0028 3932 93 S6.00+000 , 1993 Pergamon Press Lid

A LEFT VISUAL FIELD BIAS FOR SEMANTIC ENCODING OF UNATTENDED WORDS

ANTHONY LAMBERT and NEISHA VOLT

Department of Psychology, University of Auckland, New Zealand

(Receioed I Noremhrr 199 1; accepted 18 Auyust 1992)

Abstract-Previous studies have shown that verbal processing in both normal individuals and a split-brain patient can be strongly affected by the semantic category of an unattrndrd word presented to the left visual field (LVF). The effect was interpreted in terms of inhibition, since responses were slower when the unattended LVF word belonged to the same category as the target word. The present experiment discriminated between two alternative explanations for this finding. Subjects were presented with two letter strings, one in central vision and one to the left or right of centre. Subjects made a speeded lexical decision to the central string, and were instructed to ignore the lateral string. When a word was presented to the LVF subjects responded more slowly when it was semantically related to the central word. When an unattended word was presented to the RVF, its semantic relationship to the central word had no effect on decision latency. This finding is discussed in relation to views of performance laterality and selective attention.

INTRODUCTION

THE DIVIDED visual field technique, whereby information is presented briefly to the left or right side of visual space has been used extensively in studying human laterality. Lateral differences in performance have been interpreted in terms of the functional specialization of the left and right cerebral hemispheres. The general pattern of results is well known. Verbal tasks usually give rise to a right-sided performance advantage: subjects perform more rapidly and accurately when the information is presented in the right visual field (RVF). Conversely, tasks with a strong spatial component often give rise to a left-sided advantage: subjects perform more rapidly and accurately when the information is presented to the left visual field (LVF). A number of exceptions to this general pattern do occur. For example, the RVF advantage for processing verbal information can be abolished or reversed when the perceptual conditions are rendered difficult, or when another (secondary) verbal task is introduced [3, 51. An interesting exception to the verbal-RVF advantage recently reported by LAMBERT et ul. [6] is related to selective attention. Across several experiments it was observed that unattended verbal information presented to the left visual field produced strong indirect effects on the performance of a verbal task. In particular, the performance of a categorization task was influenced by the category of an unattended word presented to the LVF. Response latencies were slower when an unattended LVF word belonged to the same overall category (living or non-living) as an attended word presented in central vision. This was termed the “same category slower effect”, and it was interpreted in terms of an inhibitory effect of unattended information (cf. Ref. [13]). The effect also showed that verbal-semantic characteristics may be encoded with great facility from the left visual field-though,

67

68 A. LAMBEKT and N. VOOT

paradoxically, this was most apparent when the subject was trying not to encode the (unattended) lateral word. Indeed, a third and fourth experiment showed that the effects occurred independently of conscious awareness. In a further study [7] it was found that a split-brain patient (L.B.) showed the same pattern. L.B. categorized words presented to the RVF, and showed a same category slower effect from unattended words presented simultaneously to the LVF. This surprising result suggested that in the split-brain the left hemisphere has direct (albeit nonconscious) access to verbal-semantic information processed by the right hemisphere. Hence, an intact corpus callosum is not necessary for the effect of unattended semantic information reported by LAMBERT et al. [6]; it appears that these effects may be mediated via sub-cortical pathways.

The present experiment fulfilled two aims. The first of these was simply to establish whether the effect of unattended verbal information would generalize to a somewhat different task: namely the lexical decision task of MEYER and SCHVANEVELDT [9]. The second and more important aim was to resolve an ambiguity inherent in our earlier results. The same category slower effect could have arisen via two subtly distinct mechanisms. One possibility is that the effect occurs because words in the same category are more closely related in meaning than words in different categories. That is, the effect may be driven by semantic relatedness. Alternatively, the effect may be located at the stage of response production, and occurs because words in the same category produce the same response. That is, the effect may be driven by response category.

