCE. BE SURE TO END THE SENTENCE WITH THE PROPER
PUNCTUATION. SENTENCES MAY BE EDITED WITH:
^A (CONTROL A) -- DELETES LAST CHARACTER OR WORD TYPED
^R (CONTROL R) -- RETYPES SENTENCE SO FAR
^Q (CONTROL Q) -- DELETES ENTIRE SENTENCE
Murray [Page 2]
RFC 420 CCA ICCC Weather Demo January 1973
^X (CONTROL X) -- SAME AS CONTROL Q
IN ADDITION, TWO CONTROL CHARACTERS CAN BE USED DURING THE PROCESSING
OF SENTENCES, ^E (CONTROL E) ABORTS THE PROCESSING OF THE SENTENCE
AND RETURNS TO THE READY STATE. ^D (CONTROL D) ABORTS THE WHOLE
SYSTEM, AND CAUSES IT TO RE-INITIALIZE ITSELF.
DURING SENTENCE INPUT, THE SYSTEM ATTEMPTS TO LOOK UP EACH WORD AS IT
IS TYPED, AND DO SPELLING CORRECTION. HOWEVER, SINCE THE NAMES OF
ALL THE CITIES ARE KEPT AT THE DATACOMPUTER, THE SYSTEM ACCEPTS ALL
UNKNOWN WORDS AS POSSIBLE PLACE NAMES TO BE CHECKED LATER ON IN THE
PROCESSING. IF THE SPELLING CORRECTOR CORRECTS ONE OF YOUR CITY
NAMES TO A VOCABULARY WORD, YOU CAN ERASE THE WORD AND RETYPE IT
PRECEEDED BY AN EQUAL SIGN (=). THIS FORCES THE SYSTEM TO ACCEPT THE
WORD AS A PROPER NOUN.
THE SYSTEM RETURNS TO THE READY STATE AFTER ANSWERING THE QUESTION.
TO TERMINATE A SESSION YOU MAY TYPE 'GOODBYE.', WHICH WILL LEAVE YOU
BACK AT THE TENEX COMMAND LEVEL.
IF THE SYSTEM CANNOT UNDERSTAND YOUR REQUEST, IT TRIES TO GIVE YOU
SOME EXPLANATORY MESSAGE. HOWEVER, IT'S NOT VERY GOOD AT EXPLAINING
ITS BEHAVIOR SO YOU WILL USUALLY GET THE DEFAULT MESSAGE 'SORRY, I
COULDN'T UNDERSTAND THAT REQUEST.' IF YOU REALLY WANT TO KNOW WHY,
SEND US A TRANSCRIPTIPT OF YOUR SESSION.
RESTRICTIONS - HINTS
THE SYSTEM IS NOT LINGUISTICALLY SOPHISTICATED. IT PERFORMS PASSABLY
ON SENTENCES WHICH WOULD BE CONSIDERED SIMPLE REQUESTS FOR
INFORMATION. IT DOES NOT REALLY UNDERSTAND THE 'WEATHER', RATHER IT
UNDERSTANDS HOW TO COMPOSE RETRIEVAL REQUESTS FOR WEATHER DATA IN ITS
DATA BASE.
A LARGE SET OF EXAMPLE SENTENCES CAN BE FOUND IN <WDB-
DEMO>SENTENCES.TXT. THOSE SENTENCES PRECEDE BY A BACKSLASH (\) DO
NOT WORK. AN EXPLANATORY COMMENT, PRECEEDED BY A SEMI-COLON (;),
EXPLAINS THE REASONS FOR FAILURE. SOME OF THE OTHERS MAY NOT WORK
EITHER.
THE MAJOR THINGS THAT WE KNOW DON'T WORK ARE:
1. CONJUNCTION
2. NEGATION
3. COMPLEX QUANTIFICATION
Murray [Page 3]
RFC 420 CCA ICCC Weather Demo January 1973
4. SOPHISTICATED PRONOUN REFERENCE
5. QUESTIONS NOT (!!) INVOLVING WEATHER,E.G.,
'WHAT DAY WAS YESTERDAY?'
'WHAT CITIES ARE IN MASSACHUSETTS?'
'IT' IS NOT WORKING TOO WELL THESE DAYS. IF YOU GET IN TROUBLE, TRY
SOMETHING WITHOU 'IT' THAT IS KNOWN TO WORK. 'WHAT WAS THE
TEMPERATURE IN BOSTON YESTERDAY?' IS BETTER THAN MOST - WE ARE QUITE
APT TO TEST THAT PHRASE.
THE SYSTEM DOES NOT KNOW ABOUT THE WORD 'SNOW'. THERE IS NO WAY TO
LIST ITS VOCABULARY. HOWEVER, THE SENTENCES IN SENTENCES.TXT SHOULD
GIVE YOU A GOOD IDEA OF THE WORDS THE SYSTEM UNDESTANDS.
THE AVERAGE TIME FOR UNDERSTANDING AND ANSWERING A QUESTION IS ABOUT
1 MINUTE. IF THE SYSTEM TAKES MUCH LONGER THAN THIS YOU SHOULD GET
SUSPICIOUS. IF YOU ASK A VERY BIG QUESTION, LIKE 'WHAT WAS THE
HOTTEST PLACE IN CANADA?', THE LISP SYSTEM MAY GO INTO A GC (GARBAGE
COLLENTION) LOOP. RESTARTING ALL OVER (WITH ^D) WILL BE OK, BUT THE
DATACOMPUTER MAY TAKE A LONG TIME TO RECOVER - IT DOESN'T (YET)
NOTICE THAT ANYTHING HAS GONE WRONG UNTIL IT ASKS FOR MORE INPUT.
