Documents
How Is Human Language (HLT) Progressing?
May 5, 2015
(SiiS Haw Is Human Language Technalagy (HLT) Pragressing?
FROM:
Language Analysis Mademisatian Lead (S2)
Run Date:
Editar?s intra: At the SID tawn haii meeting cf February 2011,
(pictured) briefed an Human Language Technaiagy, taais that sart thraugh SICINT vaice
caiiectian and autantaticaiiy find the mast pramising nuggets, thereby saving iinguists cauntiess
haurs. What?s happened with HLT since that time?
In 2D11 we deplayed HLT Labs ta Afghanistan, NSA Ceargia, Latin American SCS
sites, and SA Texas.
(U) Afghanistan-area targets
(SifS Afghan Regianal Operating Center (ARUCC) started using HLT
Labs ta track their targets in April, and when the analytics were successfully used ta find new
infarmatian, the missian was expanded ta include internatianal teams.* The Afghanistan
deplayment baasts same technalagical firsts assaciated with claud camputing** and includes the
full suite af analytics with Pashta speech-ta-text (STT). Recently French in the ARC were
able ta find target speakers an new selectars using speaker recagnitian.
Clur deplayment ta NSA Ceargia enables us ta partner with ta assess the
perfarmance af aur newest STT madels: Pashtc- and Farsi. These languages have limited training
data which creates challenges far STT, and we have been facused an finding applicatians that are
beneficial even far these law-resaurce languages. NSA-Ceargia traffic includes naisy VHF
callectians which seriausly degrade analytic perfarmance; hawever, can still find target
speaker cuts an unknawn frequencies.
(U) Spanish-speaking targets
(SifS Spanish is the mast mature af aur sp eech-ta-text analytics, and has higher keyward-
search accuracy than ather deplayed STT madels. We?ve had great success searching far Spanish
keywards at NSA Texas and Latin America SCS sites.
Far example, in early August a new NSA Texas user applied keyward search the
marning after his training ta find a previausly unrep arted cut fram a drug trafficking target.
Likewise, the DIC af ane cf the Latin American SCS sites recently reparted he was able ta find
fareign intelligence regarding a Cuban afficial in a fractian cf the usual time. His camment: This
same example cauld be used aver and aver by many that have ta ga aver cauntless vaice cuts tc-
finally dig that gald nugget that will turn inta a rep art.
(U) Develapment wark cantinues
(UHFCIUD) The RS research team is warking ta add new applicatians, imprave keynvard search
cap ability, enhance analytics, add new languages, and refine the user interface. Recently the
Summer Camp far Applied Language Explaratian (SCALE) -- a jaint NSAJahns
University exercise -- investigated new ways ta use the results af HLT analytics fram existing
targets ta find new targets. Research is alsa warking clasely with the SP (vaice analytics)
(SiiS Haw Is Human Language Technalagy (HLT) Pragressing?
FROM:
Language Analysis Mademisatian Lead (S2)
Run Date:
Editar?s intra: At the SID tawn haii meeting cf February 2011,
(pictured) briefed an Human Language Technaiagy, taais that sart thraugh SICINT vaice
caiiectian and autantaticaiiy find the mast pramising nuggets, thereby saving iinguists cauntiess
haurs. What?s happened with HLT since that time?
In 2D11 we deplayed HLT Labs ta Afghanistan, NSA Ceargia, Latin American SCS
sites, and SA Texas.
(U) Afghanistan-area targets
(SifS Afghan Regianal Operating Center (ARUCC) started using HLT
Labs ta track their targets in April, and when the analytics were successfully used ta find new
infarmatian, the missian was expanded ta include internatianal teams.* The Afghanistan
deplayment baasts same technalagical firsts assaciated with claud camputing** and includes the
full suite af analytics with Pashta speech-ta-text (STT). Recently French in the ARC were
able ta find target speakers an new selectars using speaker recagnitian.
Clur deplayment ta NSA Ceargia enables us ta partner with ta assess the
perfarmance af aur newest STT madels: Pashtc- and Farsi. These languages have limited training
data which creates challenges far STT, and we have been facused an finding applicatians that are
beneficial even far these law-resaurce languages. NSA-Ceargia traffic includes naisy VHF
callectians which seriausly degrade analytic perfarmance; hawever, can still find target
speaker cuts an unknawn frequencies.
(U) Spanish-speaking targets
(SifS Spanish is the mast mature af aur sp eech-ta-text analytics, and has higher keyward-
search accuracy than ather deplayed STT madels. We?ve had great success searching far Spanish
keywards at NSA Texas and Latin America SCS sites.
Far example, in early August a new NSA Texas user applied keyward search the
marning after his training ta find a previausly unrep arted cut fram a drug trafficking target.
Likewise, the DIC af ane cf the Latin American SCS sites recently reparted he was able ta find
fareign intelligence regarding a Cuban afficial in a fractian cf the usual time. His camment: This
same example cauld be used aver and aver by many that have ta ga aver cauntless vaice cuts tc-
finally dig that gald nugget that will turn inta a rep art.
(U) Develapment wark cantinues
(UHFCIUD) The RS research team is warking ta add new applicatians, imprave keynvard search
cap ability, enhance analytics, add new languages, and refine the user interface. Recently the
Summer Camp far Applied Language Explaratian (SCALE) -- a jaint NSAJahns
University exercise -- investigated new ways ta use the results af HLT analytics fram existing
targets ta find new targets. Research is alsa warking clasely with the SP (vaice analytics)
and TransX [translatibn, transcription and transliteratian tn ensure HLT Labs capabilities
are included in the carp crate splutipn for enterprise in 21312.
Mere abbut HLT Labs is available here.
See a related article about HLT here.
(U) Nntes:
The internatibnal teams were hunt the Analysis and Research Cell (ARC), Task Farce
31B, and Cuntbined pint Special Uperatiuns Task Fbrce
Specifically, the Afghan deployment is the first use bf DISTILLERY and
CLDUDBASE en a
and TransX [translatibn, transcription and transliteratian tn ensure HLT Labs capabilities
are included in the carp crate splutipn for enterprise in 21312.
Mere abbut HLT Labs is available here.
See a related article about HLT here.
(U) Nntes:
The internatibnal teams were hunt the Analysis and Research Cell (ARC), Task Farce
31B, and Cuntbined pint Special Uperatiuns Task Fbrce
Specifically, the Afghan deployment is the first use bf DISTILLERY and
CLDUDBASE en a