The lexical decision task devised by MEYER and SCHVANEVELDT [9] allowed us to tease apart these two explanations for our previous finding. In the experiment, subjects made lexical decisions to letter strings presented in central vision. Subjects were simultaneously presented with a second letter string to the left or right of centre. These two letter strings either belonged to the same response category (i.e. two words or two nonwords) or to different response categories (a word or a nonword). In addition, on trials where two words were presented the words were either related (e.g. dog-cat) or unrelated (e.g. dog-tin).

If the response category explanation is correct then response latencies should be slower on trials where the central letter string and an unattended letter string presented to the LVF belong to the same overall response category. According to the response category explanation this should occur even though pairs of nonwords have (by definition) no semantic relationship, and the pairs of words included equal numbers of related and unrelated words. A planned comparison assessing the effect of response category for unattended letter strings presented to the LVF was used to test this explanation.

If the semantic relatedness explanation is correct then response latencies should be slower when a centrally presented word is related, rather than unrelated to an unattended word presented to the LVF. According to the semantic category explanation this effect should occur even though both related and unrelated word pairs share the same response category. A planned comparison assessing the effect of semantic category for unattended words presented to the LVF was used to test this explanation.

METHOD

Subjects Thirty subjects were used of which 17 were female and 13 were male. All subjects were right-handed and their first

language was English. Participation in the experiment was voluntary.

LVF SEMANTIC ENCODING 69

Apparatus

The stimuli were produced on an Apple IIe computer, connected to a monitor equipped with fast decay phosphor. The computer generated the stimuli and timed the reactions. Responses were indicated on the keyboard of the computer using the letters N and V. Subjects were seated approx. 48 cm from the screen with their heads resting on a headrest. The room was well lit and quiet.

Stimuli

The stimuli were 144 words and 144 nonwords ranging in length from three to five letters. There were 72 word-associated pairs based on stimulus materials used by MCKDDN and RATCLIFF 181. Representative examples of

the word associates used are: JUMP-~LEAP, HEAD-FOOT, KIND-NICE, LEAF-TREE. In addition there were five practice pairs of words.

The nonwords were based on stimulus materials used by RUBENSTEIN et al. [12]. The nonwords were also three to five letters long and were orthographically and phonologically legal as well as pronounceable. None of the nonwords used was homophonic with a real word. There were 144 nonwords plus 10 practice nonwords. Representative examples of the nonwords are: JUND, PROT, SULT, HARN, CLOTE, SLINT.

Procedure

Subjects were instructed to fixate on the centre of the screen at all times. Each trial began with a central fixation cross presented for 1250 msec. Immediately after the cross a central letter string was presented for 150 msec, and a lateral letter string was presented for 33.3 msec. Onset of the two letter strings was simultaneous. The inner edge of the lateral letter string was 4” away from the centre of the screen. Character size of both central and lateral letters was 0.4 x 0.6”. Subjects were instructed to concentrate on responding to the central letter string, and to ignore any other information appearing on the screen. The subjects’ task was to decide whether the central letter string was a word or a nonword. Subjects responded by pressing the keys N or V using the left and right forefingers. Half of the subjects responded by pressing N for a word and V for a nonword, and the other half pressed V for a word and N for a nonword.

The subjects were told to respond as quickly as possible without making too many mistakes. Although the task was simple, subjects were told to guess if unsure, and that an incorrect response would be signified by a “beep” on the computer. Any responses which took longer than 5 set were not collected. After a response (or 5 set) the next trial commenced. There were 10 practice trials which were not recorded followed by four blocks of 36 experimental trials.

Design

There were 144 experimental trials. There were equal numbers of trials on which: the central letter string was a word or a nonword; the lateral letter string was a word or nonword; the lateral string was presented to the LVF or RVF. Thus, there were 18 trials with each combination of central string (word or nonword), lateral string (word or nonword), and lateral visual field (LVF or RVF). On the 36 trials where both the central string and the lateral string formed a word there were equal numbers of trials on which the lateral word was semantically related or unrelated to the central word, and on which the lateral word appeared in the RVF or LVF. Thus, within these 36 trials there were nine trials with each combination of lateral visual field (LVF or RVF) and semantic relatedness (related or unrelated). All experimental conditions varied randomly from trial to trial. There were eight different random trial sequences which were assigned randomly to subjects.