DATA
THE DATA FROM OCT 1 THRU 16 IS ALMOST COMPLETE. BETWEEN OCT 17 TO 25
IT IS FREQUENTLY SKETCHY. WE PLAN TO RELOAD EVERYTHING SOMETIME
USING SEPTEMBER'S DATA, SO WE WILL HAVE A COMPLETE MONTH OF CLEAN
DATA. NOTE: THIS DATA HAS BEEN EXTENSIVELY SUMMARIZED - IT IS
PROBABLY NOT USEFUL FOR PEOPLE INTERESTED IN WEATHER.
THE OCTOBER DATA IS ABOUT 2400 PAGES OF REAL DATA, PLUS ABOUT 200
MORE FOR THE INVERSION. THERE ARE ABOUT 4000 CITIES. THE CITY NAMES
HAVE BEEN CHANGED TO THE LOCAL CITY WHENEVER POSSIBLE. I.E. 'BOSTON'
IS 'BOSTON' NOT 'LOGAN'.
'ALASKA' GOT LOST SOMEHOW. DON'T BOTHER ASKING FOR IT.
SAMPLE DATALANGUAGE
THESE EXAMPLES ON THIS AND THE FOLLOWING PAGES WORK ON A VERY OLD
VERSION OF THE DATACOMPUTER. THEY ARE NOT THE FINAL FORM OF
DATALANGUAGE THAT SHOULD BA AVAILABLE SOON.
Murray [Page 4]
RFC 420 CCA ICCC Weather Demo January 1973
THE FOLLOWING QUESTION TO 'WEATHER' GENERATED TWO REQUESTS TO THE
DATACOMPUTER. THE FIRST USES THE CHECK PORT TO VERIFY THAT 'BOSTON'
IS A VALID CITY NAME, AND THE SECOND USES THE ANSWER PORT TO RETRIEVE
THE DATA. THE ACTUAL REQUESTS THAT ARE SENT TO THE DATACOMPUTER DO
NOT HAVE THE CARRIAGE RETURNS THAT MAKE THEM READABLE - THEY ARE SENT
AS ONE VERY LONG LINE. THE '^Z' IS THE END OF DATA MARKER.
TO WEATHER
HAS IT RAINED IN BOSTON RECENTLY?
TO THE DATACOMPUTER:
OR CHECK.PLACE,STATION WITH CITY EQ 'BOSTON'
PLACE.LP1 = '('
PLACE.CITY = STATION,CITY
PLACE.RP1 = ')'
PLACE.BSN = STATION.BSN
PLACE.LP2 = '('
PLACE.REGION = STATION.REGION
PLACE.RP2 = ')'
END
FROM THE DATACOMPUTER:
(BOSTON )725090(MASSACHUSETTS )
^Z
TO THE DATACOMPUTER:
FOR THE STATION WITH (REGION EQ 'MASSACHUSETTS' AND CITY EQ 'BOSTON')
FOR ANSWER.ANS OBSERVATION WITH GE '273' AND DATE LE '300'
ANS.LP1 = '('
ANS.CITY = STATION.CITY
ANS.RP1 =')'
ANS.DATE = OBSERVATION.DATE
ANS.LP2 = ')'
ANS.DATA = PRECIP
ANS.DATA1 = ' '
ANS.DATA2 = ' '
ANS.RP2 = ')'
END
END
THE FOLLOWING IS THE DESCRIPTION OF THE FILE AS IT IS STORED IN THE
DATACOMPUTER.
CREATE WEATHER FILE LIST
STATION STRUCT
BSN STR(6), I=D
CITY STR(22), I=D
REGION STR(22), I=D
WORLD STR(22)
OBS LIST (31)
OBSERVATION STRUCT
DATE STR(3)
TEMPERATURE STRUCT
MIN STR(4)
MAX STR(4)
END TEMPERATURE STRUCT
PRECIP STR(4)
WINDS STRUCT
SPEED STR(4)
GUSTS STR(4)
DIRECTION STR(4)
END WINDS STRUCT
VISIBILITY STR(4)
CLOUDS STR(4)
GENERAL STR(4)
PRESSURE
END OBSERVATION STRUCT
END STATION STRUCT
END WEATHER FILE
Murray [Page 6]
RFC 420 CCA ICCC Weather Demo January 1973
THE FOLLOWING PORT DESCRIPTIONS ARE USED BY WEATHER TO RETRIVE DATA
FROM THE DATACOMPUTER.
CREATE CHECK PORT LIST
PLACE STRUCT
LP1 STR (1)
CITY STR (22)
RP1 STR (1)
BSN STR (6)
LP2 STR (1)
REGION STR (22)
RP2 STR (1)
END PLACE STRUCT
END CREATE CHECK
CREATE ANSWER PORT LIST
ANS STRUCT
LP1 STR (1)
CITY STR (22)
RP1 STR (1)
DATE STR (3)
LP2 STR (1)
DATA STR (4)
DATA1 STR (4)
DATA2 STR (4)
RP2 STR (1)
END ANS STRUCT
END CREATE ANSWER
Murray [Page 7]
RFC 420 CCA ICCC Weather Demo January 1973
SAMPLE RECORD
THE FOLLOWING DATA IS SAMPLE RECORD - THE FIRST ONE IN THE FILE. IN
THE DATACOMPUTER IT IS STORED WITHOUT THE CARRIAGE RETURNS NEEDED TO
MAKE IT PRINT NEATLY.
