Perceptual and Motor Skills, 1979, 48, 935-942. @ Perceptual and Motor Skills 1979
W O R D CONTEXT EFFECTS O N LETTER RECOGNITION1 MARION M. JACEWICZ University of Tulsa
Stmzmasy.-24 subjects were asked to determine whether a target lener was present in a tachistoscopically exposed word. T h e presence of the target was detected faster if the cargec was clearly sounded in the word, that is, the target G in the word TTGER mas detected fascer than the target G in the word RTGHT. This effect was stiongcsr among subjects whose over-all response time was below the median value T h ~ sresult was interpreted as evidence-that (1) a word is recognized as a whole before ics component parts are analyzed, ( 2 ) the transformation of the word from visual to acoustic code occurs before the analysis of the individual letters, and ( 3 ) individual differences in efficiency of target detection depend on a subject's over-all response time.
There exists compelling recent evidence that even under highly divergent experimental conditions a word-like stimulus is recognized as a unit before its components parts are recognized. Johnson (1975) observed that subjects could identify a target word faster than they could identify a target letter within a word. Travers ( 1973) showed that backward masking or sequential oneposition letter presentation affected the accuracy with which subjects identified a word but did not affect the identification of random letter strings. Although exceptions are noted in the literature (McClelland, 1976), the bulk of these studies indicate that with word-like stimuli, the whole word may be the salient perceptual unit even when the subject's task does not necessarily require apprehension of a word. This repeated demonstration of the advantage of words over random letter strings has been explained in terms of a "unitization hypothesis." The word is recognized first, as a higher order unit, and closer analysis of its lower level components takes place secondarily, and then only if it is required by the task. Since words are clearly processed differently from random letter strings, the major problem becomes understanding the source of this difference. On the one hand, some recent researchers have focused on the visual-spatial nature of word recognition. Pollatsek, Well, and Schindler (1975) following a technique developed by Eichelman (1970) required subjects to judge whether tachistoscopically presented word pairs were same or different. They showed chat an advantage for words appeared when a judgment of difference could be based only on a physical mismatch (for example, WORD,W O R ~was judged to be different faster than OR^, O R W ~ ) .Further, they found no significant re'This research was carried out at Florida International University in Miami, Florida. Special thanks go to Patti Fisher and Gregory Maguire for assistance in data collection. Parts of this study have been presented at the Southeastern Psychological Association Meeting. New Orleans. March. 1976. Requests for reprints should be sent to Marion M. Jacewicz, Psychology Program. University of Tulsa. 600 South College. Tulsa, Oklahoma 74104.
936
M. M. JACEWICZ
duction in the advantage of words over non-words when the word pairs were homophones than when they were non-homophones. Both findings indicate the decision concerning the match between word pairs probably proceeds through a visual rather than verbal comparison. Similarly, Baron and Thurstone (1973, Experiment 6) reported results inviting a similar conclusion from a study designed to show whether or not pronounceability was a source of the word advantage. Two sets of stimuli were used. The homophonic set included word pairs which sounded alike but were spelled differently, e.g., FORE, FOUR. The set of pronounceable controls included word pairs (SORE, SOUR) which sounded differently and were spelled differently. Subjects had to determine which of the words in the pair was presented. One would expect if word-sound information was important in word recognition, subjects would be more accurate in recognizing words from the set of pronounceable controls than in the set of homophones. In fact, there was no difference in subjects' accuracy between the two conditions. These findings minimize the importance of the phonetic properties of words in tachistoscopic word-recognition process. On the other hand, other evidence allows the conclusion that word recognition must involve an early phonetic component. Mezrich (1973) showed that the advantage in recognizing letters within words over recognizing single letters was actually reversed when the subject was required to vocalize the recognized