Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: The training of scent detection dogs using samples of explosives or their chemical precursors
Accepted 22 November 2013 is a well-established and documented practice. However an area of canine odor detection
Available online 1 December 2013
that remains under-studied regards a trained dog’s perception of an explosive odor when
more than one odorant is combined to produce a mixture. The first objective of our study
Keywords:
was to determine whether training adult Labrador Retrievers (n = 20) to detect the scent
Dog
of chemically pure potassium chlorate (PC) was sufficient to produce generalization to PC-
Olfactory
Explosives based explosive mixtures that contained a novel component. We found that the majority
Odor mixtures of dogs (87%) trained with pure PC alone did not correctly signal the presence of one or
more of four PC-based explosive mixtures. Our second objective was to determine whether
training dogs using the separated components found in the PC-based explosives would
subsequently enhance detection. Dogs were then trained using a novel static odor delivery
device that safely segregated the PC and non-PC components and presented a merged odor
to the dog. A statistically significant improvement in percentage of dogs detecting PC-based
mixtures after training with the separated components compared to training with PC alone
was seen with Mixture 1 (27–100%, P < 0.0001), Mixture 2 (40–81%, P = 0.0229), Mixture 3
(38–94%, P = 0.0004), and Mixture 4 (69–100%, P < 0.005). The results of this study highlight
the potential limitations of dogs trained to detect a single odor to then recognize the odor
when mixed with other substances. The odor delivery device developed for this study rep-
resents an important and effective training option that may reduce the need for using a
final PC explosive mixture in canine training.
© 2013 Elsevier B.V. All rights reserved.
0168-1591/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.applanim.2013.11.010
L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93 85
remain about the underlying processes of explosive scent number of factors including the particular physicochemical
detection in this and other species. properties and identities of the odorants (Derby et al., 1996;
Training dogs to detect explosives presents several chal- Kay et al., 2005; Laska and Hudson, 1993; Linster and Smith,
lenges. First, the types of explosives found in IEDs reflect 1999), the complexity of the mixture (Livermore and Laing,
local availability and can vary widely from region to region. 1998a), the relative intensities of the odors (Livermore and
Explosives commonly found in IEDs include organics Laing, 1998b), and olfactory enrichment (Mandairon et al.,
(e.g., 1,3,5-hexahydro-1,3,5-trinitrotriazine [RDX]; 2,4,6- 2006).
trinitrotoluene [TNT]), inorganic oxidizers (e.g., ammo- Our study addressed several of these challenges. In our
nium nitrate [AN], potassium chlorate [PC]), or a com- experiment we explored whether dogs trained to detect
bination (e.g., Amatol – a RDX and AN mixture) (Kopp, one component of an explosive will correctly detect a novel
2008). Further, the use of homemade explosives (HMEs) mixture containing that component. Our second experi-
has recently become more common than commercial and ment assessed the performance of dogs trained to reliably
military explosives (Östmark et al., 2012). Consequently, detect the components of a mixture using a novel device
the absolute and relative amounts of explosive precur- that maintained separation of the components in distinct
sors found in HMEs can vary widely. A second challenge compartments while presumably producing a merged odor
is presented when the base explosive is further modified presentation. The use of this device addresses the need to
with additional gelling agents (e.g., wax or petroleum jelly), develop training aids that can be handled safely and yet
fuels (e.g., diesel fuel or kerosene), or extraneous distract- provide the representative odor profile of the explosive of
ing odors (Kopp, 2008). Therefore, most target (explosive) interest.
odors encountered by dogs under field conditions are
comprised of a combination of many different substances
2. Animals, materials and methods
(Harper et al., 2005), which may differ from those used in
training. Learning to respond upon detection of a trained
2.1. Animals
odor, then, may not generalize to detection of novel odor
combinations. Thus, generalization from trained compo-
Nine male and 11 female field-trial-bred Labrador
nents to novel configurations, such as mixtures composed
Retrievers ranging from 2 to 5 years in age were used.
of trained and untrained odors, is an important feature
All dogs were procured by a private military working dog
of canine scent detection. The concept of generalization,
training firm (K2 Solutions, Southern Pines, NC, USA) from
in which an animal must efficiently and appropriately
field-trial-breeding kennels throughout the United States
respond to novel stimuli based on prior experience with
(USA). Dogs were individually housed in 4.6 m × 1.5 m out-
different stimuli, has been extensively studied (Ghirlanda
door kennels at the K2 Solutions Canine Training Center
and Enquist, 2003; Rilling, 1977; Spence, 1937); how-
(K2) located in Southern Pines, NC, USA. This facility is
ever, olfactory generalization of compounds of military
designed, equipped, and operated to comply with Title 9,
interest in dogs remains largely unexplored (Johnston,
Code of Federal Regulations, parts 1–3 and with the USA
1999).
