Emotional language processing:

How mood afects integration

processes during discourse
comprehension
1. Introduction
Feeling cheerful, satisfed, or gloomy is a common experience that can color much of our cognition:
Research in psychology shows that mood has a pervasive efect on cognitive processes, in particular on
udgment and memory. ! positive or negative mood can in"uence cognitive processing in two diferent
ways: it can facilitate processing of information whose content matches the mood#s valence$the
process our research investigates$or it can determine the processing strategy employed during tas%
performance &stimulus'oriented and local for sad mood, heuristic'(ased and glo(al for happy mood) for
reviews, see Clore and Huntsinger, 2007 and Martin and Clore, 2001*. In the present study, we use ++, to
investigate the impact of mood on discourse comprehension) in particular, how mood afects the
integration of valenced information and how mood afects integration processes that depend on
inferences. -o date, little is %nown a(out the in"uence of mood on language processing (eyond the
single word, and in particular on discourse processing. !s a result, the neural mechanisms (y which
mood afects linguistic processes are poorly understood.
.ur hypothesis is that mood can in"uence semantic processing in the context of discourse
comprehension, as mood constitutes a %ind of emotional framing, though not a linguistic context. /hile
theories of mood difer on the mechanisms that lin% mood to cognition, they agree in implicitly viewing
mood as a form of context that can drive information processing. Indeed, it is (ecause mood ofers such
a general context for all cognitive activity that its efect can (e so pervasive. 0ood#s in"uence on
cognition has in fact (een documented in several domains, including social udgment, persuasion,
memory, and attention &Clore and Huntsinger, 2007 and Martin and Clore, 2001*. .ur own prior wor% has
documented its efect in discourse comprehension &Egidi & Gerrig, 2009*. Recently mood has also (een
shown to in"uence event'related potentials at the word and sentence level &Chwilla et al., 2011, Federmeier et
al., 2001, Pratt and ell!, 200" and #issers et al., 2010*.
.ur investigation focuses on the 1233 efect, as in discourse comprehension it often re"ects the
process of semantic integration of a critical word with prior context &for reviews, see utas and Federmeier,
2011 and $au et al., 200"*. -he 1233 amplitude is larger for words that are less expected given prior
context, thus suggesting that the 1233 indexes the degree of di4culty involved in comprehension.
5iferent types of contextual factors can have direct (earing on the meaning construction of a
sentence. ! frst type of contexts are those that (elong to the linguistic communication experience.
-hese include contextual factors specifc to language or semantic %nowledge evo%ed (y the text, such
as general world %nowledge and prior narrative context &e.g., Hagoort et al., 200% and &an 'er(um et al., 200)*.
.ther contexts that originate within the linguistic exchange and can afect the 1233 are pragmatic
factors auxiliary to the linguistic stream proper. +xamples are co'speech gestures &e.g., He slips on the
roof and rolls down accompanied (y a gesture mimic%ing wal%ing6rolling down) Ha*ets, ita, +hao, ,-!.re(,
& Hagoort, 2010*, spea%er#s voice &e.g., child#s6adult#s voice uttering the sentence Every evening I drink
some wine before going to sleep) &an 'er(um, &an /en 'rin(, 0esin(, os, & Hagoort, 200"*, and emotional
'rain and $anguage
intonation &e.g., positive6negative words preceded (y a sentence uttered with congruent6incongruent
prosody) +1hirmer, ot-, & Friederi1i, 2002*. -here are also extralinguistic contexts that can afect the 1233,
such as a comprehender#s self'view or value system & &an 'er(um et al., 2009 and 3atson et al., 2007*. For
example, self'referential words that are inconsistent with one#s positive self'view elicit greater 1233
than consistent words &3atson et al., 2007*, and so does the word good &as compared to bad* in a sentence
li%e A society that condones abortion is a good society, when the statement clashes with the
comprehender#s values &&an 'er(um et al., 2009*.
/e (elieve that mood is a context of the latter type) it originates outside the linguistic communication,
is not language'specifc, does not convey explicit linguistic information, and in the case of the
comprehender#s mood examined here, is not conveyed in the linguistic exchange examined. Instead, it
constitutes a set of conditions on the receiver#s side, under which all information is processed. .ur
theoretical stance is that mood can in"uence the integration process in discourse comprehension (y
creating constraints on what would (e a ftting ending. 7onsistent with this hypothesis, mood has (een
shown to afect the 1233 in sentence comprehension when it is used to induce diferent processing
strategies &Chwilla et al., 2011 and Federmeier et al., 2001*. Federmeier et al. 420015 investigated the efects of
positive and neutral moods on semantic memory organi8ation (y using sentence pairs. -he study found
that positive mood, as compared to neutral mood, was associated with a decrease in 1233 amplitude
(etween unexpected items of diferent categories &e.g., They wanted the hotel look more like a tropical
resort. So along the driveway they planted rows of pines/tulips*. Chwilla et al. 420115 studied the efect of
happy and sad mood on high and low clo8e'pro(a(ility sentences &e.g., the pillows are stued with
feathers/books* and found that the 1233 efect was difused over (oth hemispheres for happy
participants, (ut limited to the right for sad participants. 9oth results are ta%en as evidence of mood'
dependent processing styles, in line with research in other cognitive domains. :appy mood &as
compared to sad or neutral mood* has (een shown to facilitate semantic processing and increase the
cognitive "exi(ility that leads to creative outcomes. ;ad mood, on the contrary, promotes a narrow
focus on external stimuli and analytic processing & Clore and Huntsinger, 2007, /a&is, 2009 and 6sen, 1999*.
