A behavioral and electrophysiological investigation of masked semantic

A behavioral and electrophysiological investigation of masked semantic priming Giordana Grossi State University of New York at New Paltz The First Russian Conference of Cognitive Science Kazan, October 9 -12, 2004

Background n Conscious perception is associated with attention n Data from both normal and brain-lesioned individuals have shown that certain aspects of visual stimuli can be accessed pre-attentively (i. e. , in the absence of awareness) – These aspects include not only visual features, but also more abstract representations, such as the meaning of words

Semantic priming paradigm n People are faster at processing words or objects (targets) if these stimuli are preceded by related as compared to unrelated items (primes) – E. g. , Meyer & Schvaneveldt, 1971 Unmasked time * tiger Prime (500 ms) LION SOA (500 ms) • Lexical decision task (LDT): Is the target a real word or not? Target (500 ms)

Priming without awareness n This facilitation can occur in the absence of overt identification of the prime – E. g. , Marcel, 1983; Humphreys et al. , 1991; Ferrand & Grainger, 1992, 1994; Dehaene et al. , 1998 Masked time ####### tiger Prime (67 ms) LION Target (500 ms) SOA (67 ms) • Participants are not aware of the presence or identity of the prime in the masked condition

Semantic priming effect 600 Response time (ms) 580 560 540 * * Related words (tiger-LION) Unrelated words (table-LION) 520 500 480 460 With awareness Priming effect Without awareness

Automatic Spreading Activation (ASA) theories Semantic network (Collins & Loftus, 1975): words in the lexicon (LTM) are connected by associative relationships Priming effect: - The prime activates associate nodes - Pre-activated (primed) nodes are processed faster than non-primed nodes (forward process)

Unmasked and masked priming effects might reflect different mechanisms n Unmasked priming effects: mix of automatic and controlled processes n – ASA (automatic) – Strategies (controlled) – Post-lexical semantic matching (controlled) n Masked priming effects – ASA Masked priming effects are considered a reflex of how our mental lexicon is organized

Event-Related Potentials (ERPs)

ERPs are sensitive to word meaning parietal tiger - LION vs. table - LION target word onset Grossi & Neville, 2000

Experiment 1: Methods n Participants: 32 right-handed participants from the University of Oregon n Stimuli: Four prime conditions – – n Identical Semantic Unrelated Neutral sun-SUN moon-SUN tale-SUN XXXX-SUN Questionnaire on prime awareness – Participants who could read the primes in the masked condition were excluded from the analyses

Sequence of events Unmasked Priming Masked Priming PRIME MASK 1000 tiger 1000 ####### PRIME 500 TARGET LION 67 tiger TARGET 67 LION 500 time (msec)

ERP Methods n Event-Related Potentials were recorded from 29 electrodes in an extended International 10 -20 System montage n On-line recordings were referenced to the right mastoid and rereferenced to averaged mastoids in the final data averaging

ms Experiment 1: RTs * *** p<0. 0001; * p<0. 05

N 400 effect (300 -500 ms)

Early fronto-temporal effect (starting at 160 ms) and small N 400 effect (400 -500 ms)

Contextual effects: Relatedness Proportion (RP) n ASA theories: masked priming effects index word recognition processes that are automatic and contextinsensitive – The prime does not provide an episodic representation that can help us process the target: masked and unmasked priming effects reflect different processes n Masked priming effects should not be affected by factors that generally affect unmasked priming – Relatedness Proportion (RP): proportion of related pairs in a list (e. g. , 80% vs. 20%) – Unmasked priming (long SOAs): larger priming effects with high RP (e. g. , Neely, 1991)

Alternative explanations of masked priming effect: retrospective theories n Retrospective theories of priming (e. g. , Whittlesea & Jacoby, 1990) – The prime creates an episodic representation (which remains unconscious) that can be recruited to assist in the encoding of the target: this recruitment is under contextual control (few preceding trials) n RP can affect masked priming – E. g. , Bodner & Masson, 1997, 2001: larger priming effects with high RP n No changes in LTM: masked priming effects may not say much about how words are organized in the lexicon

ASA and retrospective theories make different predictions n ASA theories: RP does not affect masked semantic priming n Bodner & Masson (1997, 2001, 2003): RP affects masked semantic priming – Larger priming effects with high RP

Experiment 2 a: Behavioral experiment Methods n n Participants: 40 participants from SUNY at New Paltz Procedure (Bodner & Masson, 2003) – Two conditions (between-Ss) time ####### tiger Prime (50 ms) LION Target (500 ms) SOA (50 ms) • High RP: 80% (n=20) • Low RP: 20% (n=20) n Prime awareness experiment (LDT on the primes)

Reaction times * ** ** p<0. 01; * p<0. 05 Priming, F(1, 38)=19. 132, p<0. 0001) n No interaction between priming and RP (p=0. 7): no evidence of RP effects at short SOA (50 ms) and without awareness of the prime n

% “word” responses Prime awareness n Participants were not able to discriminate between word and nonword primes – No differences between Hits and False Alarms (FA) rates (p>0. 2) – Participants were not aware of the identity of the prime

Experiment 2 b: ERP experiment Methods n Participants: 20 participants from SUNY at New Paltz n Procedure: – GO-NO GO categorization task: participants were required to press a button if the target was a proper name (e. g. , Mary, James) • Measure of the N 400 effect without the contamination of response-related potentials – Order: Ss participated in both RP conditions (order was counterbalanced across Ss)

Behavioral results RT ACC % High RP 553. 61 (65. 34) 97. 3 (2. 85) Low RP 551. 74 (63. 71) 95. 7 (5. 44) n n Proper names (n=20) –No significant effects (all p’s > 0. 1) Prime awareness (n=19) – Mean % Hits=60. 21 (17. 31) – Mean % FA=62. 32 (17. 5) n Participants were not able to discriminate between word and nonword primes (p > 0. 4)