Department of Defense (DoD) Directive 3216.01 to guaran-
Generalization between a stimulus compound (e.g.,
tee the humane, safe, and necessary use of dogs. Animals
a tone and a click presented together) and its ele-
were fed twice a day with commercial adult large breed dog
ments generally occurs in one of two ways (Bouton
food (Purina ProPlan, St. Louis, MO, USA) and had unlimited
et al., 2012). Elemental, or analytical, processing occurs
access to water. Experienced K2 trainers carried out the
when each element in the compound forms individ-
training and handling of the dogs for the duration of the
ual stimulus–response associations, which are retained
study. All experiments were performed during a 5-week
regardless of combining with other stimuli (Rescorla,
period under ambient weather conditions. Maximum daily
1972). In regards to olfactory processing, if combining two
temperatures ranged from 13 to 33 ◦ C (daily maximum
or more odorants does not alter their individual proper-
mean temperature = 27 ◦ C). All procedures were approved
ties such that the original odors remain identifiable in the
by the North Carolina State University (NCSU) Institutional
resulting mixture, conditioned responses to the individual
Animal Care and Use Committee.
odors should generalize to the mixture (Linster and Smith,
1999). Alternatively, when conditioned stimuli are com-
bined the resulting compound may be perceived as a novel 2.2. Odorants and containers
configuration (Pearce, 1987), referred to as configural or
synthetic processing. Odorants may combine to create a Odor delivery was accomplished using a custom-built
novel, blended mixture in which individual characteris- inverted “T”-shaped odor delivery device (Fig. 1). The odor
tics of each odorant are altered, creating the perception delivery device was constructed using commercially avail-
of a new and unique odor. In this case, conditioning to able 7.62 cm diameter polyvinyl chloride (PVC) pipe fittings
an individual component of the mixture does not produce (JM Eagle, Los Angeles, CA, USA). The two terminal end-caps
generalization to a mixture that contains the component, of the “T” allowed for the placement of PC-based mate-
and vice versa (i.e., conditioning to a mixture does not rials of interest (one or both arms were used depending
produce generalization to the individual components pre- upon the experimental phase). Each terminal end-cap was
sented individually) (Derby et al., 1996; Laing and Francis, attached to a 45◦ elbow which was joined to a “tee” joint
1989; Staubli et al., 1987). The degree of generalization fitting that terminated with a drain fitting and perforated
from components to mixtures may vary depending on a drain cover lid. Separate odor delivery devices were used
86 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
each time. Distractor odors were placed in the seven other produced on site by chemists associated with the USA
odor delivery devices. On Trials 3 and 5 only the seven Naval Surface Warfare Center, Indian Head Division. Mix-
distractor odor delivery containers were present. Finally, tures weighed approximately 200 g and were comprised
Trials 8–10 consisted of control trials to check for the possi- of 140–180 g of PC and 20–60 g of a non-PC component.
bility of dogs responding based on the odor of the packaging The amount of PC and non-PC component used in each
materials used to contain the target odor during training mixture varied depending upon the chemical composition
instead of the PC odor. Three types of packaging materials of the final explosive mixture tested. Because the iden-
were tested, one trial at a time: a cotton bag, a plastic con- tities of the non-PC components used to formulate the
tainer (Tupperware Corporation, Orlando, FL, USA), and a final mixtures were to remain undisclosed, the PC-based
glass container. On each of these trials, four odor delivery mixtures used in Phase 2 and beyond are designated as
devices were present, one of which contained the packag- Mixture 1, Mixture 2, Mixture 3, and Mixture 4. Dogs were
ing material and the other three were left empty. not previously exposed to the non-PC component at any
All dogs were successively run through a trial before the time.
next trial type was set up. Prior to beginning the session, Phase 2 was completed over 2 days, with one session
a randomized list was generated to designate the order in each day consisting of six trials each. Testing was per-
which the dogs were run. The order was re-randomized formed in sequence with all dogs. Eight odor delivery
after Trials 3 and 6. At the start of each trial, a dog was devices were present in a circular formation on each trial.
walked to the testing arena on-leash by a K2 trainer and On Trials 1 and 6, all of the odor delivery devices were
positioned in the designated start area. The dog was then empty. On Trials 2 and 4, distractor odors were placed
removed from the leash and given a command to search. inside each device. On Trials 3 and 5, one odor delivery
During the search trainers remained at the arena entrance, device, randomly chosen, was removed and replaced with
avoiding eye contact with the dog and maintaining neutral a device containing Mixture 1 (Trial 3) or Mixture 2 (Trial 5).