7onsistently, the diferent scalp distri(ution of the 1233 found (y Chwilla et al. 420115has (een ta%en to
re"ect a more "uent integration process for the happy mood &see also 7t1hle!, 6lardi, & Enloe, 200) for
processing emotional words*. <nli%e these studies, the present experiment does not test the impact of
diferent moods as inducers of diferent processing strategies. Rather, it tests how moods afect the
integration of diferently valenced contents, ma%ing only minimal assumptions on processing styles
diferences (etween moods.
.ur experiment investigates two aspects of the possi(le efect of mood on discourse comprehension.
-he frst is whether mood facilitates the integration of textual information that matches the mood in
valence, although this information is as consistent with prior context as mismatching information. -he
second is whether mood has a more general efect on discourse integration) in particular, whether it
afects diferently discourse integration that involves drawing an inference as compared to discourse
integration that does not. ;pecifcally, we used ++, to examine whether listeners# moods afect how
they comprehend story endings, as measured (y the amplitude and latency of 1233 pea%s. -he moods
we study are mild happiness, mild sadness, and a neutral mood, all induced experimentally. 9y inducing
a neutral mood we refer to a situation in which (oth positive and negative mood valence were reduced
(y using a manipulation matched to that used for the induction of happy and sad moods, following a
well'%nown practice in studies of the efect of mood on cognition &e.g., Forgas, 2007 and H8n-e and Me!er,
199"*. .ur frst prediction was that mood (iases comprehension so that content congruent with one#s
mood is easier to process. -his would correspond to a mood congruence efect &'ower, 19"1* that is often
(ehaviorally re"ected in facilitated processing and (etter memory for mood'congruent information and
in the formulation of mood'congruent udgments and evaluations &Martin & Clore, 2001*. 0ood congruence
has largely (een shown when people experience mild happiness or sadness, and is typically defned (y
comparing the result patterns o(tained for participants experiencing these two moods &e.g., Fiedler et al.,
2001 and 9owe et al., 2007*. 9ecause of this, prior studies cannot always determine whether mood
congruence results from processing facilitation of congruent information or hindrance of mood
incongruent information. .ur study, however, included a neutral mood group that allowed addressing
this issue.
In our experiment, participants in a happy, sad, or neutral mood listened to stories that could have
either a positive or a negative ending &e.g., Trev passed the e!am, vs. Trev failed the e!am, as in the
frst story presented in -a(le 1*. If comprehension is afected (y mood, the 1233 would show a pattern
consistent with mood congruence: :appy participants would show a larger 1233 for negative endings
as compared to positive endings, and sad participants would show the opposite pattern, thus re"ecting
the contrast in understanding the story ending as incongruent, similar to typical 1233 results. -his
hypothesis fnds some support in a study (y 7hung et al. &1==>* that reports a similar pattern, though in
the context of a udgment tas%. -he experiment examined the efect of happy and sad moods on the
comprehension of a positive, negative or neutral word that concluded a very short story while
participants made two evaluations: 1. whether the story turned out as they expected and ?. whether
participants could identify with the story. -he 1233 was largest for inconsistent words, and larger for
consistent words that mismatched the valence of the mood. In our experiment, we chose to study
passive comprehension and did not as% participants to perform any udgment while they were listening
to the stories. .ur rationale was that, although some studies ta%e udgment and passive comprehension
as converging measures of language comprehension &e.g., Rapp and ,errig, ?33> and Rapp and
@endeou, ?33A*, others have shown that the two can yield diferent results and may therefore tap into
diferent cognitive processes & +gidi and ,errig, ?33> and +gidi and ,errig, ?33=*.
