Documents
Classification Guide for Human Language Technology (HLT) Models
May 5, 2015
UN CLASSIF IEfo OR OFFICIAL USE ONLY
NATIONAL SECURITY AGENCY
CENTRAL SECURITY SERVICE
(U) Classificatian Guide far
Human Language Taahaalagy (HLT) Madala
2-20
E?'autive Date: 13 Ma}r E?ll
Illaplutz,r Direut?r fur Analysis
and Pmduuti?n
Categlry: 1.4
On: 35 yaar?
ENDURSED
IZ'Iaputg,r Aaa?aiate Direct?r far
and Recurda
UNCLASSIFIEDH FOR OFFICIAL USE ONLY
UN CLASSIF IEfo OR OFFICIAL USE ONLY
NATIONAL SECURITY AGENCY
CENTRAL SECURITY SERVICE
(U) Classificatian Guide far
Human Language Taahaalagy (HLT) Madala
2-20
E?'autive Date: 13 Ma}r E?ll
Illaplutz,r Direut?r fur Analysis
and Pmduuti?n
Categlry: 1.4
On: 35 yaar?
ENDURSED
IZ'Iaputg,r Aaa?aiate Direct?r far
and Recurda
UNCLASSIFIEDH FOR OFFICIAL USE ONLY
UN CLASSIF IEfo OR OFFICIAL USE ONLY
CLASSIFICATIDN GUIDE TITLEINUIHIEER:
(U) Human Languagn (PILT) Mndala, 2-2?
PUBLICATION DATE: 18 May ZUII
DFFICE DF DRIGIN: (U) Human Languagn
Pnc: 961-3?32a
GRIGWAL CLASSIFICATI nN AUTHORITY: (UHFDUD) Daputy
Dirantnr Analyaia and
Descrip than of Information
Classi?cation!
Marl-[Inga
Canaan
Remarks
A. (U) General
A. 1. Tl]: fact that has;
I-ILT nand fnr:
I
I Langnagn
I Langnagn
I pnaknr
I
I antivil}r
I Annmal},r
I
UNCLASSIFIED
A.2. fact that I-ILT am
nbtainnd. at lnaat in part. by
atatiatina dnrivnd ?nm
UNCLASSIFIED
A3. 'Ihn fact that I-ILT alan
andin film?; In Ian and
pdnritiznd linguists?;
UNCLASSIFIED
A.4. 'Ihn fact that statistics in a
can gnmratnd ?'nm n1" man},r audin
UNCLASSIFIED
A.S. 'Ihn fact that ?aw am
mgnlarlj,r adding [n aggmgatn
Datum nf'
UNCLASSIFIED
Ali. Tl]: fact that SIGINT vnicn
[nnl furthm? idnnti?nd] can
idnntif'ind an:
I maln nr fnmaln
I a apnci?n langnagn
I a apnci?n langnagn
I a apnci?c apnaknr
I a nf' wnrda
I n1"
UNCLASSIFIED
dntaila such as which
apnni?n langnagn. nr dialnnt. n1?
apnaknr am Cnnault
applinab 1n SICINT
I-ILT uand fnr:
I
I Langnagn
Sen anmka
nf'
I-ILT nand and
Langnagn ia
nf' nand [n train
up In TCI
UNCLASSIFIEDNFOR OFFICIAL USE ONLY
UN CLASSIF IEfo OR OFFICIAL USE ONLY
CLASSIFICATIDN GUIDE TITLEINUIHIEER:
(U) Human Languagn (PILT) Mndala, 2-2?
PUBLICATION DATE: 18 May ZUII
DFFICE DF DRIGIN: (U) Human Languagn
Pnc: 961-3?32a
GRIGWAL CLASSIFICATI nN AUTHORITY: (UHFDUD) Daputy
Dirantnr Analyaia and
Descrip than of Information
Classi?cation!
Marl-[Inga
Canaan
Remarks
A. (U) General
A. 1. Tl]: fact that has;
I-ILT nand fnr:
I
I Langnagn
I Langnagn
I pnaknr
I
I antivil}r
I Annmal},r
I
UNCLASSIFIED
A.2. fact that I-ILT am
nbtainnd. at lnaat in part. by
atatiatina dnrivnd ?nm
UNCLASSIFIED
A3. 'Ihn fact that I-ILT alan
andin film?; In Ian and
pdnritiznd linguists?;
UNCLASSIFIED
A.4. 'Ihn fact that statistics in a
can gnmratnd ?'nm n1" man},r audin
UNCLASSIFIED
A.S. 'Ihn fact that ?aw am
mgnlarlj,r adding [n aggmgatn
Datum nf'
UNCLASSIFIED
Ali. Tl]: fact that SIGINT vnicn
[nnl furthm? idnnti?nd] can
idnntif'ind an:
I maln nr fnmaln
I a apnci?n langnagn
I a apnci?n langnagn
I a apnci?c apnaknr
I a nf' wnrda
I n1"
UNCLASSIFIED
dntaila such as which
apnni?n langnagn. nr dialnnt. n1?