body position in order to minimize unintentional cueing. This sequence of trials was repeated on the second session
On target odor trials (2, 4, and 6), the dog was rewarded the next day except Trial 3 used Mixture 3, and Trial 5
with a Kong® toy upon displaying a correct response at the used Mixture 4. In this phase, trials were conducted in
target location. An additional trainer outside of the testing the same manner as the Phase 1 test trials (see Section
arena and out of view of the dog tossed the toy into the 2.4.1), except the dogs were not rewarded for displaying
arena immediately after a correct response was exhibited. a positive response to the target container. Upon display-
The dog was then removed from the test arena while the ing a positive response the dogs were called out of the
next dog was retrieved by a second trainer. Responses were test arena by the trainer who remained neutral. If a dog
recorded as either positive (dog displayed response within did not indicate a response upon the first pass of the tar-
approximately 1 m of the device) or negative (dog searched get device, the trial continued to allow the dog to search
target device but failed to display correct response). Upon a the remainder of the devices before being called back by
negative response, dogs were allowed to continue search- the trainer. Responses were scored as positive (dog dis-
ing the remainder of the devices in the arena and then played immediate response within 1 m of target odor) or
were removed with no reward. On non-target trials (1, 3, 5, negative (dog failed to respond to the target odor). False
7–10), dogs were allowed to search the test arena and each responses (responding to a non-target device) were also
device sufficiently before being called back by the trainer. recorded. The testing order of dogs was randomized for
Any responses to non-target devices were recorded as a each mixture.
false response. False responses were ignored by trainers, The first two trials of each session in this phase served as
and the dog was allowed to continue searching the remain- exclusion criteria trials based on false responses. Dogs that
der of the test arena before being removed. In order to move exhibited a false response on both Trials 1 and 2 of either of
on to the next experimental phase, a criterion was required the two sessions were removed from testing at this point.
of a positive response on all three of the target odor trials This was implemented in order to eliminate dogs display-
and only dogs that met this criterion continued to the sub- ing high rates of false responding which could invalidate
sequent experimental phase. False responses in this phase their future test results.
were recorded and analyzed, but not used as exclusionary
criteria for moving to the next phase. In between subjects, 2.4.3. Phase 3 – evaluation of novel training device
the odor delivery devices were wiped down with a cloth The purpose of this phase was to evaluate the effective-
and the test arena was lightly raked. After every fourth trial, ness of the odor delivery device as a training method for
all odor delivery devices were removed from the area and odor presentation of binary mixture components (i.e., PC
wiped down, and the entire test arena was raked. Three and a second component). In this phase, PC was placed in
data collectors observed from outside of the testing arena one end cap of the device, while the non-PC component
and independently recorded trial information and results. was placed in the remaining end cap. The design of the
Additionally, all trials were videotaped. device ensured that PC and the second component would
remain physically separated throughout the experiment;
2.4.2. Phase 2 – generalization tests however, the air from each is presumably combined by dif-
The goal of this phase was to evaluate whether dogs fusion and presented as a merged odor to the dog. Ratios
that reliably detected pure PC in Phase 1 would respond and amounts of the two components remained consistent
to a PC-based mixture (i.e., generalization of trained with the amounts used in the actual mixtures during the
responses to novel odor combinations). Each mixture was test phase (see Section 2.4.2). In this phase, dogs received
88 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
Table 1
Individual percent correct for potassium chlorate (PC) detection assessment trials (Phase 1) and PC-based mixture generalization test trials (Phases 2 and
3). Individual total percentage of false responses for the entire session is also reported.
Dog name Phase 1: PC validation Phase 2: generalization test Phase 3: training evaluation
Percent correct Percent false Percent correct Percent false Percent correct Percent false
Individual false response rates over the 12 trials ranged 3.3. Evaluation of novel training device (Phase 3)
from 0 to 25%, with seven of the 16 dogs displaying no false
responses in the entire session (Table 1). False response Next, dogs (n = 16) received additional training corre-
rates were highest during distractor trials (Fig. 2) and on sponding to individual performance during each of the
Trial 2 (Fig. 3). Table 2 shows individual dogs’ results for Phase 2 generalization tests for 7 days until a criterion of
each of the PC-based mixture test trials for the Phase 2 10 consecutive positive responses per odor was achieved.
