Content uploaded by Katherine R Naish
Author content
All content in this area was uploaded by Katherine R Naish on Aug 12, 2014
Content may be subject to copyright.
1 3
Exp Brain Res
DOI 10.1007/s00221-012-3367-2
RESEARCH ARTICLE
To eat or not to eat? Kinematics and muscle activity
of reach-to-grasp movements are influenced by the action
goal, but observers do not detect these differences
Katherine R. Naish · Arran T. Reader ·
Carmel Houston-Price · Andrew J. Bremner ·
Nicholas P. Holmes
Received: 15 August 2012 / Accepted: 28 November 2012
© Springer-Verlag Berlin Heidelberg 2012
is going to be brought to the mouth or placed until after
the grasp has been completed. This suggests that the early
kinematic differences are either not visible to observers, or
that they are not used to predict the end-goals of actions.
These data are discussed in the context of the mirror neuron
system.
Keywords Reach to grasp · Kinematics ·
Electromyography · Action prediction ·
Movement planning · Pre-grasp
Introduction
Reaching towards a fruit bowl to retrieve a shiny red apple
is an action that requires little thought. In performing this
movement, it is likely that the acting individual knows
exactly what they are going to do with the apple once it has
been grasped. But, to an observer, the goal of the action
might not be so clear. Will the fruit be brought directly from
the bowl to the mouth of the grasper, handed to the child who
is standing nearby, or moved aside so that a different piece
of fruit can be reached? Being able to predict the post-grasp
action might be useful to a hungry onlooker, and it appears
that humans are able, from an early age, to anticipate actions
after viewing only the initial stages of movement (Sebanz
and Shiffrar 2007; Southgate et al. 2010).
Some researchers have attributed our ability to predict
actions to the mirror neuron system (MNS; Fogassi et al.
2005). Discovered in monkeys, mirror neurons are cells
in the premotor cortex which respond during both action
execution and action observation (di Pellegrino et al.
1992). Studies using techniques such as transcranial mag-
netic stimulation (e.g. Fadiga et al. 1995) and functional
imaging (e.g. Iacoboni et al. 1999) provide evidence for a
Abstract Recent evidence suggests that the mirror neuron
system responds to the goals of actions, even when the end
of the movement is hidden from view. To investigate whether
this predictive ability might be based on the detection of
early differences between actions with different outcomes,
we used electromyography (EMG) and motion tracking to
assess whether two actions with different goals (grasp to eat
and grasp to place) differed from each other in their initial
reaching phases. In a second experiment, we then tested
whether observers could detect early differences and predict
the outcome of these movements, based on seeing only part
of the actions. Experiment 1 revealed early kinematic differ-
ences between the two movements, with grasp-to-eat move-
ments characterised by an earlier peak acceleration, and
different grasp position, compared to grasp-to-place move-
ments. There were also significant differences in forearm
muscle activity in the reaching phase of the two actions. The
behavioural data arising from Experiments 2a and 2b indi-
cated that observers are not able to predict whether an object
Electronic supplementary material The online version of this
article (doi:10.1007/s00221-012-3367-2) contains supplementary
material, which is available to authorized users.
K. R. Naish (*) · A. T. Reader · C. Houston-Price · N. P. Holmes
School of Psychology and Clinical Language Sciences,
University of Reading, Earley Gate, Whiteknights,
Reading RG6 6AL, UK
e-mail: k.r.naish@pgr.reading.ac.uk
K. R. Naish · N. P. Holmes
Centre for Integrative Neuroscience and Neurodynamics,
University of Reading, Earley Gate, Whiteknights,
Reading RG6 6AL, UK
A. J. Bremner
Department of Psychology, Goldsmiths, University of London,
New Cross, London SE14 6NW, UK
Exp Brain Res
1 3
similar system in humans. Such studies indicate that activ-
ity in the human motor system is similar when a person
observes a goal-directed action and when they perform
the same action. According to recent reports, the MNS is
activated even when part of an action is hidden from view
(e.g. Villiger et al. 2011), suggesting that this system has
the ability to predict the outcomes or goals of actions after
seeing only the beginning. It is theorised that the appropriate
‘motor chain’ is activated in the observer at the beginning
of the action; thus, an internal representation of the viewed
movement is activated regardless of whether the entire
action is seen (Fogassi et al. 2005). As the MNS is said to
‘transform visual information into knowledge’ (Rizzolatti
and Craighero 2004), an important question to ask is: what
visual information allows an observer to predict the out-
come of an action? Previous studies suggest that observers
can infer a great deal from viewing the kinematics of move-
ments alone (e.g. Runeson and Frykholm 1983; Manera et
al. 2011). So, do differences in the movements themselves
allow us to predict whether an object will be, for example,
brought to the mouth or thrown away?
To answer this question, we first need to know whether
the initial stages of movement do in fact differ according to
the goal of an action. Indeed, it has been shown that reach
kinematics are influenced by what is going to be done
with an object once it has been grasped. Marteniuk et al.
(1987), for example, found that a reach-to-grasp movement
that was followed by fitting an object into a ‘tight fitting
well’ was characterised by a longer overall movement time,
lower peak velocity and a longer deceleration phase, com-
pared to a reach-to-grasp movement followed by throwing
the object into a relatively large box. Similarly, Ansuini
et al. (2006) found that reaching movements were slower
and were characterised by a more gradual shaping of the
hand to the object when the post-grasp movement required
greater precision. Another study found that reaching move-
ments towards a bottle were longer in duration when the
subsequent action was pouring liquid from the bottle before
placing it down, compared to when the post-grasp move-
ment was simply placing (Schuboe et al. 2008). Grasp
position is also affected by the end-goal of a movement,
with objects being grasped lower down when they are to be
moved upwards, and higher up when they are to be moved
downwards (Cohen and Rosenbaum 2004; Schuboe et
al. 2008). In addition, the initial stages of reach-to-grasp
movements vary according to the size (e.g. Marteniuk et
al. 1987; Pryde et al. 1998; Armbrüster and Spijkers 2006),
texture (e.g. Fikes et al. 1994) and weight (e.g. Eastough
and Edwards 2007) of the object being grasped.
In this study, we were specifically interested in how
reach-to-grasp movements differ depending on whether the
grasped object is going to be brought to the mouth or placed
in another location, and whether the object is a food or a
non-food item. Hand-to-mouth actions are one of the most-
practiced movements performed by humans, emerging first
at the foetal stage of development (for review see Rochat
1993; Bremner and Cowie in press), and remaining a large
part of our daily movement repertoire throughout life. In
monkeys, at least, there are mirror neurons which respond
selectively to the observation of eating and food-related
actions (Ferrari et al. 2003; Fogassi et al. 2005). In Fogassi
et al. (2005) study, some neurons responded selectively to
the observation of grasp-to-eat, or to grasp-to-place actions
even when the final eating or placing action was hidden
from view, indicating that the monkey MNS differentiates
eating and placing actions from an early phase of the move-
ment. To ascertain whether this differentiation was based
on kinematic differences between the movements, Fogassi
also analysed the hand trajectory, velocity and finger aper-
ture of the monkeys’ movements. To investigate whether the
differences in mirror neuron responses could be explained
by kinematic differences, Fogassi and colleagues included
two different ‘grasp-to-place’ conditions, one in which the
monkey placed the object in a container next to its mouth
and another in which the monkey placed the object on a
table in front of its body. The data showed similar mirror
neuron responses to the observation of both grasp-to-place
conditions, both of which were different to the response to
observing the grasp-to-eat movement. Whilst these results
indicate that kinematic differences do not explain the differ-
ent neural responses to the observation of eating and placing
actions, Fogassi et al. did indeed find that grasp-to-eat and
grasp-to-place movements differed in terms of peak velocity
and peak finger aperture.
As food-directed actions are also used as stimuli in
human MNS research (e.g. Cheng et al. 2007; Sartori et al.