apnaknr am Cnnault
applinab 1n SICINT
I-ILT uand fnr:
I
I Langnagn
Sen anmka
nf'
I-ILT nand and
Langnagn ia
nf' nand [n train
up In TCI
UNCLASSIFIEDNFOR OFFICIAL USE ONLY
UNCLASSIF IEfo OR OFFICIAL USE ONLY
USA, AUS, CAN, GEE, NZL.
Although it it?; poaaihlc that
nacd to train moch
may hayc a highcr
anch'or morc rcatrictiyc
than SECRETNREL,
original audio cannot hc
from modcl.
i5; atrt't'icicnt to
protcct thia of Inodcl.
Dircctor for
Analysis; and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A.3. HLT apcalccr rccognition modch;
Ecl'l't?l?k?.
Conanlt applicahlc SIGINT
gnidancc: Classification and
forcign ahonlcl hc in
accordancc 1with highcat
classification and moat lcatt'ictiyc
that applica to
cntitica nacd in moch
Dircctor for
Analysis and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A3. (U) HLT aconatic mochs; naccl for
I
I F?honctic tokcnization
of
HLT aconatic moclcla i5;
npon classi?cation of
nacd to train Inodcl,
up to SECRETHREL TCI USA,
AUS, CAN, GER, NZL. Although
it is; poaaihlc that
nach to train moch may hayc a
hi classification antifor morc
rcatrictiyc ty than
original anclio
cannot hc from tho
moch. SECRETNREL i5
ant'?cicnt to plotcct this?; of
moch
Dircctor for
Analysis and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A. if}. (U) HLT langnagc moclcla nacd for
I
Conanlt applicahlc SIGINT
gnidancc: Classification and
forcign ahonlcl hc in
UNCLASSIFIEDH FOR OFFICIAL USE ONLY
UNCLASSIF IEfo OR OFFICIAL USE ONLY
USA, AUS, CAN, GEE, NZL.
Although it it?; poaaihlc that
nacd to train moch
may hayc a highcr
anch'or morc rcatrictiyc
than SECRETNREL,
original audio cannot hc
from modcl.
i5; atrt't'icicnt to
protcct thia of Inodcl.
Dircctor for
Analysis; and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A.3. HLT apcalccr rccognition modch;
Ecl'l't?l?k?.
Conanlt applicahlc SIGINT
gnidancc: Classification and
forcign ahonlcl hc in
accordancc 1with highcat
classification and moat lcatt'ictiyc
that applica to
cntitica nacd in moch
Dircctor for
Analysis and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A3. (U) HLT aconatic mochs; naccl for
I
I F?honctic tokcnization
of
HLT aconatic moclcla i5;
npon classi?cation of
nacd to train Inodcl,
up to SECRETHREL TCI USA,
AUS, CAN, GER, NZL. Although
it is; poaaihlc that
nach to train moch may hayc a
hi classification antifor morc
rcatrictiyc ty than
original anclio
cannot hc from tho
moch. SECRETNREL i5
ant'?cicnt to plotcct this?; of
moch
Dircctor for
Analysis and Production may
approyc, on a caac-hy-caac haaia,
forcign rclcaac of moclcla
containing non-
information.
A. if}. (U) HLT langnagc moclcla nacd for
I
Conanlt applicahlc SIGINT
gnidancc: Classification and
forcign ahonlcl hc in
UNCLASSIFIEDH FOR OFFICIAL USE ONLY
UNCLASSIF IEfo OR OFFICIAL USE ONLY
I Fhanatia takanizatian
with tha highaat
and must laatriativa
that appliaa ta tha
ttn'gatad antitiaa andfar aantant af'
tha uaad in tha Inadal.
(U) Tha Ellaput}r Diraatar far
Analysis; and Fraduatian Ina},r
apprava. an a haaia.
faraign af' Inadala
aantaining atharwiaa nan-
inf'annatian.
A.11. (U) aatiwitj,r dataatian mandala NIA
using ayllahla 1'ata aati'trit}r dataatian FCIE CIFFICIAL
(SRSAEI) USE CINLT
A. 1 2. (U) Antanalj,r dataatian mandala NIA
FCIR CIFFICIAL
USE CINLT
B. (U) Mada]
?utput
E.1 (U) Clutput af' languaga (U) Raaulta ganarall},r indiaata tha
tnndala FDR CIFFICIAL languaga and tha
USE CINLT dagraa af' aanfidanaa in tha
datal'tninatian. a. g. ?Farsi with
913% aanf'idanaa.? This;
infartnatian tna},r laquil?a pt?ataatian
aa whan aatnhinad with
athar dataila ragtn?ding tha input
data.