generalization test (‘pre’). The two dogs (‘Audi’ and ‘Kody’) with 100% response accu-
racy in Phase 2 were included in the remaining phases as
controls. All dogs (n = 16) were then re-tested on the four
PC-based mixtures. Total percentage of dogs correctly res-
ponding to each mixture ranged from 81 to 100% across the
four mixtures (Table 2). Positive responses to the PC-based
mixtures were significantly higher than responses to the
PC-based mixtures in Phase 2 (Fig. 4); significant improve-
ments were seen with Mixture 1 (2 = 22.03, P < 0.0001,
n = 30, df = 1), Mixture 2 (2 = 5.18, P = 0.0229, n = 30, df = 1),
Mixture 3 (2 = 12.53, P = 0.0004, n = 32, df = 1), and Mix-
ture 4 (2 = 7.86, P < 0.005, n = 32, df = 1). Thirteen of the 16
dogs showed a positive response to all four mixtures (100%
response rate); two dogs showed a positive response to
three of the four mixtures (75% response rate), and one dog
performed with a 50% response rate (Table 1). Individual
dogs’ responses to each of the four mixtures before (‘pre’)
and after (‘post’) Phase 3 odor merger device training are
reported in Table 2.
In Phase 3, individual false response rates by dogs
ranged from 0 to 42% over the 12 trials, with nine of the
16 dogs showing no false responses in the entire session
Fig. 3. Overall false response rates (%) as a function of trial number for (Table 2). False response rates were highest during dis-
Phases 1–3. False response rates were calculated by dividing the sum of tractor trials (Fig. 2) and on Trial 2 (Fig. 3).
false responses for each trial in a phase by the total number of dogs tested
in the phase. Phase 1 assessed detection of pure potassium chlorate (PC)
and was completed over 10 consecutive trials, with PC tested on trials 3.4. Variation of odorant quantities
2, 4, and 6. Phases 2 and 3 assessed detection of PC-based mixtures and
were performed in two separate 6-trial sessions, with mixtures tested on
trials 3 and 5. Trial data for the two sessions in each of Phases 2 and 3 are All dogs tested demonstrated a positive response on
combined for that phase. each of the eight trials (100% response rate) in which
90 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
Individual responses to each potassium chlorate (PC)-based mixture after initial odor training with chemically pure PC (‘pre’), after training with an odor delivery device that contained the physically segregated
PC and non-PC components (‘post’), to the four non-PC mixture components presented without PC (‘0-PC’), and to mixtures containing increased (‘+PC’) and decreased (‘−PC’) amounts of PC. Individual responses
−PC
100
nda
nda
nda
nda
nda
nda
nda
nda
nda
1
1
1
1
1
100
+PC
nda
nda
nda
nda
nda
nda
nda
nda
nda
1
1
1
1
1
1
0-PC
nda
nda
nda
nda
nda
nda
nda
nda
nda
14
1
0
0
0
0
0
0
Post
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
100
Mixture 4
Pre
1
1
0
0
0
0
1
1
1
1
1
1
1
1
0
1
69
−PC
100
nda
nda
nda
nda
1
1
1
1
1
1
1
1
1
1
1
100 Fig. 4. Percentage of dogs demonstrating a positive response to the four
+PC
nda
nda
nda
nda
1
1
1
1
1
nda
nda
33
1
1
1
0
0
0
0
0
100
nda
nda
nda
4. Discussion
1
1
1
1
1
1
1
1
1
1
1
1
nda
nda
1
1
nda
nda
nda
0
0
0
0
0
0
0
0
0
0
40
1
1
1
1
1
0
0
0
0
0
0
0
0
nda
1
1
nda
nda
1
1
nda
nda
1
1
1
1
1
0
0
0
0
0
0
0
27
1
1
1
0
0
0
0
0
0
0
0
0
0
Total (%)
Harley
Brutus
Sandy
Drake
Table 2
Annie
Pirate
Sadie
Moss
Kody
Audi
Indy
Dye
is possible that the lack of responses to PC-based mixtures amount of PC, although this cannot be entirely ruled out.
by dogs that were proficient in detecting pure PC was due Nevertheless, our finding that dogs demonstrating profi-
to changes in the perception of PC caused by its interac- ciency in detecting a trained amount of pure PC did not
tion with the non-PC component. This may also explain reliably generalize to the PC-based mixtures, whether due
the variation seen between responsiveness to each of the to changes in amount of PC or interaction with fueling
four mixtures, in that the specific components in each mix- agents, is important and a further understanding of what
ture may have interacted with PC in different ways and factors may facilitate generalization is needed.