N 400 effect (300 -500 ms)

N 400 effect (300 -500 ms)

Both N 400 and early frontal effect

No N 400 effect and no early frontal effect

n=10 Both N 400 and early frontal effect

n=10 Both N 400 and early frontal effect

n=10 Reduced and later N 400 and no early frontal effect

n=10 No N 400 effect and no early frontal effect

Discussion and final considerations n No clear support for retrospective theories of priming, at least as formulated by Bodner & Masson – No behavioral RP effects (see also Rosa & Perea, 2004) – No robust and clear-cut ERP effects, although more data are needed n The N 400 effect in masked conditions might reflect ASA mechanisms (lexical organization). Although, contextual factors seem to play a role – The N 400 effect was present, absent, or reduced depending on what block was presented first (e. g. , the N 400 effect was present in the high RP condition only when this condition was presented first) – Do we need to rethink automaticity (e. g. , Posner & Snyder, 1975)?

Many thanks to my students and collaborators n SUNY at New Paltz – – – – Joshua Boggan Patricia Broome Teressa Del Santo Joanna Doerfer Jennifer Earle Amertah Perman Stella Quinn Jamie Slonim – Aram Agajanian, Dariann Zielinski, Bruce Blocke n University of Oregon – Helen Neville – Brain Development Lab crew

Stimuli Identical primes Semantic primes Unrelated primes Neutral primes Word moon cat pen XXXXX targets MOON DOG DOG N=128 Nonword targets N=128 Word primes peel GLIK Neutral primes XXXXX LABE NOTES Prime-target pairs were rotated across the priming condition across four groups of subjects and four lists of stimuli, such that no subject saw any single prime or target more than once, but each subject received all four experimental conditions. Identical, semantic and unrelated primes were matched on frequency and length, so were targets across the four lists

Behavioral results * ** Unmasked Masked Semantic priming was found in both unmasked and masked conditions

RP*Priming*Hem*Order, (F(1, 18)=3. 98, p=0. 06). n Separate ANOVAs for each order (high RP or low RP first): the priming effect was larger when Ss were exposed to the high RP block first. When the first block was high RP, the priming effect was significant (p<0. 03) and even larger for the low RP block, which was presented second, whereas it was not significant when the first block was the low RP. n – N 400 effect was larger in the low RP block when the block was seen as second. Moreover, the difference between high and low RP block was not that big, but, if any, depended on the order of the blocks: significant differences were obtained only when the high RP block was received first.

Masked 160 -200 ms 200 -300 ms 300 -500 ms Unmasked

Questions n What is the nature of masked semantic priming effects? – Is priming with and without awareness mediated by the same cognitive processes? – Are identical or non-identical neural systems involved in semantic priming with and without awareness?

Previous ERP research Masked priming experiments n Pratarelli et al. , 1989; SOA = 33 msec – No N 400 effect BUT earlier effect starting around 200 msec n Brown & Hagoort, 1993; SOA=740 ms – No N 400 effect n Mc. Carthy & Nobre, 1993 – No N 400 effect in the unattended visual field n Kiefer, 2002; SOA=67 ms – N 400 effect, no early effects (prime was followed by mask)

The role of the SOA Effects different from the typical N 400 effect have been found in implicit semantic ERP tasks (Pratarelli et al. , 1989) n Except for Pratarelli et al. (1989), masked semantic priming studies have employed long SOAs (e. g. , 500 ms) n Greenwald et al. (1996): activation of semantic information that rapidly decays (~200 ms) n n Hypothesis: early effects reflect the activation of highly refractory lexical systems

Unmasked and masked priming in retrospective theories n Priming effects are explained in terms of influence of memory that is – Conscious in unmasked priming • Episodic and strategic processes – Unconscious in masked priming • Episodic n Similar cognitive processes are involved in the two priming conditions – The same factors can affect both conditions

Interim discussion n Evidence of behavioral and ERP priming effects for both unmasked and masked condition n Masked condition – N 400 effect (400 -500 ms): considered to be an index of ASA in the lexicon (Kiefer, 2002) – Early fronto-temporal effect (~160 ms) • Generally observed at short SOAs (Pratarelli et al. , 1989; Misra & Holcomb, 2003)

Interim summary N 400 effect (300 -500 ms): F(1, 18)=6. 81, p<0. 02 n Trend for early fronto-temporal effect (220 -320 ms): priming, F(1, 18)=3. 6, p<0. 08 n No interaction between priming and RP for either effect n n Order effects: The priming effect was larger when Ss were exposed to the high RP block first – RP*Priming*Hem*Order (F(1, 18)=3. 98, p=0. 06) • High RP first: N 400 effect was significant (p<0. 03) • Low RP first: N 400 effect was not significant

Interim discussion The N 400 effect was present in both RP conditions n Presence of early frontal effects, although weak n Lack of robust RP effects on N 400 and early effect n – These effects were somewhat smaller in the low than high RP blocks, but this difference was due to the order of presentation of the two blocks

Between-subjects analyses: comparisons of first blocks (high vs. low RP) n The N 400 effect was – Significant as main effect (F(1, 18)=4. 45, p<0. 05) – Present in both high and low RP blocks, and larger over the posterior sites – More robust in the high RP condition; it also started later in the low RP – Not affected by RP (Priming*RP, p=0. 58) n ERPs at anterior sites (220 -320 ms) were – Larger to unrelated than related trials (F(1, 18)=3. 83, p<0. 07) – Although the interaction between priming and RP was not significant, priming was significant in the high RP condition (F(1, 9)=6. 83, p<0. 03), but not in the low RP condition