13.2. (U) Clutput af' gandar NIA (U) Raaulta ganarall}r indiaata tha
lnudala FCIE DFFICIAL gandar and tha dagt'aa
USE CINLT af' aanf'idanaa in tha datartninatian.
a. g. ?Mala with T591: aanf'idanaa.?
This?; infartnatian Ina},r laquil?a
pt?ataalian aa whan
aalnhinad with athar dataila
ragtn'ding tha input data.
13.3. (U) Clutput af' apaakar
tnndala
Saa Eatntn?ka.
(U) and fal?aign
of tha 1?aau1ta ahauld
ha tha aalna as; tha input data.
EA. (U) Clutput af'
and phunatia talcanizatian inadala
Saa Eatntn?ka.
(U) and fal?aign
of tha 1?aau1ta ahauld
ha tha aatna aa tha input data.
E.S. (U) Clutput af' languaga
and phanatia talianizatian Inadala
Saa Eatntn?ka.
(U) and fal?aign
of tha raaulta ahauld
ha tha aatna aa tha input data
tha 1?aau1ta ravaal apaai?a
infartnatian uaad in tha Inadal that
is?; prataatad at a highar laval than
tha input data; in this?; tha
raaulta raquira pl'ataatian at tha
laval af' tha tnndal.
(U) Nata: in 25 yatn?a indiaataa that tha inf'artnalian ia far 25 ??atn tha data a daautnant i5;
at?aatad {11? 25 yatn?a ??atn tha data at this ariginal whiahavar ia 1ata1?.
UNCLASSIFIEDNFOR OFFICIAL USE ONLY
UNCLASSIF IEfo OR OFFICIAL USE ONLY
I Fhanatia takanizatian
with tha highaat
and must laatriativa
that appliaa ta tha
ttn'gatad antitiaa andfar aantant af'
tha uaad in tha Inadal.
(U) Tha Ellaput}r Diraatar far
Analysis; and Fraduatian Ina},r
apprava. an a haaia.
faraign af' Inadala
aantaining atharwiaa nan-
inf'annatian.
A.11. (U) aatiwitj,r dataatian mandala NIA
using ayllahla 1'ata aati'trit}r dataatian FCIE CIFFICIAL
(SRSAEI) USE CINLT
A. 1 2. (U) Antanalj,r dataatian mandala NIA
FCIR CIFFICIAL
USE CINLT
B. (U) Mada]
?utput
E.1 (U) Clutput af' languaga (U) Raaulta ganarall},r indiaata tha
tnndala FDR CIFFICIAL languaga and tha
USE CINLT dagraa af' aanfidanaa in tha
datal'tninatian. a. g. ?Farsi with
913% aanf'idanaa.? This;
infartnatian tna},r laquil?a pt?ataatian
aa whan aatnhinad with
athar dataila ragtn?ding tha input
data.
13.2. (U) Clutput af' gandar NIA (U) Raaulta ganarall}r indiaata tha
lnudala FCIE DFFICIAL gandar and tha dagt'aa
USE CINLT af' aanf'idanaa in tha datartninatian.
a. g. ?Mala with T591: aanf'idanaa.?
This?; infartnatian Ina},r laquil?a
pt?ataalian aa whan
aalnhinad with athar dataila
ragtn'ding tha input data.
13.3. (U) Clutput af' apaakar
tnndala
Saa Eatntn?ka.
(U) and fal?aign
of tha 1?aau1ta ahauld
ha tha aalna as; tha input data.
EA. (U) Clutput af'
and phunatia talcanizatian inadala
Saa Eatntn?ka.
(U) and fal?aign
of tha 1?aau1ta ahauld
ha tha aatna aa tha input data.
E.S. (U) Clutput af' languaga
and phanatia talianizatian Inadala
Saa Eatntn?ka.
(U) and fal?aign
of tha raaulta ahauld
ha tha aatna aa tha input data
tha 1?aau1ta ravaal apaai?a
infartnatian uaad in tha Inadal that
is?; prataatad at a highar laval than
tha input data; in this?; tha
raaulta raquira pl'ataatian at tha
laval af' tha tnndal.
(U) Nata: in 25 yatn?a indiaataa that tha inf'artnalian ia far 25 ??atn tha data a daautnant i5;
at?aatad {11? 25 yatn?a ??atn tha data at this ariginal whiahavar ia 1ata1?.
UNCLASSIFIEDNFOR OFFICIAL USE ONLY