to differing degrees. For example, we found that Mixture It is also important to recognize that differences in
1 was associated with the lowest percentage of positive individual animal responses exist despite training. For
responses (25%), indicating that the non-PC component example, we found that less than 50% of the dogs trained
used in this mixture may be more reactive with PC than to pure PC subsequently responded to either Mixture 2
other non-PC components tested. In contrast, Mixture 4 or 3. This finding suggests that perception of these mix-
was associated with the highest positive response rate tures varied greatly between dogs. Moreover, some dogs
(71%) suggesting that the non-PC component in this mix- demonstrated a 100% response rate to the four mixtures,
ture was either more inert or produced minimal changes while some dogs failed to respond to all or most of the mix-
to the odor properties of pure PC. However, given that the tures. Differences in individual responsiveness, sensitivity,
first mixture tested received the lowest response rate and motivation, and experience may contribute to individ-
the last mixture tested received the highest, it is possible ual differences in odor mixture discrimination (Laska and
that order effects were also a contributing factor. Hudson, 1993; Rabin and Cain, 1990). While training with
Our findings are somewhat surprising since previous PC alone appeared to be sufficient to produce generaliza-
studies have shown that dogs are capable of discriminat- tion to PC-based mixtures for some dogs, this method did
ing a target odorant when mixed with various amounts of not produce reliable detection in all dogs that otherwise
another odorant (Johnston, 1999). However, previous stud- consistently detected trained amounts of pure PC.
ies held amounts of the trained target substance constant Our second main objective was to determine whether
throughout tests in which it was mixed with other odors. training the dogs with a blended odor that was developed
In the current study, dogs were initially trained with 450 g from the separated components could enhance detection
of PC but test mixtures contained reduced amounts of PC of the physically mixed PC and non-PC components. To
(140–180 g) due to the explosive properties of the mix- this end, we developed an odor delivery device that safely
ture. It is possible then that the lack of responses to the contained the PC and non-PC components but kept the
PC-based mixtures in Phase 2 was due to the variation in materials physically segregated while presenting a combi-
the amount of PC present in the mixture compared to the nation of the two odorant air streams to the dog. The odor
amount used in training. One possible effect of reducing the delivery device was composed of commercially available
overall amount of PC is that the intensity of the odor was PVC pipe fittings and could be easily constructed and used.
also reduced. Changes in stimulus intensity are known to We found that training with the separated components
produce changes in responding, sometimes referred to as contained within the device significantly improved the
stimulus intensity dynamism (Hull, 1949), an effect that is positive detection rate for PC-based explosives. After train-
unique to stimulus intensity and differs from other stim- ing with the separated components in the device, overall
ulus dimensions such as wavelength, size, and frequency positive response rates to the four PC-based explosive mix-
(Mackintosh, 1974). In classic studies of Pavlovian general- tures increased from 27–69% to 81–100%. Individual dog
ization as well as instrumental generalization, increasing performance also became more uniform. During the first
the intensity of the conditioned stimulus produced an generalization test (Phase 2), only two of the 16 dogs
increase in responding; similarly, a reduction in respon- responded with 100% accuracy. Of the remaining dogs
ding was seen to stimuli less intense than the conditioned that scored below 100% on the first test and proceeded to
stimulus (see Mackintosh, 1974 for review). receive further training using the odor delivery device dur-
In order to evaluate the possibility that varying the ing Phase 3, accuracy increased upon re-test of the mixtures
amount of odorant used could cause the dogs to fail to for 12 of the 14 dogs.
generalize from standard training amounts to novel quan- Overall, these results suggest that in this case, exclu-
tities, additional trials were performed with increased and sively training dogs with a standard amount of a single
decreased amounts of PC. We found that neither increasing target substance was not sufficient to produce reliable gen-
nor decreasing the amount of PC in the mixtures affected eralization to mixtures containing that substance, and that
the dogs’ ability to recognize the target odor and all dogs in order to obtain high levels of positive detection of the
responded to each of the eight novel variations presented. It target binary mixtures, dogs needed to be trained with
is possible, however, that these variations were not robust both of the components. However, results from the Phase
enough to cause a disruption in detection. It is also pos- 3 trials in which dogs were trained to the segregated com-
sible that further training and experience with the odors ponents and subsequently tested on the explosive mixture
during Phase 3 facilitated the dogs’ ability to generalize to indicated that the two odorants did not need to be physi-
the variations. However, a fixed amount of PC was used cally mixed during training in order to produce subsequent
during training (corresponding to the ratios used in mix- generalization to the mixture.
tures during test trials) and the positive responses to the Training and testing of the dogs was constrained by
novel variations suggests that the failure to generalize to certain standard operating procedures used by the K2 facil-
the PC-based mixtures was not likely due to the change in ity that were beyond the investigators’ control. First, for
92 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
safety reasons, trainers were not blinded to the location from Trial 1 (blank trial) to Trial 2 (distractor trial) in Phases
of the target device, presenting the possibility of uninten- 2 and 3 (Fig. 3). Phase 1 consisted of a pattern of alternat-
tional cueing by gaze, posture, and proximity to the dog or ing non-target and target trials, which the dogs may have
odor (Lit et al., 2011). However, dogs in the current study learned. In the following phases, the first target odor trial
were always worked off-leash and handlers remained at a did not appear until the third trial, and thus dogs may have
distance while avoiding eye contact and maintaining neu- anticipated an odor trial on Trial 2. However, given this type
tral posture. The observers and data collectors remained of temporal based responding, false responses would also
outside of the test arena, separated by wooden walls that be expected to occur during the final four non-rewarded
obstructed the dog’s view of the people from the chest trials of Phase 1, in which no false response were recorded.