RESULTS

Response cutegory analysis

Latencies. To test the response category hypothesis median correct response latencies for each condition were entered into a three-way analysis of variance. The factors were: Response Category of the unattended letter string (same vs different); Visual Field of the unattended letter string (LVF vs RVF); and Central Letter String (word vs nonword). Response latencies together with accuracy data for these conditions are shown in Fig. 1.

There was a significant main effect of Central Letter String on response latency [F (1, 29) = 63.94, P < O.OOl]. Centrally presented words had shorter decision latencies than centrally presented nonwords. No other main effects or interactions approached significance in this analysis. A planned comparison of the effect of response category for LVF letter strings did not approach significance [F< 11.

Accuracy. The number of errors made in each condition was also examined, by an analysis of variance with the same factors as for the latency analysis. There was a marginally

70 A. LAMBEKT and N. VOOT

525 “l&.3’

Nonwords :: 500

-0

: (8.9) Different (7.8)

t; a a 425

(6.5) Same (8.0) A

4 Words (8.2). Different (6.9)

LVP RVF

Fig. 1. Mean reaction times (msec) as a function of visual field and response category for both words and nonwords presented centrally. Error rates (“56) are shown in parentheses.

significant main effect of Central Letter String [F (1, 29) = 3.06, P < 0. lo]. The error rate was slightly lower for words compared to nonwords: 7.4 vs 9.2%. The interaction between Response Category and Visual Field attained significance [F (1, 29) = 5.43, P < 0.05]. This showed that when a letter string was flashed to the RVF subjects made slightly more errors when it belonged to the same response category as the target: 9.6 vs 7.3%; whereas when a letter string was flashed to the LVF subjects made slightlyfewer errors when it belonged to the same response category as the target: 7.7 vs 8.5%. A complete breakdown of error rates is shown in Fig. 1.

Semantic relatedness analysis

Latencies. To test the semantic relatedness hypothesis median correct response latencies from trials in which both letter strings were words was entered into a two-way analysis of variance. The factors were: Semantic Relatedness of unattended word (related to central word vs unrelated) and Visual Field of unattended word (LVF vs RVF). Response latencies and accuracy data for these conditions are shown in Fig. 2.

(8.3)

(8.7)

400 1 LVF RVF

Visual field

Fig. 2. Mean reaction times (msec) as a function of visual field and semantic relatedness to the central word. Error rates (%a) are shown in parentheses.

LVF SEMANTIC ENCODING 71

The main effect of Visual Field of unattended word did not approach significance [F< I]. The main effect of Semantic Relatedness was also nonsignificant [F (1,29)= 2.52, MSe = 1399, ns.]. The interaction between them was also nonsignificant [P (1,29) = 2.34, MSe = 2076, P= 0.1371. However, a planned comparison revealed a significant effect of Semantic Relatedness for unattended words presented to the LVF [F (1,29)=4.38, MSe=2076, P<O.O5]. This effect is shown in Fig. 2. Subjects responded more slowly to a central word when it was related, rather than unrelated, to an unattended word presented to the left visual field. When an unattended word was presented to the right visual field, response speed was unaffected by semantic relatedness.

Accurac~j. The number of errors made was entered into an analysis of variance with the same factors as for the latency analysis. Both main effects and the interaction between them failed to approach significance (all P>O.20).

DISCUSSION

The results strongly suggest that semantic relatedness rather than assignment to the same response category was the factor responsible for the same category slower effect that we reported previously [6, 71. When an unattended word was presented to the left visual field subjects responded more slowly when it was related, rather than unrelated, to the central target word. In this experiment both related and unrelated words produced the same response. In contrast, response latency appeared unaffected by the response category of the peripheral letter string.