.ur investigation extends to also explore the possi(ility of negativity (ias in discourse comprehension
and its relation with mood congruence. It is %nown that people show greater sensitivity to negative &as
compared to positive* information in many domains of cognition &for a review, see 'aumeister, 'ratsla&s(!,
Fin(enauer, & #ohs, 2001*. 1egative stimuli are more informative than positive stimuli, are devoted greater
attention, are detected faster, demand more ela(orated and longer processing, and are remem(ered
(etter. -hey also elicit diferent +RB responses than positive stimuli &+mith et al., 200) and :lo;sson et al.,
200"*. In discourse comprehension, negative story endings have (een found to elicit longer reading
latencies &li%e inconsistent endings do*, a fnding which has (een ta%en as evidence of more careful
processing or greater di4culty of integration &Egidi & Gerrig, 2009*. In our study, negativity (ias would (e
re"ected in greater 1233 pea%s for negative endings compared to positive endings for participants in a
neutral mood. 9ecause of the expected negativity (ias, the diference in participants# responses to
positive and negative information may (e similar for the happy and the neutral group.
.ur aim was also to investigate the process that generates mood congruence in discourse
comprehension and thus also distinguish it from negativity (ias. /e therefore tested whether mood
congruence would show as easier processing of information congruent with the mood or more di4cult
processing of information incongruent with the mood, or (oth. +asier processing would show in a
reduced 1233 for mood congruent endings for happy and sad groups with respect to the neutral group.
7onversely, greater di4culty would show in a larger 1233 for mood incongruent endings for happy and
sad groups with respect to the neutral group.
/ith respect to the topography of the efects of mood and text valence, we expected that the efects
would (e more frontally distri(uted. 1ot only are auditory 1233 efects often shifted more frontally
&utas & Federmeier, 2011*, (ut processing emotional information has (een shown to modulate the 1233 at
frontal sites &e.g., /e Pas1alis et al., 2009 and ans(e and ot-, 2007*. -his prediction is also supported (y
previous research demonstrating involvement of frontal areas in processing afect in general
&e.g., /a&idson, 200)* and mood in particular &e.g., 'a(er, Frith, & /olan, 1997) Fit-gerald et al., 2011*. Regarding
lateral distri(ution, we considered that diferent moods alter the relative (alance of left and right
hemispheres contri(utions to language processing &Federmeier et al., 2001*. 9ased on the results o(tained
(y Chwilla et al. 420115, we therefore expected that the 1233 efect would (e widespread for the happy
mood, (ut would (e restricted to the right hemisphere for the sad mood.
Finally, we examined whether mood particularly afects language comprehension involving more
su(stantial processing. It has (een argued that mood more strongly in"uences processes reCuiring a
su(stantive amount of information ela(oration &e.g., 'ower and Forgas, 2001 and Forgas, 1992*. In the current
study, the valence of half of the story endings could (e understood directly, whereas the other half
could only (e fully understood as a result of integration with prior context. /e refer to the frst type
as sentence endings, and to the second type as conte!t endings. 7onsider the two stories in 0a*le 1. -he
sentence Trev passed the e!am immediately communicates the positive Cuality of the ending, without
having to refer this sentence to prior context. In contrast, the positive meaning of the sentence The
doors were open can only (e appreciated after assimilating it to the rest of the story and ma%ing the
inference that open doors mean that the protagonist could cash her chec%. /e examined whether mood
afects this general valence implication varia(le. /e considered that the fact that context endings are
fully understood only after integration with a greater portion of linguistic context may have two
conseCuences. First, compared to sentence endings, understanding context endings involves
integration of a greater amount of information, which entails greater taxing and greater di4culty on the
integration process. -his may translate into greater 1233 pea%s for these endings. -he second
conseCuence is a delay in comprehending the valence of these endings, which could translate into a
delayed 1233 pea%. In addition, to the extent that one type of ending valence or a mood can promote
or hinder the integration process, either of these efects could (e modulated (y the valence of the
endings or (y participants# mood.
/ith respect to the scalp distri(ution of these efects, we considered that moods could have again the
strongest efects frontally. In addition, (ecause right hemisphere processing has (een shown to (e
(iased towards information maintenance in wor%ing memory &utas & Federmeier, 2000* diferent efects
(etween moods on this varia(le could (e mostly distri(uted to the right.
?. 0ethod
?.1. Barticipants
+ighty'one undergraduates at the <niversity of 7hicago participated in this experiment for class credit
or payment. -hey were all right'handed +nglish native spea%ers. -hey were randomly assigned to one
of the three mood groups &happy, sad, neutral*. ;ix of these participants failed to follow the instructions
or recalled less than D3E of the story endings) as a conseCuence, their data were not used in the
analyses &see ;ection2.2.1 for more details a(out this*. -he analyses reported here were performed on
data from AD participants, ?D for each mood group.
?.?. 0aterials
/e used > video clips to induce moods and F? test stories. Gideo clips ranged from D to F min in length.