down. Throughout the study, dogs did not appear to attend The majority of false responses occurred during dis-
to any of the people outside of the testing area as they were tractor trials, and rarely occurred during target odor or
largely out of view. This is likely due to training practices blank trials (Fig. 2). This suggests that another possible
at K2 in which dogs are trained to attend exclusively to the cause for false responses in this study was that the dis-
target objects and odors while ignoring extraneous sources tractor odors elicited responses. Items used as distractors
of distraction including other people, animals, and sounds. may have been composed of materials similar to the target
Further, the low response rates seen in Phase 2 generaliza- odor or in previous training the dogs received prior to this
tion test trials that occurred despite observers’ and trainers’ study. Finally, although care was taken to clean odor con-
awareness of the target location suggests that the dogs did tainers and containers were never used for more than one
not use human directed cues to locate the target. Addi- odorant, the possibility still exists that some cross contam-
tionally, the majority of false responses occurred on trials ination may have occurred during handling of containers.
in which observers knew that no target odor was present False responding may also simply be an artifact of train-
(Fig. 2). Although it cannot be ruled out that some dogs used ing. During initial PC training (Phase 1), dogs were not
unintentional trainer or observer cueing, we feel that it did rewarded on blank trials for correctly rejecting (not respon-
not play a significant role in the dogs’ performance. How- ding to) the non-target containers, nor was a punishment
ever, efforts should be made in future studies to minimize procedure ever used to correct a false response. Thus, the
such a possibility. relative response cost of responding (lying down) at a non-
It may be argued that false responses seen in several of target container was very low, and some dogs may have
the dogs throughout testing may weaken the interpreta- adopted this behavioral strategy. Further, dogs were never
tion of the positive responses seen. However, overall false rewarded during Phase 2 or test sessions in Phase 3 in order
response rates throughout the study were relatively low. In to prevent within-session learning that could affect inter-
Phase 1, false responses were recorded but were not used pretation of any generalization seen. Thus, false responses
as exclusionary criteria. In this phase, dogs simply needed could be interpreted as extinction effects in which the
to demonstrate proficiency in reliably detecting PC. False trained behavior initially increases when reinforcement for
responses at this stage were likely due to the novelty of the a previously rewarded response is terminated (Lerman and
training paradigm. Dogs were still early in their training Iwata, 1995). Indeed, false responses were higher in Phases
and habituating to searching the devices in the test arena. 2 and 3 than in Phase 1 (Fig. 2).
At the start of Phase 2, four dogs were excluded based on Other trials demonstrated that approximately 50% of the
displaying an excessive number of false responses (defined tested dogs showed a positive response to at least one of
as two or more in Trials 1 and 2 of either session) in order to the non-PC components when presented alone. Positive
minimize potential confounding of test data. Of the 16 dogs responses across individual non-PC components ranged
that advanced, five dogs never exhibited a false response in from 14 to 33%. These results indicate that for some dogs,
subsequent phases of the study. The remainder of the dogs responding may have been controlled by the non-PC com-
made an average of 18% false responses in Phases 2 and 3 ponent alone and was independent of the presence of PC,
combined. a possible artifact of the previous training phase in which
Few studies have systematically investigated causes the non-PC component was always part of the odor dis-
for false responses in detection dogs other than those crimination. However, this is not conclusive and responses
induced by handler cueing (Lit et al., 2011). However, false during these trials may be due to reasons previously men-
responses may occur for a variety of other reasons. One tioned. Although not implemented during this study, the
suggested cause of false responding is that dogs may learn use of the non-PC components as distractors during train-
to respond based on learned temporal patterns of rein- ing and a differential reinforcement procedure may have
forcement. Although blank searches were programmed minimized this effect.
into early training in Phase 1, dogs encountered the tar-
get odor, and thus reinforcement, more frequently. Typical 5. Conclusion
early training protocols involve numerous training prob-
lems in which dogs will encounter either continuous or The results from the present study demonstrate that
frequent intermittent reinforcement in a short period of training dogs to detect PC in its pure form may not be suf-
time in order to maintain attention and responding. As ficient to produce reliable, operational levels of detection
a result, dogs may become habituated to the pattern of of PC-based explosive mixtures, although some individ-
repetitive reinforcement and may learn to anticipate rein- ual exceptions were seen. This is an important finding
forcement during a given timeframe (Herstik, 2010). This with implications to canine scent detection training in that
hypothesis may explain the rise in false responses seen single-odor training, which is commonly used, may not
L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93 93
produce adequately trained dogs able to detect complex Hull, C.L., 1949. Stimulus intensity dynamism (V) and stimulus general-
explosive mixtures. Future studies investigating factors ization. Psychol. Rev. 56, 67–76.