2011a, b; Villiger et al. 2011), we set out to build a profile
of the kinematics and muscle activity underlying grasp-to-
eat and grasp-to place movements in humans. Interestingly,
Fogassi et al. (2005) also found that a subset of neurons that
responded most strongly to grasp-to-eat actions were more
responsive to grasp-to-place actions when the object was a
food compared to a non-food item. We therefore also inves-
tigated the extent to which movements towards food and
non-food objects differ, and whether there is an effect of
object–action congruency (i.e. whether grasp-to-eat actions
are performed differently when they are directed towards
foods versus non-foods).
The first aim of this study was therefore to establish
whether, and how, reach-to-grasp movements differ depend-
ing on whether the grasped object is going to be placed,
or brought to the mouth. To this end, in Experiment 1, we
tracked the positions of the wrist, index finger, and thumb
and recorded muscle activity using electromyography
(EMG), whilst participants performed grasp-to-place and
grasp-to-eat actions towards either a plum or a ball. Based
Exp Brain Res
1 3
on previous findings of the effect of the post-grasp action
on pre-grasp kinematics, it was anticipated that the reaching
phases of grasp-to-eat and grasp-to-place actions would dif-
fer from each other. A second aim was to establish whether
observers can distinguish grasp-to-place and grasp-to-eat
actions, after viewing only the early stages of movement.
We carried out two behavioural studies (Experiments 2a and
2b) in which participants viewed portions of actions and
were asked to predict the outcome of these actions.
Experiment 1
Methods
Participants
Ten participants aged 21–34 years (M = 26.2, SD = 4.6)
were recruited from the School of Psychology and Clini-
cal Language Sciences at the University of Reading. One
participant was left-handed by self-report, and two of the
authors (KN and NH) were participants. The experimental
procedures were approved by the local ethics committee
(refs: 2011/05/NH; UREC 11/11); participants gave written,
informed consent; and the experiments were conducted in
accordance with the Declaration of Helsinki.
Apparatus and stimuli
The position of the hand during movements was recorded
using a Polhemus Fastrak (Polhemus Inc., Colchester, VT,
USA) motion tracking system with six degrees of freedom
(X, Y, Z, azimuth, elevation and roll). Three data points
(thumb, index finger and wrist) were acquired, each at
40 Hz. EMG data were acquired with an AD Instruments
Powerlab 16/30 and two dual bioamplifiers (ADI, Colorado
Springs, CO, USA). EMG electrodes were attached to the
skin to record activity over four muscles of the participants’
dominant hand and forearm: the first dorsal interosseous
(FDI) and the thenar eminence (i.e. abductor pollicis brevis,
APB) in the hand, and the forearm flexor (i.e. flexor digito-
rum superficialis, FDS) and extensor muscles (i.e. extensor
digitorum communis, EDC). Two circular electrodes were
placed on the skin overlying each muscle, approximately
2 cm apart, in a belly-tendon fashion, and ground electrodes
were attached to the styloid process of the wrist and the lat-
eral epicondyle of the elbow. The experiment was controlled
and data were acquired using custom software written in
MATLAB 2010b (Mathworks, Inc.). All experimental and
analysis scripts are available from the last author’s website
(http://neurobiography.info).
The food object was a (red) Victoria plum (an approxi-
mate spheroid ~3.5 cm high and ~4.0 cm wide and deep).
The non-food object was an orange table tennis ball
(a sphere, 4.0 cm in diameter).
Design
The experiment followed a repeated measures design. The
four experimental conditions were derived from crossing
the two variables: action (grasp to eat and grasp to place)
and object (plum and ball). The four conditions were run in
blocks of 20 trials, with two blocks per condition, giving a
total of eight blocks per participant. For each participant, the
order of blocks 1–4 was pseudorandomised and the order of
blocks 5–8 was the reverse of 1–4.
Procedure
The participants were seated in a chair in front of a table
where the target object was placed (Fig. 1). Kinematic
markers were placed on the fingernails of the thumb and
index finger, and on the dorsal aspect of the wrist, and the
self-adhesive EMG electrodes were attached to the skin.
The markers and wires were secured and positioned so
that they did not restrict the participants’ movements. The
actions were always performed with the dominant hand,
and the participants were told that their movements, includ-
ing the speed of their movements, should be ‘as natural as
possible’. In the grasp-to-place conditions, following an
Fig. 1 Schematic diagram of the set-up of Experiment 1. ‘A’ repre-
sents the starting position, where the participant’s hand started at the
beginning of each block, and where it was brought back to between
trials. ‘B’ represents the starting position of the object in all trials,
whilst ‘C’ is the location that the object was moved to in grasp-to-
place blocks. The square represents the Polhemus transmitter unit
Exp Brain Res
1 3
auditory signal, the participant moved their hand from the
starting position (location A; 10 cm from the table edge, and
approximately 20 cm from their body, in front of them) to
the object location (location C; 40 cm from the table edge),
grasped the object and placed it in location B (20 cm from
the table edge). Participants were told to hold their hand
steady in the ‘placed’ position (location B) until a sec-
ond auditory signal three seconds after the first, at which
point they were to move the object back to location C and
immediately return their hand to the starting position to
await the next trial. In the grasp-to-eat conditions, the par-
ticipants moved their hand from location A to the object,
before bringing the object from location C to their mouth,
and holding it just in front of their mouth until the second
auditory signal. They then returned the object to location
C and moved their hand back to location A. To make the
grasp-to-eat movements as natural (i.e. as similar to an eat-
ing movement, without the object actually being put into the
mouth) as possible, the participants were told to perform
the action as if they were going to eat the object, so that the
mouth opened as the object approached the mouth, and the
final position was the object held at the open mouth. The
movement to be performed was demonstrated to the par-
ticipants by the experimenter before the start of each block,
and the participants were told when to move relative to the
auditory signals.
Kinematic data acquisition began immediately prior to
the first auditory signal and continued for 3 s. EMG data
were acquired continuously and later segmented offline
using trigger signals sent from the experimental control
computer to the data acquisition computer.
Data analysis
Kinematic data
Filtering and extraction of variables The kinematic data
(X, Y, Z position in cm) were filtered with a second-order
dual-pass Butterworth filter with a high-frequency cut-off of
10 Hz. The data of the left-handed participant were mirror-
reversed across the midline (inverting the Y values of the
markers). The position data were differentiated once to com-
pute velocity and a second time to compute acceleration.
The movement was analysed in two parts: the pre-grasp
and grasp phase (henceforth ‘reach to grasp’), and the post-
grasp phase (‘grasp to place’ or ‘grasp to eat’). For each
part of the movement, a total of 38 variables were extracted
from the kinematic data. Movement onset (RT) and move-
ment offset (ET) for the pre-grasp reach were defined based
on the velocity of the wrist after trial onset: the movement
began (RT) when the velocity of the wrist first exceeded
5 cm/s and ended (ET) when the velocity first decreased
below 10 cm/s. Movement duration (MT) was calculated
by subtracting RT from ET. The same criteria were used to
define the RT, ET and MT of the post-grasp movement, but
with the constraint that the RT of the post-grasp movement
was at least one sample (25 ms) after the pre-grasp ET.
Five variables associated with ‘grip aperture’ (the distance
between the thumb and index finger) were calculated: grip
aperture at movement offset (GAET), peak grip aperture
(PGA), time of peak grip aperture (PGAT), peak velocity
of grip aperture (PVGA) and the time of peak velocity of
grip aperture (PVGAT). The following variables were cal-
culated separately for each of the three kinematic markers
(thumb, index finger and wrist): peak velocity (PV), peak
acceleration (PA), peak deceleration (PD), time of peak
velocity (PVT), time of peak acceleration (PAT), time of
peak deceleration (PDT), path length (the distance travelled
between movement onset and offset) and position at move-
ment offset (XET, YET and ZET).