For the happy and sad mood inductions, we selected video clips that would (e efective in ma%ing
participants feel either cheerful or sad. For the neutral mood inductions we used clips that would
distract participants from any initial afective (ias they may have rather than elicit a specifc afective
response. /e created 1? clips, 2 for each mood induction condition, and used the following norming
procedure to identify the most efective clips. 1ine undergraduates at the <niversity of 7hicago
watched the clips in random order and flled out a Cuestionnaire after each clip (efore moving to the
next one. -hey rated whether the clip made them feel happy and whether it made them feel in a
generally positive afective state. -hey also rated whether the clip made them feel sad and whether it
made them feel in a generally negative afective state. /e used this two'Cuestion procedure to ensure
that the elicited mood was colored (y happiness or sadness &and not, for example, (y hope or anger*
and that it was in fact a mood &a generally positive or negative afective state*, rather than an intense
emotion. Barticipants used a scale that ranged from 1 &"ot at all* to = &E!tremely/#ery much*. 9ased on
their responses, we selected ? clips for each mood induction. 0a*le 2! shows the mean ratings for each
Cuestion for the three types of clips selected for the experiment and demonstrates how the selected
neutral, happy, and sad clips resulted in strongly diferent efects.
?.H. 5esign
!fter viewing the appropriate video clip to induce a happy, sad, or neutral mood, participants listened to
stories with positive or negative endings that could derive their valence from the sentence itself or prior
context. -hus, the design consisted of one (etween'participant varia(le: 0ood &:appy, ;ad, 1eutral*,
and two within'participant varia(les: +nding Galence &Bositive, 1egative*, and Galence Implication
&;entence, 7ontext*.
H. Results
H.1. 0ood induction
!t the end of the experiment, participants rated their happiness, positive mood, sadness, and negative
mood in four occasions during the experiment: at their arrival to the la( &T3*, after watching the frst
clip &T1*, after watching the second clip &T?*, and at the end of the experiment &TH*. 0a*le 29 shows the
mean ratings for each Cuestion. -o assess the success of the mood induction we performed two sets
of t'tests. /e frst compared participants# ratings at T3 with those given at T1, T?, and TH, to test
whether participants# mood was efectively modifed (y the presentation of the videos and whether this
modifcation lasted until the end of the experiment. 0a*le )! shows the results of these tests. -he second
set of t'tests assessed whether participants in the diferent groups were experiencing diferent moods
throughout the experiment. /e compared the ratings given (y the participants in the positive and the
sad groups with the ratings of the participants in the neutral group. -his analysis was necessary
(ecause, although the previous set of t'tests had already shown that the presentation of the video clips
polari8ed ratings, it did not guarantee that they were diferent across groups. 0a*le )9 shows the results
of these tests. .n average participants in the three groups arrived at the experiment with the same
medium level of happiness and good mood and low level of sadness and negative mood. Barticipants
who watched the neutral video clips maintained this pattern throughout the experiment. .nly after
watching the frst clip did their happiness and sadness decrease slightly. Barticipants who watched the
happy clips experienced an increase in happiness and good mood and a slight decrease in sadness and
negative mood after watching the clips, (ut not at the end of the experiment. Barticipants who watched
sad clips experienced a considera(le decrease in happiness and good mood and an even stronger
increase in sadness and negative mood after watching the two clips, (ut this efect did not last until the
end of the experiment. -he three groups were experiencing signifcantly diferent levels of happiness,
good mood, sadness, and negative mood throughout the experiment. !lthough at the end of the
experiment each mood group gave ratings that did not difer from those given at the (eginning, the
diference in ratings across groups was maintained, at least partially, until the end of the experiment.
-he positive group was still happier, in a (etter mood, and less sad than the neutral group &(ut did not
difer in negative mood*, whereas the sad mood group was still &and only* in a more negative mood.
2.D. 7onclusion
-he interest for the impact of mood on language comprehension processes has developed mostly
recently, and this is the frst study to show that incidental mood can afect the 1233 in discourse
comprehension. In addition, although the num(er of studies that approach the investigation of language
with auditory presentation is constantly increasing, this is the frst study to evaluate the impact of mood
on speech comprehension.
Recent +RB studies of language processing have shown that several types of contextual cues are ointly
integrated (y the language processing system, thus arguing for a (roader construal of what serves as a
context for discourse comprehension &Hagoort & &an 'er(um, 2007*. -hese studies ofer evidence against a
two'stage model of comprehension, where the compositional meaning of a sentence is frst constructed
and then integrated with contextual factors &Cutler & Cli;ton, 1999*. -he present study contri(utes to
strengthen a one'stage theoretical position &Hagoort, 200" and Hagoort and #an 'er(um, 2007* (y showing that
the construal of context for discourse processing should (e (roadened even further, to include mood or
emotional context that cannot (e descri(ed (y semantic or sym(olic linguistic properties. 0ost
important, (ecause mood percolates into the information integration process and changes how
discourse is understood, it must (e included in a comprehensive theory of language understanding.