Johnston, J.M., 1999. Canine Detection Capabilities: Operational Impli-
that facilitate olfactory generalization would have impor- cations of Recent R&D Findings. Institute for Biological Detection
tant implications in the training of scent detection dogs. Systems, Auburn University, Auburn, AL.
We also found that training dogs with a novel odor delivery Jones, B.M., 2011. Applied behavior analysis is ideal for the development
of a land mine detection technology using animals. Behav. Anal. 34,
device, which produced an odor mixture from segregated 55–73.
components, improved generalization to odor mixtures. Kay, L., Crk, T., Thorngate, J., 2005. A redefinition of odor mixture quality.
Despite certain limitations, this device represents a safe Behav. Neurosci. 119, 726–733.
Kiddy, C.A., Mitchell, D.S., Bolt, D.J., Hawk, H.W., 1978. Detection of estrus-
and effective training option that reduces the need for using
related odors in cows by trained dogs. Biol. Reprod. 19, 389–395.
a final PC-based explosive mixture in canine training. Kopp, C., 2008. Technology of improvised explosive devices. Defense
Today, 46–49.
Laing, D.G., Francis, G.W., 1989. The capacity of humans to identify odors
in mixtures. Physiol. Behav. 46, 809–814.
Acknowledgements Laska, M., Hudson, R., 1993. Discriminating parts from the whole: determi-
nants of odor mixture perception in squirrel monkeys, Saimiri sciureus.
This work was funded through a contract to K2 Solu- J. Comp. Physiol. A 173 (2), 249–256.
Lasseter, A.E., Jacobi, K.P., Farley, R., Hensel, L., 2003. Cadaver dog and
tions, Inc. from the United States Office of Naval Research
handler team capabilities in the recovery of buried human remains in
(ONR). Lisa Albuquerque, Michael Hoglund and other mem- the southeastern United States. J. Forensic Sci. 48 (3), 617–621.
bers of the ONR provided logistical and technical support. Lerman, D.C., Iwata, B.A., 1995. Prevalence of the extinction burst and its
attenuation during treatment. J. Appl. Behav. Anal. 28 (1), 93–94.
Lisa Albuquerque also designed and constructed the odor
Linster, C., Smith, B.H., 1999. Generalization between binary odor mixtures
delivery device and holds a provisional patent (61/814,385) and their components in the rat. Physiol. Behav. 66 (4), 701–707.
for the device. We thank the USA Naval Surface Warfare Lit, L., Schweitzer, J.B., Oberbauer, A.M., 2011. Handler beliefs affect scent
Center, Indian Head Division and K2 Solutions personnel detection dog outcomes. Anim. Cogn. 14 (3), 387–394.
Livermore, A., Laing, D.G., 1998a. The influence of chemical complexity
for facilitating this study as well as Joseph Albuquerque for on the perception of multicomponent odor mixtures. Percept. Psy-
his input on technical matters. chophys. 60 (4), 650–661.
Livermore, A., Laing, D.G., 1998b. The influence of odor type on the dis-
crimination and identification of odorants in multicomponent odor
mixtures. Physiol. Behav. 65 (2), 311–320.
References Mackintosh, N.J., 1974. The Psychology of Animal Learning. Academic
Press, London/New York.
Adams, G.J., Johnson, K.G., 1994. Sleep, work, and the effects of shift work Mandairon, N., Stack, C., Linster, C., 2006. Olfactory enrichment improves
in drug detector dogs Canis familiaris. Appl. Anim. Behav. Sci. 41, the recognition of individual components in mixtures. Physiol. Behav.
115–126. 89, 379–384.
Bouton, M.E., Doyle-Burr, C., Vurbic, D., 2012. Asymmetrical generaliza- Östmark, H., Wallin, S., Ang, H.G., 2012. Vapor pressure of explosives: a
tion of conditioning and extinction from compound to element and critical review. Propell. Explos. Pyrot. 37, 12–23.
element to compound. J. Exp. Psychol. Anim. Behav. Process. 38 (4), Pearce, J.M., 1987. A model for stimulus generalization in Pavlovian con-
381–393. ditioning. Psychol. Rev. 94, 61–73.
