I PL 86-36 50 USC 3605 I MPINIII tJvwwm oom uJlill GJ 1 1 wv till l lliH ls ID11 Ull ultilU lilUIQl UIW i r TO PULL A PONYAL •••••••• ••••••• __ 1_______ 1 MUSINGS ABOUT THE AG - 22 IATS •••• •••• Cecil Phillips •••• •••••• 3 • 11118 DOOVM•R N R JHfl NBIIWl B M1111U I ti • • 'FGPIIG R Approved for Release by NSA on 04 - 26-2021 pursuant to E O 13526 MOR Case 109413 I Non - Responsive IOP S'BCRfi Published Monthly by Pl Techniques and Standards for the Personnel of Operations VOL III No MARCH 1976 'FOP BROH8'f l I EO 3 3b 3 PL 86-36 50 USC 3605 · •i ' •• er OEORR• PL 86-36 50 USC 3605 lffllRRt• TO iUl l·-A ''PONYAL' · I · _·· t · ' ' A642 • It almost ne w er fail - • Ii the trtJJscriber •1 · uaderstands •v Fy word 111tat t1te Russi rf is saying the luuian re ea · th• statement sever- where It Is easY t6 suy till Chi li1 Slilg Ibid a times B t if the tra xdbf canMt i e has to be an adjective but how to prove it o t a word or t1to -- and tll1s• happens most f In English an adjective or noun used ad'Ulently when thl words are ac i ticai ones -- t i 'e jectivally does not Russian at t he pther end of1 t e lin says t a follows it ''Ponyalfn ''Uh-huh tht s a'tement is not r11peated a d the conversa tion'fitk es • completely differenf tatk Because of thiJ fact of Ufe we Ru1siah transcribch-s l uurseives ponyal pullers n ' we caJ l t e• i oriota s job o f listeniq and rel1stening to ht wikno • So as I was pulling this ponyal rd or wot'ds while si1111ltaneoulJy researching listened the appropriate adjectival endt ie - iltip e possibilities in' the p 8ilfble ing It wasn I there I listened elsewhere on '9arch aids p lling a ponyat 1 • • • the reel for the•i ord to be repeated no other t • occurrences Every tµie that a transcriblr piikl up a • Well then what own ite111 sound ape she Aal i-Ll l iMiil WAIU Wioloii Wllill _ I LIUIA 1 ke It sounded like I looked in clifferent I • various dictionaries fo ive that began I ·that way asswning that the ending had been spoken I • but'J las inaudible to me no results How about t tacldllg on the usual adjectival endings and then I I lookinl_up the possibilities in the reverse dieI ike I Again I I no resu I 1 Okay thep tear the word apar t l Well the I combination ak is highly improbable in spoken 1 ussian Beca Jse of the feature of consonant 1 I voicing in Russian combination of a a plus a k I would be pronouneed as el or ag · even though the • s# lling of the w61d would not indicate the W barw 1mwmcialt 1'ltnl ll lL rJ Jht then Change and check all the f-T1 -c 't -r o 'n a-r 1 e_s_a_g__a n-- '_ _ t a-t --s ouldn' t be 1110re than few thousand possibilities to check Add t e preswned adjectival'lndin and check the rffersttdictionaries for etc -· thirt's oply another S00f sb7k siibilities to che Jt But still I could find nothing to fit ng t rou t e secon time starte to the W U ll Ulli L Qen thoujlh I didn't •write down verb•tim what the speaker was say-• ing I found that I had to supply inaudible • know ---- ---T -r -- - -1 the speaker was refer prefixes or othll'r parts of words myself I did ring to I i u1 1 1111 lY new that he wasn't dis• that either by r lying on my previous encounters • cussing the __ __ or cut of a dress • Baclc to square one Maybe the missing word • with the words i si• ilar phrases using various • •dictionaries to fill in the gaps in my knowain't an ad ective I'd been thinking along • ledge or by finding the phrase repeated else where on the taJftl For example IIY experience told IP that thf phrase in this sepent was ce by a noun but still not ing came c ose o tile sounds that I heard No it had to be an But after re uilding as much of the verbatiw adjective Getting desperate I checked other transcripts utterance as I cfuld I was still left with one unknown stretch _ Was the un- for that date and case notation Maybe the same phrase had been used on a different reel in a known item a single word Two words The logic similar situation No such luck of the utterance called for an ad ective The 0 rt- ctif •- I I 1 March 76 CRYPTOLOG • Page l 'IQP ill Rli'f IJftlllDlt I I BO 3 3b 3 PL 86- 36 50 osc 3605 I Non - Responsive • T8P 8110DIIT Bl'IIBRJ -· Wait l minute Anpther brains enn In Russian an Z is often lli 'Sheard for 411 z and vice versa Not only tha • but an unsftessed a often represente an o in tH e Russian di ionary spelling as opposed to G s pronunciati Pn Unfort D ately • the quali t - c f my tape lllllde it virtually impo sible to kne where the i tress was in the unknown word so•n w I had a•p etty large number of• additiona1·•o