Trials were excluded automatically by the analysis
script if they met certain criteria, which differed for each
part of the action due to the considerable kinematic differ-
ences between the pre- and post-grasp movements. Most
of the excluded trials were due to hand movement at the
start of the trial (before the first auditory signal), unsmooth
trajectories or participants missing the start signal. For the
pre-grasp, trials were excluded if the RT was less than
100 ms, if the calculated ET was less than 50 ms after the
RT or if the path length was shorter than 17 cm or longer
than 60 cm. For the post-grasp, trials were automatically
excluded if ET was less than 50 ms after the RT, if the
ET was within 100 ms (four samples) of the last recorded
sample (i.e. very late, slow or long movements), if path
length was shorter than 12 cm or longer than 90 cm or if
the velocity at the end of the trial was greater than 10 cm/s
(i.e. the hand was still moving at trial offset). Further tri-
als were excluded on visual inspection if they were clear
outliers within the dataset of the participant. In total, 2.9 %
of trials were excluded from further analysis (of 1,600 tri-
als in total, 40 trials were excluded automatically, and a
further six trials manually).
Data reduction Because many of the kinematic variables
were correlated with one another (e.g. peak velocity and
peak acceleration of the wrist in participant 1, r = 0.822,
p < .001), principal component analyses (PCAs) were con-
ducted to reduce the number of redundant kinematic vari-
ables, thereby decreasing the number of statistical com-
parisons required and simplifying the interpretation of the
data. It has been suggested that PCA is preferable to factor
analysis where the primary goal of analysis is data reduction
(de Vaus 2002). Oblique (direct oblimin) rotation was used,
as this method of rotation is recommended if it is expected
that some components might correlate with each other (e.g.
Costello and Osborne 2005).
Exp Brain Res
1 3
Pre-grasp A preliminary PCA with all 37 variables
(excluding ET, which was redundant given that MT was
based on RT and ET) suggested seven components. How-
ever, as these were difficult to interpret and the rotation
failed to converge, we decided to analyse the variables asso-
ciated with temporal properties of the movement separately
from the spatial variables. Thus, two separate PCAs were
conducted with the aim of reducing the 37 variables into
meaningful temporal and spatial components. The first of
these PCAs comprised 23 temporal variables: RT, MT, PV,
PA, PD, PVT, PAT, PDT, PGAT, PVGA and PVGAT. The
Kaiser–Meyer–Olkin (KMO) measure of sampling ade-
quacy confirmed that the data were suitable for this analy-
sis (KMO = 0.942). The second PCA (KMO = 0.480) was
conducted on the 14 spatial variables: ET position (XET, YET
and ZET) and path length for all three markers, in addition
to GAET and PGA.
Post-grasp Whilst our main focus of interest was the
pre-grasp movement, the kinematics of the post-grasp
movement phase (placing or bringing to the mouth) were
nonetheless recorded and analysed. Grasp-to-place and
grasp-to-eat movements were analysed separately because
of the large kinematic differences between the two. A PCA
was conducted with all 32 variables (the same variables as
in the pre-grasp analysis, but excluding variables relating to
grip aperture, as this did not change during the post-grasp
movement).
EMG data
Muscle activity was recorded throughout the whole experi-
ment, but only data arising from the pre-grasp part of the
trial were analysed. The data were acquired at a sampling
rate of 1 kHz and bandpass filtered at 1–1,000 Hz by the
data acquisition software. For offline analysis, the data were
then rectified and filtered with a second-order dual-pass
Butterworth filter with a bandpass of 1–15 Hz.
EMG data were analysed by three paired t tests con-
ducted on each sample. Due to the large number of data
points generated and analysed for every trial, bootstrapping
was carried out to provide 95 % confidence intervals for any
significant differences found in muscle activity between dif-
ferent conditions. Bootstrapping was carried out for three
epochs of the data: standardised across the whole pre-grasp
movement (by interpolating to 1,000 samples), relative to
RT (99 ms before to 100 ms after RT; 200 samples) and
relative to the peak acceleration (99 ms before to 100 ms
after PAT; 200 samples). For each participant’s dataset, the
data were randomly resampled with replacement across tri-
als and conditions, separately for each muscle and partici-
pant. For each iteration of resampling, within-participants t
statistics were calculated by comparing the average of two
resampled conditions with the average of two others (i.e.
similar to the main effects of action, object and their interac-
tion). We then calculated, for each muscle in each dataset,
the number of significant samples (i.e. where one data point
was significantly higher (t(9) > 2.26, p < .05, two-tailed) for
one of the contrasted conditions than the other), as well as
the number of sequences and the length of each sequence of
significant samples. This process was repeated for 10,000
iterations, and distributions were constructed of: (a) the total
number of significant samples per iteration, (b) the number
of sequences of significant samples per iteration and (c) the
lengths of significant sequences of samples. From these null
distributions, p values were assigned to the sequences of
significant samples found in the real data.
All data treatment and analysis were carried out using
MATLAB (R2010b) and SPSS (version 19).
Results
Pre-grasp kinematics
Extracted temporal and spatial movement components For
both PCAs, components were extracted which had eigen-
values exceeding Kaiser’s criterion of 1. In interpreting the
components, we referred mainly to the pattern matrix, but
loadings in the structure matrix were also considered when
the two matrices differed considerably. Variables were inter-
preted as loading onto a component ‘significantly’ if they
had a loading value of 0.4 or above (Stevens 2002).
The 23 temporal variables were reduced to five com-
ponents, which together explained 75.1 % of the variance.
Component loadings can be seen in Online Resource 1.
Component 1 was interpreted as representing ‘slowness of
movement’, with positive loadings of PD, MT, PGAT and
PDT, and negative loadings of PV, PA and PVGA. Compo-
nent 2 had positive loadings of MT, PVT and PDT, so was
interpreted as representing long movements with late peak
velocities. Component 3 represented time of peak accelera-
tion of the thumb and the index finger, with PAT of both
loading negatively. Component 4 had positive loadings of
RT and PVGAT; thus, a high score represented movements
that started later and in which the fastest finger-thumb open-
ing occurred later. Finally, Component 5 represented time
of peak acceleration of the wrist, with only this one variable
loading significantly.
The 14 spatial variables were reduced to five compo-
nents, which together accounted for 80 % of the variance.
Component loadings can be seen in Online Resource 2.
Component 1 seemed to represent movements with a posi-
tion at ET in which the thumb and index finger were further
forward on the object (further from the participant’s body)
and the wrist was further back (closer to the participant’s
body), and in which the index finger was further right and
Exp Brain Res
1 3
the thumb further left. Component 2 comprised high posi-
tive loadings of XET position of the thumb and index finger
(i.e. the thumb and index finger were further back on the
object), and high positive loadings of ZET for all markers
(i.e. the thumb, index finger and wrist were all higher up at
ET). This component was interpreted as grasp position; spe-
cifically, a high score represented a grasp position in which
the hand was further back and higher up on the object. Com-
ponent 3 represented path length, a high score on this com-
ponent representing a short path. Component 4 represented
an ET position in which the thumb and wrist were further to
the left (negative loadings of YET). Component 5 had nega-
tive loadings of PGA and GAET. The mean and standard
deviation of all the individual kinematic variables prior to
the PCA are provided in Table 1.
Effects of action and object type on movement compo-
nents Bonferroni correction was used to protect against
type 1 error. This was done separately for the temporal
and spatial analyses. Because there were five components
extracted for each of these, the alpha significance criterion
for all tests was decreased to 0.01.
Separate analyses of variance (ANOVAs) were con-
ducted to assess the effects of action and object on each
of the extracted components. Main effects of action on
both Component 3 (PAT of the index finger and thumb;
F(1,9) = 12.91, p = .006, η2 = 0.589) and Component 5
(PAT of the wrist; F(1,9) = 17.21, p = .002, η2 = 0.657)
were revealed. In both cases, peak acceleration occurred
earlier in grasp-to-eat than in grasp-to-place movements.
There was also a significant interaction between action
and object on Component 5 (F(1,9) = 12.88, p = .006,
η2 = 0.588). Post hoc ANOVAs showed that peak accel-
eration occurred earlier in grasp-to-eat than grasp-to-place
movements for both the plum and the ball, although the
simple effect of action was larger for actions towards the
ball (F(1,9) = 32.44, p < .001, η2 = 0.783) than the plum
(F(1,9) = 5.60, p = .042, η2 = 0.384). Separate analyses
of placing and eating actions showed that peak acceleration
was reached earlier for grasp-to-place actions towards the
plum than those towards the ball (F(1,9) = 5.46, p = .044,
η2 = 0.378), whereas there was no significant difference
between grasping-to-eat the plum and grasping-to-eat the
ball (F(1,9) = 3.70, p = .086, η2 = 0.291). The average peak
acceleration times for each condition are plotted in Fig. 2.