Though the effect of response category fell far short of significance, Fig. 1 shows a trend (albeit a small one) for latencies to be slower when a word was presented centrally, and the LVF letter string belonged to the same response category (i.e. was another word). This trend within the “word central” condition is consistent with an interpretation favouring semantic relatedness over response category as the critical factor here. Remember that the semantic category effect of LAMBERT et al. [6] was interpreted in terms of an inhibitory effect from unattended information (cf. Ref. [13]). In the present experiment, this inhibitory effect operates wirhin the condition in which a word is presented centrally and a word is presented to the LVF. This means that an inhibitory effect of semantic relatedness may contribute to an effect of response category, within the condition in which a word is presented centrally. When this contribution is factored out (by excluding those trials on which the central word is related to the LVF word), the contrast between same category and different response category trials became truly negligible (Word central, LVF same response category: 410 msec; Word central, LVF different response category: 412 msec).

While response latency was the central measure in this study, analysis of error rates revealed an interaction between response category and visual field. When unattended letter strings were presented to the LVF accuracy was slightly better when the letter string belonged to the same response category. The reverse was true for letter strings presented to the RVF. This was an unexpected result and is not predicted by either the semantic relatedness or the response category explanation for our earlier finding. In Experiments 1 and 2 of LAMBERT et al. [6] when distracters were presented to the LVF accuracy was lower

for related/same category words (21.4% error vs 1.8% and 13.8% error vs 6.3%, respectively). Overall, it can be concluded that neither the error data nor the latency data provide any support for the response category explanation of LAMBERT et al.‘s [6] “same category slower effect”.

12 A. LAMBERT and N. VOOT

At a general level, the results are consistent with a classic late selection view of attention in which all inputs, both selected and unselected are encoded to a complex semantic level. In the present experiment, and in our earlier reports unattended information produced an inhibitory effect related to the semantic attributes of the distractor word (cf. Ref. [13]). However, this effect has been confined largely to trials where the distractor is presented to the LVF. Since RVF words were also unattended why was an effect not found for these distracters also? The answer is provided by Experiments 3 and 4 of LAMBERT et al. [6], which showed that spatial attentional bias is a crucial factor mediating the effect of unattended information. In Phase 1 of these experiments, subjects were required to attend and respond to centrally presented target words, while ignoring lateral distracters. In Phase 2 attentional bias was assessed by presenting a lateralized version of the word categorization task. Subjects with a RVF advantage in the lateralized task showed an inhibitory effect from LVF distracters together with a facilitatory effect from RVF distracters, when attending centrally in Phase 1. Subjects with a LVF advantage showed precisely the reverse pattern. The importance of attentional bias as a factor controlling the effect of unattended information was confirmed in Experiment 4, in which distracters were presented above and below fixation rather than to left and right, and in a further study by FRANCIS [4], which has replicated the findings of Experiment 3.

This pattern of results was interpreted as consistent with a dynamic view of selective attention, in which there is activation of semantic representations for some unattended objects-those appearing in locations towards which there is a bias or predisposition to shift attention; and inhibition of semantic representations for other objects-those appearing at locations for which there is no bias or predisposition for an attention shift. Lateralized versions of the lexical decision task employed in the present experiment have almost always produced a RVF performance advantage (see BEAUMONT [2] for review). Hence, our central prediction was that an inhibitory effect would be found for LVF distracters. Although both facilitation from biased locations and inhibition from unbiased locations have been observed 161, the latter effect on its own has been the more common pattern in our experiments [4,6, 71, and this is borne out in the present results.

A corollary to this interpretation of our findings is that the present results also support the conclusion that selective attention is an important factor mediating visual field asymmetry in processing verbal information. The controversy between attentional and structural explanations for performance asymmetry in divided visual field tasks is a long standing issue in laterality research, and it seems likely that both structural and attentional factors play a role. The present results together with our earlier findings show that even though a clear RVF advantage is typically found for lateralized verbal tasks such as lexical decision, individuals are nevertheless highly sensitive to unattended verbal&semantic information presented to the left visual field. As others have noted (e.g. see ALLPORT [I]) indirect measures, such as interference or priming, may produce a very different picture of information processing to that provided by direct measures, such as categorization or identification. The present results show that verbal-semantic information can be encoded with great facility from unattended information presented to the LVF. If neural structure (i.e. the more direct connections linking each visual field with the contralateral hemisphere) were the sole factor responsible for visual field asymmetry then the pattern of results would be unaffected by manipulation of selective attention. Further support for the view that selective attention is an important factor mediating visual asymmetry comes from a recent study by MONDOR and BRYDEN [lo]. These authors found that RVF advantages observed for letter identification and lexical

LVF SEMANTIC ENCODING 73

decision were eliminated when a peripheral cue provided advance information concerning the location of a forthcoming stimulus.