binations to la with 0 to yield something •But hell f re and damnation No11hing What tb do We1j at this c 111pietely desperate pofnt anothcu repeatedly• ob erved feature of S GINT life•reared its h ad serendipity Have ou ever nd'ticed that jlftc 'r you come home from-spending ioj ars at work tryjng to locate a cenain word the word j 11111ps out at you from the 1'5 ny Tners «rapper on your- kitchen counter h happened' again this' time Just as I was ready to thro• 'in the spol Be l memo crossed my•deslt Ofle word jU111p ouf and hit my eye -- the Engliih abbreviati on CQB close bf busines For 50111e reason rt sunested to me that maibe the w inl I was 106kingofor was an abbreviatian itself ot the firs part of a telescoped wor consist nj ofw revi Sed adjective tacked' onto a ord I roto the reverse dictiOJlal'f too t e word as my juap-of point erut then tried pre xing it t with all th9-J11Wlil l·Ulll • iii1 1m1 Wu J 1Wi t- 1 _ IIAl1LJU __ I ready tried __ --- ----- - - --- -- 'I in all their My first impression had been col't'ect The missing item was indeM an adjective but it had been telesco ed for b•evit the unabbreviated term iA P 'l '' 'l l ' i l - which is much more o a 110u u auU11ption that the speaker was slurring the ending of the adjective had been wron1 The speaker wasn't using the ending of the word at all i nstead the adjective and noun had been fused into a single language unit This telescoping of words also called the for111at ion of portmanteau words is a very common occurrence in Russian and is not unc01111110n in other languages since all languages respond to the needs and requirements placed on them We transcribers of Russian or of other languages must remain constantly aware of the ways in which foreign langual es behave like our own language New words are being coi ned every day in order to meet the demands of technology and the need to collll IWlicate more and 1110re information in less time If it is happening in English it is also happening in all the other living languages of the world March 76 • CRYPTOLOG • Page 2 '118P 81911M' t Jll lBBst I -- --- - --- ---------- ----88HPIB11Hflftt MUSINGS ABOUT THE AG-22 IATS ilfJWWW1'ts w • • __ __ -1 1 • ' Cecil Phillips C03 Th8 foZ Zowing tU'tiote is reprinted 1' m C-LINERS C Group ltrchine Processing Information Buturtin Vol 3 No Aug- wst Sep 1' 0ctobe1' 197$ In November 1960 1S years ago the first tests of the AFSAV-D 311 a prototype of the AG-22 were conducted at Rothwesten Germany At the tiH of the tests those of us in the ADVA- GENS Joint Mechanization Group had great hopes that in 4 or S years the D-311 would have great impact on the nature of traffic analysis We thought that the full-text input and carefully designed editing and formatting prograas would eliminate lllUCh of the work of TECSUM preparation and punching of paper tapes at the site and the card pwlching and editing at NSA We realized that the results would not be as good as very carefully hand-prepared reports but our tests showed that the average error rate of our editing and formatting programs was about the same as that of manually prepared TECSUMs Several recent events have sharply confimed what I have suspected for some time that is the computer records generated totally automati• cally fro• AG-22 and IATS are ot very poor quality These events include detailed discussions with each of the offices in A B G and W about computer needs for the future and discussions with non-NSA elements receiving feedback from these auto-tic processes The opinion seems almost universal that the output is very poor if one expects specific infol'llation such as cipher text in a format suitable for cryptanalysis The same is true if one is looking for a unique degarbled set of callsigns for each network I am equally sure this applies to any kind of specific unique information I do not know exactly why the goal has not been realized but I suspect that the computer programs have not been as tightly tailored to the input as were our first experiments There may also be more variation in the data and perhaps more variation in the way that it is copied Another factor which may have been present is that our first experiments probably had the effect of stimulating the operators to copy with better than average care If I am right about the probable causes of the poor quality then there are a very limited set of potential solutions for improving the results As I see these alternatives we can • try to develop 110re precise and more sophisticated programs • introduce extra edit steps into