Because PAT of the wrist was the only variable to load
significantly onto Component 5, additional analyses were
carried out using the original variable. These confirmed the
same main effect (F(1,9) = 11.25, p = .008, η2 = 0.555)
and interaction (F(1,9) = 6.48, p = .031, η2 = 0.018).
The same post hoc tests performed on the original PAT
variable showed significant differences between grasp-
to-eat plum and grasp-to-eat ball actions (F(1,9) = 6.02,
p = .037, η2 = 0.400), and between grasp-to-eat and grasp-
to-place actions towards the ball (F(1,9) = 26.35, p = .001,
η2 = 0.745).
Analyses of the spatial components revealed a signifi-
cant main effect of action and a marginal effect of object
on grasp position (Component 2). Objects were grasped
higher up and further back in grasp-to-place as compared to
Table 1 Mean (SD) of all pre-grasp kinematic variables
Thumb Index Wrist
RT (ms) n/a n/a 317 (100)
MT (ms) n/a n/a 732 (136)
PV (cm/s) 80.4 (18.0) 87.9 (18.9) 68.9 (15.8)
PA (cm/s/s) 635 (246) 685 (262) 450 (191)
PD (cm/s/s) −368 (169) −434 (193) −273 (134)
PVT (ms) 331 (88.4) 293 (84.6) 320 (74.7)
PAT (ms) 102 (70.1) 102 (70.1) 111 (68.8)
PDT (ms) 569 (173) 570 (183) 545 (163)
Path (cm) 35.4 (3.10) 38.7 (3.28) 29.8 (2.75)
XET (cm) 36.6 (0.834) 30.6 (0.768) 44.8 (1.74)
YET (cm) −1.45 (0.938) 1.52 (1.28) 7.99 (2.12)
ZET (cm) 1.23 (0.611) 2.79 (0.977) 7.05 (1.33)
PGA (cm) 3.65 (1.23)
PGAT (ms) 541 (191)
PVGA (cm/s/s) 20.3 (9.25)
PVGAT (ms) 251 (151)
GAET (cm) 2.63 (1.14)
Fig. 2 Interaction between action type and object type on the time
(in milliseconds) of peak acceleration of the wrist (Experiment 1).
The grey bars represent actions towards the plum, and the white bars
represent actions towards the ball. Error bars represent 95 % confi-
dence intervals
Exp Brain Res
1 3
grasp-to-eat actions (F(1,9) = 13.56, p = .005, η2 = 0.601),
and the grasp was higher and further back on the plum than
on the ball (F(1,9) = 10.20, p = .011, η2 = 0.531). These
were the only effects which reached significance after Bon-
ferroni correction, but there were also marginally significant
effects of object on Component 1 (F(1,9) = 7.94, p = .020,
η2 = 0.469) and of action on Component 3 (F(1,9) = 6.78,
p = .029, η2 = 0.430). Actions with the plum had higher
scores on Component 1 than actions with the ball; thus,
participants’ grip on the plum was different to their grip
on the ball. The second marginal effect indicated shorter
path lengths for grasp-to-place as compared to grasp-to-eat
actions.
Post-grasp kinematics
Extracted temporal and spatial movement components
Grasp to place Analysis of all 32 variables (KMO = 0.866)
revealed six components. Component loadings are shown in
Online Resource 3. Component 1 was interpreted as rep-
resenting ‘slowness of movement’ as it was characterised
by positive loadings of RT, MT, PD and PVT, and negative
loadings of PV and PA. Component 2 represented move-
ments with a late PD of the index finger and an ET posi-
tion in which the thumb and index were further forward, the
thumb was to the left and the index finger to the right and
low down on the object. Component 3 was characterised by
a long path length and an ET position in which the wrist
was further back on the object. Component 4 was charac-
terised by a late PV and late PA time of the thumb and the
index finger. Component 5 was characterised by a late PD
of the wrist, and an ET position in which the wrist was fur-
ther to the left. Component 6 represented movements with
a late PA of the wrist, and an ET position in which the wrist
was further forward, and both the wrist and the thumb were
higher up.
Grasp to eat Analysis of all 32 variables (KMO = 0.858)
revealed five components; component loadings can be seen
in Online Resource 4. Component 1 represented movements
with a late RT, long MT, low and late PV, low PA, and a
high and late PD. Component 2 represented movements
with long path lengths and in which the thumb, index fin-
ger and wrist were all closer to the body at ET. This was
interpreted as representing either the distance from location
A to the participant’s mouth, or how close to their mouth
the participants held the object. Component 3 represented
grasp position, a high score on this component representing
movements in which the thumb, index finger and wrist were
all further right, and the index finger and wrist were higher
up at ET. Component 4 represented movements with late PA
times for all markers, and late PV of the thumb and index
finger. Component 5 represented movements with a short
path length of the thumb and in which all markers were
lower at ET; thus, a high score on this component could rep-
resent movements in which the hand was held lower at the
face.
Effects of object type on grasp-to-place and grasp-to-eat
movement components There were no significant effects
of object on the post-grasp movement after Bonferroni cor-
rection. In both the grasp-to-place and grasp-to-eat data,
the only effect of object which was near-significance was
on Component 1. In both cases, this component appeared
to represent ‘slowness of movement’, a high score on this
component representing movements with a long RT, long
MT, low and late PV, low PA, and high and late PD in all
markers. For the grasp-to-eat movement, this component
also had significant loadings of the position (‘left–right’)
of the thumb and index, with both being further to the left
at ET. In both movements, it was found that actions with
the plum were slower than actions with the ball (grasp to
place: F(1,9) = 6.25, p = .034; grasp to eat: F(1,9) = 6.80,
p = .028).
EMG data
Based on the multiple comparisons carried out on data
from four different muscles, differences in muscle activity
between conditions (eat vs. place; plum vs. ball; interaction
between action and object) were only considered significant
when p was .01 or lower. Contrasting the grasp-to-eat and
grasp-to-place trials revealed differences in activity in the
FDS. For the standardised data (1,000 interpolated samples
across the whole pre-grasp movement, from four samples
before RT to four samples after ET), activity in the FDS was
significantly higher (bootstrapped p < .0001) in grasp-to-
eat than in grasp-to-place movements at two phases of the
pre-grasp movement. The first significant sequence of time-
points started from trial onset, showing higher activation in
the FDS in grasp-to-eat trials from approximately 300 ms
(average RT) before the movement had started (a sequence
of 102 consecutive samples; p < .0001). The second phase at
which activation was greater in grasp-to-eat movements was
from approximately 170 ms after trial onset (76 samples;
p = .0053). The greater activity in eating compared to plac-
ing trials was confirmed by the analysis of the data relative
to RT, with one significant sequence starting 100 ms before
the RT (117 samples; p < .0001). A plot of the activity in
all muscles, comparing grasp-to-eat with grasp-to-place
actions, is provided in Fig. 3.
Discussion
The initial movements involved in object-directed actions
vary depending on the properties of the object itself (e.g.
Exp Brain Res
1 3
Marteniuk et al. 1987) and the action following the initial
movement (e.g. Schuboe et al. 2008). Our data confirm
these previous observations, demonstrating effects of both
target object and subsequent action on reach-to-grasp move-
ments. In addition, we found an interaction between object
and action type, which could reflect an incongruity effect
of grasping-to-eat a non-food item, or simply an effect
of object which is only apparent when the item has to be
moved upwards to the mouth.