One final point concerns the split-brain study described earlier. When categorizing words presented to the RVF, the split-brain patient L.B. showed an inhibitory effect from distracters presented to the LVF. This shows that interaction between attended and unattended objects presented at different spatial locations may be mediated subcortically, presumably involving cortico-collicular interaction. The effect observed here may have occurred via similar pathways. However, since our subjects were all neurologically intact, purely cortical mediation of the effect is also possible. An interesting possibility that we are currently studying is that the effect reported here reflects an unconscious process whereby new unattended visual events are monitored, in order to check for significance, informativeness, or perhaps danger value. In the event of a positive outcome an attention movement may be triggered, bringing the new object into focal attention, enabling further evaluation and, if necessary, responding. Collicular involvement would certainly be expected here, in view of the evidence linking the superior colliculus with attention movements in both humans and monkeys [I 1, 141.

REFERENCES I. ALLPORT. A. Visual attention. In Foundation9 of‘ Cognitive Science, M. P~SNER (Editor), pp. 631-682. MIT

Press, Cambridge. MA, 1989. 2. BEAUMOX;T, J. G. Divided Visual Fir/d Studies of Cerebral Orgmrsation. Academic Press. New York, 1982. 3. BRALXHAW, J. L., HICKS. R. E. and ROSE. B. Lexical discrimination and letter string identification in the two

visual fields. Brctin Lang. 8, 10-18, 1979. 4. FRAXCIS, M. Unpublished experiments, University of Auckland. 5. HELLIC;E, J.. Cox. P. and LITVAC, L. Information processing in the cerebral hemispheres: selective hemispheric

activation and capacity limitations. J. r.up. PsJ&I~.: Gen. 108, 251~ 279. 1979. 6. LAMHERT. A. J., BEARD, C. T. and THOMPSON. R. J. Selective attention, visual laterality, awareness and

perceiving the meaning of parafoveally presented words. Q. J. e\-p. Pspchol. 40A, 615-652, 1988. 7. LAMBFRT. A. J. Inter-hemispheric interaction in the split-brain. Nruropsychologia 29, 941-948, 1991. 8. McKoo~, G. and RATCLIFF. R. Priming in episodic and semantic memory. J. rerh. Lrarn. wrh. Beh~. 18,

463-480, 1979. 9. MFYER, D. and S~HVANEVELDT, R. Facilitation in recognising pairs ofwords: evidence ofa dependence between

retrieval operations. J. e\-p. Ps)chol. 90, 227-234, 1971. IO. MONDOR, T. A. and BRYDEV, M. P. On the relation between visual spatial attention and visual field

asymmetries. Q. J. rup. Psycho/. 44A, 529-556, 1992. I I. POSNER, M. I.. COHEN, Y. and RAFAL. R. Neural systems control of spatial orienting. Pl~il. Trans. Roy. SW.

Land. B 298, I87 198. 1982. 12. RLIHENSTEIN. H.. LEWIS. S. and RIXENSTEIN, M. Evidence for phonemic recoding in visual word recognition. J.

rerh. Learn rwh. Behur. IO, 645-657, 1971. 13. TIPpER. S. P. The negative priming effect: inhibitory priming by ignored objects. Q, J. exp. Psycho/. 37A,

571 590, 1985. 14. W~R~Z, R. H. and MOHLER, C. W. Organisation of monkey superior colliculus: enhanced visual response of

superficial layer cells. J. lVeuroph~~sio/. 39, 745-765. 1976.