the process using interactive computer terminals or • make changes in the way that data is copied I have serious doubts about the first of these alternatives since it would probably consume more good progr11111111er resources than are available The second alternative has real merit particularly with the expected expansion of the nUJaber of available interactive terminals However it may also suffer from a lack of available manpower to do the editing and correction I believe that the third alternative -changes in the way that the traffic is copied -- offers the best hop for the immediate future Thus it is my contention that we should take i11111ediate steps to modify coding procedures or at least to test sorae possible changes The kinds of changes I would suggest are outlined below In the non-message instructions I think that we should change the morse copying concept from the idea of copying everything to the idea of swnmarizing or something more comparable to gisting in voice communication The emphasis should be on getting one good representation of the callsigns and callup s rather than all kinds of garbled versions While it is theoretically possible to produce computer algorithms to degarble and SU111111arize as a practical 11 tter a human of IIOdest skill can almost certainly do better Equally the copying of chatter could be more sharply focused on unusual items by allowing the operator to give a co111111ent on the nature of routine chatter and to copy only the unusual chatter verbatim I suppose that the procedures mentioned above would require some better understanding by the operators of what is important but I can't help believing that they would be better motivated if they knew more about the targets and why they are being copied March 76 • CRYPTOLOG 0PWIBIPffl tJ ·- ' 1 Page 3 l lld•IW I Iii illllllllf lllldllll LII 91112 ·---·· I PL 86-36 50 USC 3605 l • As far as 11essagcs are concerned I think there are some 1110re• pecific things that might be done I think ti i operators could do a little 110re to format aessage To a considerable extent the op tor already formats messages What I am s1 esting is that we go as far as possible towaid' operator formatting I would also couple ack i'i ional tagging with the tighter formatt unfortunately tlfe AG-22 ICSR-37 does not permit of corrections s aich as is desirable but with TENNIS and MARQPN SHIELD there are excellent possibilities fiat c orrection of copy and for production of a Dffl Site version from repeated transmissions •• l11 any case a strong · concentration on mes ages might unearth other ways of producing be ter -copy even with the There are a coup e'Of argunients given KSR-l7 • • against more tagging add '4' re formatting by the operator One of these is that he is already overburdened Since•J kave never been an operator I cannot deal with•thits directly but it seems to me that by essenihg the total amount of copy and treating thl op9rator a little more intelligcmtly we can•ce tainly get a better product • • The other argumen giyen ls that the editing programs must assume•missed ugs and therefore it makes no differen e wkethe'r the operator tags or not There is soie truth n this asswaption but it misses the fact tlfat elen an operator doing a poor job can•put n same tags that the program can never put in •• In • sense program and operator are comp emem ary 'the program doing a better job than the opetatorj hen everything is routine but the opera or is infinitely better with any kind of•varh tion er unique situation TodotheAG-22 obwau om1needsboth and both should be as good as on can p ctically aet To SUID up I beli ve it•is tillle we took a drastic look at the vai HF erse i copied Fro• what I know of COPES a sdrt has been made toward what is i11p0rflllnt i the trafic on a specific case or target'basis • I think we can add some general goals add impruve thi product At the very least we ciyt redute the yolume of records generated from AG-21 input and produce more useful data bases • Conaents anyone Ill' Pm ltips was Chief of tM ADVAGENS Jr ri nt Msc haniaation Gztoup which conducted the fiNt ts is of an Alr22like devi c e in 1860 awl• 2861 at Roth• II westen and 1 Ger-many · Carrie B11ny and '6 re also membfll'B of the group l annatadt - 1um fJf- i c AFSAV-D 311 t8llte Mr PhiZl ip _ mtmtions IJel'e d out by Ill' L J 4 ii6-' 0ltilllifliiliiiNiiMltlllill ih lili Li -•1N ¥ t8il8ifll ' March 76 11------------------------ CRYPTOLOG Page £QNIIDIN 51 r L 4 111112•0 1111 U811i T umn1 1S ·- ---- - - ------------ --- mnx Responsive i