Our finding that peak acceleration occurs earlier in grasp-
to-eat than in grasp-to-place actions might be due to the fact
that the subsequent hand-to-mouth action requires greater
control or precision than placing the object. This would
support the results of Schuboe et al. (2008), who reported
peak acceleration occurring earlier in reaching movements
when the post-grasp action involved pouring liquid from
the grasped object compared to when the object was sim-
ply placed on a shelf. The difference between grasp-to-eat
and grasp-to-place actions is especially interesting in the
light of the recent finding that the MNS responds differ-
ently to observing the initial stages of movement depending
on whether the object is subsequently placed or brought to
the mouth (Fogassi et al. 2005). Although Fogassi and col-
leagues’ control measure (the inclusion of a grasp-to-place
movement in which the object was placed in a container next
to the mouth) indicated that differences in MNS responses
to eating and placing actions could not be explained by
kinematic differences, this does not exclude the possibility
that kinematic differences can be detected by the MNS and
contribute to the MNS response in humans.
The interaction between post-grasp action and target
object in our data is a novel finding, which could be inter-
preted in different ways. We investigated the effects on time
of peak acceleration by analysing both the component scores
and the original variable. As peak acceleration time was the
only variable to contribute significantly to the component,
this discussion will focus on the differences revealed when
the raw peak acceleration time was compared between the
conditions. Whilst no difference was found between reach-
ing for the plum and reaching for the ball in the grasp-to-
place condition, peak acceleration occurred significantly
earlier in reaches towards the ball than in reaches towards
the plum when the subsequent action was bringing to the
mouth. If an earlier peak acceleration reflects more care
being taken in the movement (e.g. for movements requiring
greater precision), then it is possible that participants took
more care over movements in which the post-grasp action
and target object were incongruent, that is, ‘eating’ the ball.
Indeed, an effect of object action congruency was found by
Begliomini et al. (2007), who looked at the kinematics of
‘natural’ movements, in which the type of grasp was appro-
priate to the object (a precision grip for a small object, or
a whole hand grasp for a larger object), and ‘constrained’
movements, in which the grasp type was not the most
Fig. 3 Differences between
conditions in EMG activity
recorded over four muscles of
the hand and forearm (Experi-
ment 1). Bold lines represent
mean activity, and fine lines
represent the standard error. The
plot compares activity in each
of the muscles during grasp-to-
eat and grasp-to-place actions;
zero on the y-axis represents no
difference between the two con-
ditions. The black bars below
the FDS data highlight the
points at which the data deviate
significantly from zero, that is,
where activity was significantly
greater in the grasp-to-eat than
in the grasp-to-place conditions
Exp Brain Res
1 3
appropriate for the object (a whole hand grasp of the small
object, or precision grip for the large object). Begliomini et
al. found that the time of movement initiation, as well as the
time of peak grip aperture, occurred later for constrained
than for natural movements.
Alternatively, the earlier peak acceleration time in this
case might simply reflect a more ‘awkward’ movement, as
the action is not one that we would normally perform. It is
equally possible that the interaction may simply reflect the
fact that grasping to eat is an action which requires more
care as the object is transported upwards to the mouth, and
so the effect of object is apparent only for this action, that
is, if grasp-to-eat movements require more care or precision,
the kinematics may be more susceptible to properties such
as the smoothness of the object. As discussed below, the dif-
ference in grasp position could indicate the need for a more
secure grip on the ball than the plum. If this is the case, the
interaction between action and object on peak acceleration
time could reflect a combination of the more ‘careful’ move-
ment and less graspable object. This interpretation would
be consistent with the findings of Sartori et al. (2011a, b)
and their suggestion that there is an interplay between object
affordances and the end-goal of actions. Sartori and col-
leagues compared grasp-to-move and grasp-to-pour move-
ments towards a bottle which was either full or half-full of
water, and either concave (easier to grasp) or straight-edged.
Movement time was found to be longer when the action was
‘pouring’ compared to when it was simply moving the bot-
tle, arguably due to the greater precision and care required
for pouring actions. This difference, however, was not
found when the bottle was a concave shape, indicating that
the effect of action was only present for the less graspable
object. Furthermore, some effects of action and object shape
(greater peak grip aperture for pouring than moving, and for
pouring from the straight bottle compared to pouring from
the concave bottle) were only present when the bottle was
half-empty (i.e. it weighed less). The findings of Sartori and
colleagues, and our own finding of an object–action inter-
action, could indicate that effects of action type (namely,
how precise an action is) are dependent on the stability of
the grasp on the object. When the grasp is relatively secure
(for example on the plum compared to the ball, or on the
concave bottle compared to the straight bottle in Sartori’s
experiment), there are no discernible effects of the action
itself.
The difference in grasp position between the plum and
the ball—specifically, that the index finger and thumb were
placed further back and higher up on the plum than on the
ball—could be due to the fact that a plum would more often
be grasped with the intention to eat, and this grasp position
would result in the fingers being less of an obstacle when the
plum is bitten into. As this difference between the objects
was not influenced by the subsequent action (i.e. there was
no interaction between action and object in this variable),
this would have to be an automatic positioning of the hand
in response to an object that could be eaten, rather than a
grasp determined by the immediate intention of the actor.
Another possibility is that the grasp on the ball, with the
thumb and index finger lower down and further forward
on the object, would have enabled a more secure grip on
the ball, which had a more slippery surface than the plum.
It could be argued that the higher-friction surface of the
plum, as well as its less symmetrical shape, ‘affords’ grasp-
ing more than does the texture and shape of a table tennis
ball. Indeed, affordances such as the shape and texture of
objects are known to influence the kinematics of reaching
and grasping (e.g. Mon-Williams and Bingham 2011; Flat-
ters et al. 2012). The effect of action on grasp position indi-
cated that the finger and thumb were placed lower down on
the object when it was to be brought to the mouth rather than
placed on the table. As well as potentially providing a more
secure grip on the object when it was going to be brought
to the mouth rather than placed down, this could also reflect
an effect of the final position of the object, supporting pre-
vious findings that grasp height is inversely related to final
position height (Cohen and Rosenbaum 2004; Schuboe et
al. 2008).
In terms of movement execution, our results confirm
what has been found previously. The fact that kinematics
are influenced from an early stage by the end-goal indicates
forward planning of movements, with movements either
following an internal model defined before the movement
begins, or being corrected online during the movement based
on sensory feedback resulting from movement execution
(see Desmurget and Grafton 2000, for review). These data
could also have implications for the recognition of actions
during action observation; if the kinematics of movement
differ according to what is going to be done with an object
before the object is reached, perhaps it is possible to pre-
dict actions from seeing only the beginning of movements.
Based on these findings, Experiments 2a and 2b explored
whether people can predict actions based on viewing the ini-
tial reach-to-grasp stage of movements alone.
Experiment 2a
In this experiment, participants viewed videos of the same
reach-to-grasp actions that were performed in Experiment
1 and were asked to predict whether the object was going
to be placed or eaten. The duration of movement viewed
by participants on each trial was determined by their (cor-
rect or incorrect) response on the previous trial according to
an adaptive staircase procedure. This enabled us to estimate
how much of the action participants needed to see to cor-
rectly predict the movement outcome.
Exp Brain Res
1 3
Methods
Participants
Twelve healthy participants, aged between 19 and 55 years
(M = 29.2; SD = 11.1), were recruited from the School of
Psychology and Clinical Language Sciences at the Univer-
sity of Reading. None of the participants were actors in the
video stimuli. The procedures of Experiments 2a and 2b were
approved by the local ethics committee (refs: 2012/035/NH;
UREC 11/11). Participants gave written, informed consent,
and the experiments were conducted in accordance with the
Declaration of Helsinki.