Cornu, J.N., Cancel-Tassin, G., Ondet, V., Girardet, C., Cussenot, O., 2011. Pfiester, M., Koehler, P.G., Pereira, R.M., 2008. Ability of bed bug-detecting
Olfactory detection of prostate cancer by dogs sniffing urine: a step canines to locate live bed bugs and viable bed bug eggs. J. Econ. Ento-
forward in early diagnosis. Eur. Urol. 59 (2), 197–201. mol. 101 (4), 1389–1396.
Dean, E.E., 1972. Training dogs for narcotic detection. Technical report No. Pickel, D., Manucy, G.P., Walker, D.B., Hall, S.B., Walker, J.C., 2004. Evidence
LWL-CR-6ODJ71. http://www.dtic.mil/cgi-bin/GetTRDoc?Location= for canine olfactory detection of melanoma. Appl. Anim. Behav. Sci. 89,
U2&doc=GetTRDoc.pdf&AD=AD0749302 (5.02.13). 107–116.
Derby, C.D., Ache, B.W., Kennel, E.W., 1985. Mixture suppression in olfac- Rabin, M.D., Cain, W.S., 1990. Attention and learning in the perception of
tion: electrophysiological evaluation of the contribution of peripheral odor mixtures. In: Laing, D.G., Cain, W.S. (Eds.), Perception of Complex
and central neural components. Chem. Senses 10, 301–316. Smells and Tastes. Academic Press, Sydney, pp. 173–188.
Derby, C.D., Hutson, M., Livermore, B.A., Lynn, W.H., 1996. Generaliza- Rescorla, R.A., 1972. “Configural” conditioning in discrete-trial bar press-
tion among related complex odorant mixtures and their components: ing. J. Comp. Physiol. Psychol. 79, 307–317.
analysis of olfactory perception in the spiny lobster. Physiol. Behav. Rilling, M., 1977. Stimulus control and inhibitory processes. In: Honig, L.,
60 (1), 87–95. Staddon, W.K.J.E.R. (Eds.), Handbook of Operant Behavior. Prentice-
Faust, A.A., de Ruiter, C.J., Ehlerding, A., McFee, J.E., Svinsas, E., van Rhee- Hall, Englewood Cliffs, NJ, pp. 432–480.
nen, A.D., 2011. Observations on military exploitation of explosives Spence, K., 1937. The differential response in animals to stimuli varying
detection technologies. In: Proc. SPIE 8017, Detection and Sensing of within a single dimension. Psychol. Rev. 44, 430–444.
Mines, Explosive Objects, and Obscured Targets XVI, 801716, May 23, Staubli, U., Fraser, D., Faraday, R., Lynch, G., 1987. Olfaction and the “data”
2011, http://dx.doi.org/10.1117/12.886391. memory system in rats. Behav. Neurosci. 101, 757–765.
Fischer-Tenhagen, C., Wetterholm, L., Tenhagen, B.A., Heuwieser, W., Walczak, M., Jezierski, T., Gorecka-Bruzda, A., Sobczynska, M., Ensminger,
2011. Training dogs on a scent platform for oestrus detection in cows. J., 2012. Impact of individual training parameters and manner of taking
Appl. Anim. Behav. Sci. 131, 63–70. breath odor samples on the reliability of canines as cancer screeners.
Furton, K.G., Myers, L.J., 2001. The scientific foundation and efficacy of J. Vet. Behav. 7, 283–294.
the use of canines as chemical detectors for explosives. Talanta 54, Watts, H.G., 2009. The consequences for children of explosive rem-
487–500. nants of war: Land mines, unexploded ordnance, improvised
Gazit, I., Terkel, J., 2003. Explosives detection by sniffer dogs following explosive devices, and cluster bombs. J. Pediatr. Rehabil. Med. 2,
strenuous physical activity. Appl. Anim. Behav. Sci. 81, 149–161. 217–227.
Ghirlanda, S., Enquist, M., 2003. A century of generalization. Anim. Behav. Willis, C.M., Church, S.M., Guest, C.M., Cook, W.A., McCarthy, N., Bransbury,
66, 15–36. A.J., Church, M.R.T., Church, J.C.T., 2004. Olfactory detection of human
Harper, R.J., Almirall, J.R., Furton, K.G., 2005. Identification of dominant bladder cancer by dogs: proof of principle study. Br. Med. J. 329 (7468),
odor chemicals emanating from explosives for use in developing 7.
optimal training aid combinations and mimics for canine detection. Wilson, C., 2007. Improvised explosive devices (IEDs) in Iraq and
Talanta 67, 313–327. Afghanistan: Effects and countermeasures. CRS Report for Congress.
Herstik, M., Herstik, M., 2010. The false alert: reasons and remedies. Det- http://www.fas.org/sgp/crs/weapons/RS22330.pdf (accessed
onator 37, 43–45. 5.02.13).