Stimuli
Stimuli consisted of 96 unique video clips, taken from three
actors, each showing a hand reaching for and grasping either
a plum (48 videos) or a ball (48 videos). The actors in the
videos were all white Caucasian; one was female and two
were male. The plum and ball featured in the videos were
of the same type as those used in Experiment 1. The actors
were filmed whilst performing the movement as per the pro-
tocol of Experiment 1 (performing the same grasp-to-place
and grasp-to-eat actions, in time with the same auditory sig-
nals as used in the first experiment). As in Experiment 1,
after the grasp, the object would either be brought to the
mouth or placed in location B (see Fig. 1). The videos were
filmed from three perspectives: above the actor, in front
of the actor and from the left-hand side of the actor. Faces
were occluded to eliminate distractions from the movement
itself, and cues in the actors’ head position or facial expres-
sion that might have provided information about the actors’
intentions. In the ‘above’ and ‘side’ views, only the reach-
ing arm was visible, whilst the ‘front’ views also showed
the actors’ chests, but not their necks or heads. The object
was visible throughout the video clips. Of the 96 videos,
48 showed grasp-to-place actions and 48 showed grasp-to-
eat actions. Within each of those subsets, 24 were actions
towards the plum and 24 were towards the ball, and within
each of those subsets, 8 were ‘front’, 8 were ‘above’, and 8
were ‘side’ views. The videos were presented using MAT-
LAB (R2010a) and the Psychophysics Toolbox version 3
(Brainard 1997).
Design
The video clips were presented in four blocks of 24 tri-
als, in a pseudorandomised order for each participant. A
QUEST (Quick Estimation of Threshold) adaptive staircase
procedure (Watson and Pelli 1983) was used to determine
the threshold proportion of the video required for correct
discrimination between eating and placing movements. Due
to the kinematic and muscle activity differences between
grasping the two different objects (revealed in Experiment
1), separate thresholds were determined for predicting the
goal of actions towards the plum and the ball. Each thresh-
old was found twice, in counterbalanced order across partic-
ipants: once with the video duration based on the proportion
of the pre-grasp movement only (M1), and once based on
the proportion of the whole movement duration (M2). These
two different threshold methods were used because, whilst
the two movements (grasp to eat and grasp to place) were
different in duration, the initial reach-to-grasp duration was
similar. On each trial, the QUEST procedure updated the
underlying probability distribution function for the thresh-
old, based on the participant’s responses on the previous
trials. Thus, depending on whether the participant made a
correct or incorrect prediction of the post-grasp movement,
they would be shown a lesser or greater proportion (respec-
tively) of the movement in the next trial. The standard
Psychtoolbox QUEST parameters were used (beta = 3.5,
delta = 0.05, grain = 0.01, pThreshold = 0.82), and stair-
cases began with a stimulus duration corresponding to the
pre-grasp and grasp movement.
Procedure
Participants were seated in front of a computer screen and
keyboard, and were asked to observe the presented actions
closely. After each video, they made a judgement as to
whether the action would result in the object being brought
to the mouth (‘EAT’) or placed on the table in location B
(Fig. 1; ‘PLACE’). On each trial, after the QUEST-deter-
mined length of video was played, a screen displaying the
question ‘Eat or place?’ was presented and remained until
the participant made a key press response. Participants were
told to press the ‘e’ key on the keyboard if they thought the
object would be brought to the mouth, or the ‘p’ key if they
thought the object would be placed. Feedback was given on
each trial: the word ‘Correct’ or ‘Incorrect’ was presented
immediately after participants made their response.
Data analysis
For both M1 and M2, threshold was generated for each par-
ticipant for both ball and plum conditions. These threshold
proportions were then adjusted by multiplying M1 with the
mean M1 video duration and M2 with the mean M2 video
duration. This gave a threshold measured in numbers of
frames (i.e. the number of frames participants needed to
view in order correctly to predict the outcome of the action).
A 2 × 2 repeated measures ANOVA was used to look for
differences in threshold, with threshold type (M1 and M2)
and object type (plum and ball) as predictor variables. The
percentage of ‘eat’ responses given by the participants
Exp Brain Res
1 3
was calculated separately for the ball and plum trials, and
a paired t test was conducted to assess whether there was
any bias in participants’ prediction of action depending on
whether the action was towards a plum or a ball. A further
t test was run to assess whether there was any difference
in the total number of ‘eat’ and ‘place’ responses given. To
investigate whether participants were better at predicting the
actions from one angle than another, thresholds were calcu-
lated for each of the three angles (front, above and side), and
a within-subject ANOVA was performed with accuracy as
the dependent and angle as a predictor variable.
Results
There was no significant effect of threshold method
(F(1,11) = 4.292, p = .063 η2 = 0.281), or object
(F(1,11) = 0.549, p = .474, η2 = 0.048), and no interac-
tion between threshold type and object (F(1,11) = 1.572,
p = .236, η2 = 0.125). The mean threshold duration for M1
was 28.3 ± 0.3 frames and for M2 was 27.8 ± 0.3 frames.
The mean threshold for plum was 27.9 ± 0.2 frames and for
ball was 28.1 ± 0.3 frames. The overall mean duration was
28 ± 0.2 frames. Whilst there was no difference between
plum and ball trials in the percentage of ‘eat’ responses
given by participants (t(11) = −1.747, p = .108), partici-
pants were significantly more likely to judge actions as ‘eat’
(53 %) than ‘place’ (47 %; t(11) = 3.246, p = .008).
To confirm that the participants were not able accurately
to predict the subsequent action from viewing the pre-grasp
alone, and to find out how many frames were required for
participants to perform significantly above chance level,
we calculated the accuracy of the participants’ predictions
after seeing different lengths of videos. Figure 4 shows
the percentage accuracy across all participants, for each
proportion of movement viewed (ranging from one frame
before the grasp to 14 after it). Because the number of
frames viewed on each trial was based on each partici-
pant’s accuracy on previous trials, not all participants saw
all lengths of video. Therefore, there are some proportions
of movement (for example, six frames before the grasp)
for which the percentage accuracy is based on very few
trials. Consequently, the number of frames before or after
the grasp was included in Fig. 4 only if the number of trials
available fell inside the central 95 % of the distribution of
trials per frame number. The figure clearly shows that par-
ticipants needed to see at least six frames after the grasp,
in order to perform at a level significantly above chance
(50 %).
Separate analyses of participants’ responses to each cam-
era angle revealed mean thresholds of 26, 30 and 27 frames,
respectively, for the front, above and side angle trials. Thus,
participants needed to see approximately 16 % (front), 37 %
(above) and 20 % (side) of the post-grasp movements in
order to accurately predict the action. The ANOVA showed
that the participants’ predictions were most accurate when
viewing the videos filmed from the ‘front’ angle, and least
accurate when viewing the videos taken from the ‘above’
angle (F(1.7, 18.6) = 24.36, p < .001).
Discussion
The results of Experiment 2a indicate that people need to
see at least part of the post-grasp action before they can
correctly predict the action outcome. The mean duration
Fig. 4 Percentage accuracy of
predictions based on viewing
different proportions of move-
ment. The average percentage
accuracy (across all partici-
pants) on the y-axis is plotted
against the number of frames
viewed relative to the ‘grasp’
frame. Error bars represent
95 % confidence intervals. The
solid black horizontal line on
the figure depicts the number
of frames required to perform
at 82 % accuracy (the threshold
level used by QUEST in our
experiment). The dotted black
line marks the 50 % accuracy
(chance) level, showing that
participants had to see six or
more frames after the grasp to
perform significantly above
chance
Exp Brain Res
1 3
required for accurate prediction was 28 frames (~1167 ms),
which represented slightly different stages of the action in
each of the videos used, but was always later than the time
at which the kinematic differences were detected in our
first experiment. The average duration of M1 (up until the
start of the lift) shown in the videos was 916 ms, and the
duration of the post-grasp movement was 945 ms. There-
fore, participants needed to see around 27 % (251 ms) of
the post-grasp movement in order to correctly predict the
final action.
The absence of a significant effect of object indicates that
there was no difference in participants’ ability to predict the
outcome of actions towards the plum and the ball. There
was also no difference between the proportions of ‘eat’ and
‘place’ responses when participants were observing actions
towards the plum compared to when they were observing
actions towards the ball. So, perhaps surprisingly, viewing
a plum did not appear to bias participants towards an ‘eat’
response (nor did the ball bias participants towards a ‘place’
response). Interestingly, the number of ‘eat’ responses was
higher overall than the number of ‘place’ responses, indi-
cating that people were more likely to interpret actions as
eating than placing.
The fact that the participants were more accurate in
predicting actions viewed from the front, and least accu-
rate for those viewed from above, may be explained as an
effect of familiarity. Both eating and placing actions are
commonly viewed from a frontal viewpoint. In social eat-
ing contexts, it is common for people to eat meals facing
each other. Similarly, we often view placing actions from
the front, for example, when a person hands us an object or
takes an object from us. In contrast, it is quite rare for us to
view actions from above, so it is likely that the participants
had little experience of viewing the actions from this angle.
The increased performance for the front view compared to
the side view is perhaps the most interesting difference, as
it is not uncommon to view grasp-to-eat or grasp-to-place
actions from the side. One explanation for the difference is
that people are better at predicting actions viewed from the
front because we pay greater attention to actions directed
towards us, and thus have greater visual experience of the
finer details of these movements. When a hand is reaching
towards an observer’s body (as is perceived in the ‘front’
view video clips), it is likely that the observer is expected
to perform an action themselves (for example, in the case
of shaking hands, or when giving or receiving an object).
Therefore, it would make sense that we pay more attention
to actions viewed from this angle and would consequently
be better at predicting the outcome of such actions. It must
be noted, however, that even from the ‘front’ viewpoint,
the participants could only correctly distinguish eating and
placing actions after seeing a considerable portion of the
post-grasp movement.
In summary, the data from Experiment 2a indicate that
participants need to see more than the initial phase of a reach-
ing action before they can judge the intention of that action.
Whilst there may be distinct differences in grasp kinematics
and muscle activity associated with pre-grasp movements
of grasp-to-eat and grasp-to-place actions (Experiment 1),
observers are either not able to detect these differences, or
do not use them to predict whether a grasping action will
lead to eating or placing.
Experiment 2b
Our final experiment considered whether people are able
to predict post-grasp actions based on observing (exclu-
sively) the hand’s position on an object. Experiment 1
showed that the index finger and thumb are placed lower
down and further forward on the target object when the
object is going to be brought to the mouth, compared
to when it is going to be placed. Although the results of
Experiment 2a indicate that observers are unable to infer
the subsequent action based on viewing the grasp posi-
tion, we wanted to confirm this by isolating the view of the
hand on the object. In the videos presented in Experiment
2a, the grasp itself was always preceded by the pre-grasp
movement, so it is possible that other factors impeded par-
ticipants’ ability to predict the action. By isolating grasp
position, we aimed to establish whether observers can
predict actions based on the differences in grasp position
alone. To this end, participants in this study were presented
with a single frame of video which showed the hand’s final
position on the grasped object before the object was lifted
from its starting position.
Methods
Participants
Twelve healthy participants aged 18–33 years (M = 25.5,
SD = 4.43) were recruited from the School of Psychology
and Clinical Language Sciences at the University of Read-
ing. None of them had participated in Experiment 2a. Two
of the authors (KN and AR) were participants.
Stimuli
The same video clips presented to participants in Experi-
ment 2a were used in this experiment, but this time only
the ‘grasp’ frame of each video was presented. For vid-
eos in which more than one video frame showed the hand
grasping the object in its starting location, the frame
immediately before the hand started to lift the object was
selected.
Exp Brain Res
1 3
Design and procedure
All 96 video frames were presented in a pseudorandomised
order for each participant, in four blocks of 24 trials. On each
trial, the single video frame was presented for 1,000 ms. As
in Experiment 2a, this was followed by an ‘Eat or place?’
prompt, which remained on the screen until the participant
responded by pressing the ‘e’ or ‘p’ key to denote ‘eat’ or
‘place’.
Data analysis
We used the binomial distribution to assess whether the
number of correct responses for each participant was greater
or lesser than that expected by chance and a t test against the
chance level of 0.5 across participants. As participants were
more likely to give an ‘eat’ response in Experiment 2a, we
also ran a t test to compare the number of ‘eat’ and ‘place’
responses in this experiment.
Results
The number of correct predictions ranged from 42 to 55
(M = 49.5, SD = 3.90), so the mean percentage of correct
responses was just, but not significantly, higher than chance
(M = 51.6 %, SD = 4.06; t(11) = 1.334, p = .209). The
binomial tests confirmed that none of the participants per-
formed significantly above or below chance. In addition,
there was no difference between the number of ‘eat’ and
‘place’ responses given by participants (t(11) = −0.650,
p = .529).
Discussion
These data confirm the results of Experiment 2a, that observ-
ers are not able to predict whether an object is going to be
placed or brought to the mouth, based on seeing either the
pre-grasp movement or the grasp position alone.
General discussion
Studies on the neural effects of action observation have sug-
gested that the mirror neuron system (MNS) can anticipate
action outcomes from an early stage in the observed action
(Umiltà et al. 2001; Fogassi et al. 2005; Villiger et al. 2011).
The present experiments aimed to investigate whether the
predictive ability of the MNS might be due to visible differ-
ences between movements with different goals. To address
this question, we compared muscle activity and movement
kinematics between grasp-to-eat and grasp-to-place actions,
and between actions towards food and non-food objects.
After discovering differences in kinematics and muscle
activity based on the subsequent action and object type, we
ran two behavioural experiments to assess whether people
are able to detect these differences and use them to predict
what is going to be done with an object after it has been
grasped.
The results of Experiments 2a and 2b indicate that,
despite differences in the pre-grasp stage of movement, par-
ticipants are not able to distinguish grasp-to-eat and grasp-
to-place movements until they have seen at least part of the
post-grasp movement. This suggests that either observers do
not detect the early kinematic (and muscular) differences, or
that they do not utilise such differences to predict the out-
come of observed actions. This finding is surprising, as the
results of previous studies suggest that observers can pre-
dict and recognise actions, and even infer properties of the
acting individual and the object, based on viewing the kin-
ematics of movements (e.g. Runeson and Frykholm 1983;
Manera et al. 2011; Stapel et al. 2012). We know from previ-
ous studies that adult participants are good at making goal
predictions on the basis of action observation, including
predicting whether an action is going to be cooperative or
competitive (Manera et al. 2011; Sartori et al. 2011a), pre-
dicting the weight of an object to be lifted (from the way the
actor approaches the object), and predicting whether or not
an actor has ‘deceptive’ intentions (Runeson and Frykholm
1983; Sebanz and Shiffrar 2007).
There remains the possibility that the kinematic differ-
ences between the grasp-to-eat and grasp-to-place actions
depicted in our stimuli were too subtle to guide observers’
predictions. However, this seems unlikely to be the case
given that action prediction in other studies can often occur
in particularly pared down circumstances. For example,
in work by Stapel et al. (2012), observers had to predict
whether a walking individual was going to start crawling
or to continue walking. Observers were able to predict the
subsequent action accurately, even in trials in which all con-
textual information was eliminated.
Our preferred interpretation of our findings is that most
people are not able to detect subtle kinematic differences
between, specifically, pre-grasp movements with different
subsequent outcomes. This has important implications for
the MNS literature, as it places limitations on the apparent
ability of the MNS to infer action goals (e.g. Villiger et
al. 2011) on the basis of kinematic differences in reach-
ing behaviours. As suggested by Fogassi et al. (2005),
the MNS response may rely on cues such as the context
in which the action is observed (e.g. an actor opening an
empty container immediately before grasping an object),
or the type of object being acted upon (e.g. a food item is
likely to be taken to the mouth). Fogassi and colleagues
did find that mirror neurons that were selective for grasp-
to-eat actions responded more strongly to grasp-to-place
actions directed towards food than grasp-to-place actions
Exp Brain Res
1 3
directed towards non-food objects. This could suggest
that the chain of motor commands associated with eating
(i.e. the ‘eating chain’; Fogassi et al. 2005, p. 666) is acti-
vated whenever an observer sees a food object, regardless
of the subsequent action. In other words, seeing a per-
son reaching towards an object which could be eaten may
lead to activation which represents the ‘eating’ goal of
the action rather than another possible goal such as ‘plac-
ing’. In our experiment, participants would have learned
quickly that both the food and non-food object could be
‘eaten’ or ‘placed’, so we did not necessarily expect par-
ticipants to predict the action based on object suitability.
In a ‘real-world’ context, however, perhaps seeing a plum
would lead observers to predict a ‘grasp-to-eat’ rather
than ‘grasp-to-place’ action.
It is, of course, possible that MNS responses are based
on a combination of kinematic and contextual cues, so that,
when the final goal of an action cannot be inferred from
the kinematics of a movement, context can be used to infer
intention. Indeed, Stapel et al. (2012) showed that action
prediction is more accurate when contextual cues are avail-
able, but that actions can be inferred using kinematic cues
alone.
The fact that participants were unable to report the kin-
ematic differences in our study does not preclude the possi-
bility that the MNS is nonetheless able to detect differences
in movement kinematics. It is possible that our participants
did not recognise these kinematic differences to an extent
that allowed them consciously to know whether an action
was going to bring the object to the mouth, but that, none-
theless, the differences were detected by a mechanism that
influences the mirror neuron response. Indeed, there is evi-
dence that some of our perception of actions takes place at
an unconscious level (e.g. see Blake and Shiffrar 2007 for a
review), so it is feasible that kinematic differences are ‘mir-
rored’ but not recognised by the observer. It should be noted,
however, that implicit processing of kinematic differences
between the actions would likely enhance the participants’
performance on the action prediction task even if they had
no conscious awareness of the differences, so we find this
explanation unlikely.
In conclusion, the present findings indicate that there are
early kinematic differences in the reach-to-grasp phase of
eating and placing actions, but these appear to be too subtle
to be detected by an observer. This places an important qual-
ification on the putative ability of the MNS to distinguish
reaching actions with different goals (e.g. eating vs. placing;
Fogassi et al. 2005) on the basis of kinematic differences
between those actions. We speculate that, where intention
cannot be inferred from kinematics, action prediction may
be based on contextual cues such as the type of object to be
acted on.
References
Ansuini C, Santello M, Massaccesi S, Castiello U (2006) Effects of
end-goal on hand shaping. J Neurophysiol 95:2456–2465
Armbrüster C, Spijkers W (2006) Movement planning in prehension:
do intended actions influence the initial reach and grasp move-
ment? Mot Control 10(4):311–329
Begliomini C, Caria A, Grodd W, Castiello U (2007) Comparing natu-
ral and constrained movements: new insights into the visuomo-
tor control of grasping. PLoS ONE 2(10):e1108. doi:10.1371/
journal.pone.0001108
Blake R, Shiffrar M (2007) Perception of human motion. Ann Rev
Psychol 58:47–73
Brainard DH (1997) The psychophysics toolbox. Spat Vis 10(4):433–436
Bremner AJ, Cowie D (in press) Developmental origins of the hand in
the mind, and the role of the hands in the development of the mind.
In: Radman Z (ed) The hand: Organ of the mind. Cambridge: MIT
Press
Cheng Y, Meltzoff AN, Decety J (2007) Motivation modulates the
activity of the human mirror-neuron system. Cereb Cortex
17(8):1979–1986
Cohen RG, Rosenbaum DA (2004) Where grasps are made reveals
how grasps are planned: generation and recall of motor plans. Exp
Brain Res 157(4):486–495
Costello AB, Osborne JW (2005) Best practices in exploratory factor
analysis: four recommendations for getting the most from your
analysis 10(7): 1–9
de Vaus DA (2002) Analyzing social science data. Sage Publications,
London
Desmurget M, Grafton ST (2000) Forward modeling allows feed-
back control for fast reaching movements. Trends Cogn Sci
4(11):423–431
di Pellegrino G, Fadiga L, Fogassi L, Gallese V, Rizzolatti G (1992)
Understanding motor events: a neurophysiological study. Exp
Brain Res 91(1):176–180
Eastough D, Edwards MG (2007) Movement kinematics in prehension
are affected by grasping objects of different mass. Exp Brain Res
176(1):193–198
Fadiga L, Fogassi L, Pavesi G, Rizzolatti G (1995) Motor facilitation
during action observation: a magnetic stimulation study. J Neuro-
physiol 73(6):2608–2611
Ferrari PF, Gallese V, Rizzolatti G, Fogassi L (2003) Mirror neurons
responding to the observation of ingestive and communicative
mouth actions in the monkey ventral premotor cortex. Eur J Neu-
rosci 17(8):1703–1714
Fikes TG, Klatzky RL, Lederman SJ (1994) Effects of object texture
on precontact movement time in human prehension. J Mot Behav
26(4):325–332
Flatters IJ, Otten L, Witvliet A, Henson B, Raymond JH, Culmer P,
Bingham GP, Wilkie RM, Mon-Williams M (2012) Predicting
the effect of surface texture on the qualitative form of prehension.
PLoS ONE 7(3):e32770. doi:10.1371/journal.pone.0032770
Fogassi L, Ferrari PF, Gesierich B, Rozzi S, Chersi F, Rizzolatti G
(2005) Parietal lobe: from action organization to intention under-
standing. Science 308:662–667
Iacoboni M, Woods RP, Brass M, Bekkering H, Mazziotta JC, Riz-
zolatti G (1999) Cortical mechanisms of human imitation. Science
286(5449):2526–2528
Manera V, Becchio C, Cavallo A, Sartori L, Castiello U (2011) Coopera-
tion or competition? Discriminating between social intentions by
observing prehensile movements. Exp Brain Res 211(3–4):547–556
Marteniuk RG, Mackenzie CL, Jeannerod M, Athenes S, Dugas C
(1987) Constraints on human arm movement trajectories. Can J
Psychol 41(3):365–378
Exp Brain Res
1 3
Mon-Williams M, Bingham GP (2011) Discovering affordances that
determine the spatial structure of reach-to-grasp movements. Exp
Brain Res 211:145–160
Pryde KM, Roy EA, Campbell K (1998) Prehension in children and
adults: the effects of object size. Hum Mov Sci 17:743–752
Rizzolatti G, Craighero L (2004) The mirror-neuron system. Ann Rev
Neurosci 27:169–192
Rochat P (1993) Hand-mouth coordination in the newborn: Morphol-
ogy, determinants, and early development of a basic act. In: Sav-
elsbergh GJP (ed) The development of coordination in infancy.
Elsevier, Amsterdam, pp 265–288
Runeson S, Frykholm G (1983) Kinematic specifications of dynam-
ics as an informational basis for person-and-action perception:
Expectation, gender recognition, and deceptive intention 112:
585–615
Sartori L, Cavallo A, Bucchioni G, Castiello U (2011a) Corticospinal
excitability is specifically modulated by the social dimension of
observed actions. Exp Brain Res 211(3–4):557–568
Sartori L, Straulino E, Castiello U (2011b) How objects are grasped:
the interplay between affordances and end-goals. PLoS ONE
6(9):e25203. doi:10.1371/journal.pone.0025203
Schuboe A, Maldonado A, Stork S, Beetz M (2008) Subsequent actions
influence motor control parameters of a current grasping action.
In: The 17th IEEE international symposium on robot and human
interactive communication, 2008. RO-MAN 2008, pp 389–394
Sebanz N, Shiffrar M (2007) Bluffing bodies: inferring intentions from
actions. Perception 36(Suppl):176
Southgate V, Johnson MH, El Karoui I, Csibra G (2010) Motor system
activation reveals infants’ on-line prediction of others’ goals. Psy-
chol Sci 21(3):355–359
Stapel JC, Hunnius S, Bekkering H (2012) Online prediction of others’
actions: the contribution of the target object, action context and
movement kinematics. Psychol Res 76:434–445
Stevens J (2002) Applied multivariate statistics for the social sciences.
Lawrence Erlbaum, Mahwah
Umiltà MA, Kohler E, Gallese V, Fogassi L, Fadiga L, Keysers C, Riz-
zolatti G (2001) I know what you are doing. A neurophysiological
study. Neuron 31(1):155–165
Villiger M, Chandrasekharan S, Welsh TN (2011) Activity of human
motor system during action observation is modulated by object
presence. Exp Brain Res 209(1):85–93
Watson AB, Pelli DG (1983) QUEST: a bayesian adaptive psychomet-
ric method. Percept Psychophys 33(2):113–120