word or letter. Hawkins, Reicher, Rogers, and Peterson (1976) required subjects to determine which of two alternative words had appeared. When the subject was set to rely on phonetic information in making this discrimination, performance was facilitated when a phonetic distinction between the alternatives was present, e.g., SOLD,COLD, and deteriorated when a phonetic cue was not available, e.g., SENT, CENT Krueger (1975) provided indirect evidence that verbal processes may underlie the superiority of a word. He showed that subjects could determine whether a target letter appeared in a word faster and with fewer errors than they could determine the presence of the target in a non-word only when the word or non-word was presented in the right visual field. Recent models of word and letter perception (Smith & Spoehr, 1974) allow For both spatial-visual and phonetic categorization but at different phases of the recognition process. The translation from a visual to a phonetic version seems to occur relatively late in the recognition process (e.g., Thorson, Hochhaus, & Stanners, 1976) and some theorists (Smith & Spoehr, 1974, p. 263) suggest it may not even be necessary in many tasks. The present experiment was designed to provide ( 1 ) a direct test of the unitization hypothesis that the word is the most compelling perceptual unit with visually. presented word stimuli and ( 2 ) evidence that the word unit is represented phonetically quite early in the recognition process even when this representation is not necessary to the task at hand. If a word is the salient
WORD
CONTEXT
IN
LETTER
RECOGNITION
937
perceptual unit with tachistoscopically presented word stimuli and if that word is available in a phonetic form, the subject should be able to recognize a target letter within that word more easily if the target letter is sounded in the word than when the target letter is unsounded. That is, it should be faster to find the letter G within the word TIGER than within the word RIGHT.
METHOD Twenty-four Florida International University undergraduate students (13 men and 11 women) participated in the study as a course requirement. The subjects were between the ages of 18 and 44 yr., with an average age of 27. Six of the 24 were bilingual in Spanish and English. Four of the 24 were left-handed. All had vision corrected to normal. Each subject participated in a single experimental session which lasted approximately half an hour. The stimuli were 60 five-letter words, each on a single card. These words along with cheir frequencies per million words (Kuqera & Francis, 1967) are listed in Table 1. Half the words had at least one letter which is not sounded at all or which is given "unusual" pronuncistion. These letters are the potential target letters. The words were chosen such that the potential target letters occurred equally ofcen in each of the five positions within the word. The other half of the stimulus words were chosen so that the potential target letter was clearly pronounced within the word. Any target letter occurred only once within the word. The design was a 2 x 2 x 5 factorial, with three repeated measures. Half of the 60 stimulus words in the stimulus set were randomly designated as containing a target letter for any one subject (target-present condition). The other half did not contain target letters (target-absent condition). It is important to realize that any one stimulus word might serve in the target-present condition for one subject and in the target-absent condition for another subject. Half the set of target-present words had the target clearly pronounceable in the word, e.g., target G in the word 1TGER. In the other half of the target-present words, the target was either not pronounced (target G in RIGHT) or given "unusual" pronunciation because of spelling rules (target I in JUICE). TABLE 1 POSITION OF
TARGET IN
WORDS IN WHICHTARGET L
2
1
E ~ IS R PRONOUNCED OR NOT
Postion of Target in Word 3
4
5
Target Letter Clearly PronouncedC
36 442
KINDS WATER 104 EIGHT 224 WOMAN 130 WORKS 7 KINGS
10 SWEAR 938 WHERE 3 SCARE 4 PETAL 95 FUNDS 67 ASIDE
7 TIGER 170 TRIED 22 FAULT 58 SMILE 34 CRIME 148 STAND
5
FLAGS 7 BARGE 48 LIMIT 3 STRUT N O SHALE 28 POUND
Target Letter Not Pronounced or Given "Unusual" Pronunciation* 13 DOUGH 76 KNIFE 7 SWORD 613 RIGHT 22 RHYME 11 JUICE 7 COUGH 106 WRITE 15 CRUDE 16 TRAINS 59 EXIST 106 SCENB 129 WRONG
9 -'I11 1 GNARL
173 HEART 76 QUIET 4 ISLES
10 GRIEF
52 RAISE 110 HEAVY
8 FRAUD 2714 WOULD 332Ll~HT
134 MAYBE 11 POSSE 3 PIZZA 20 PLAT0 1 POLKA NO ZIPPO
13 BLADE 41 ROUGH 2 COCOA 4 WEIGH 14 ARROW ~~OEARTH
*Numbers are the number of, times the word appears per million words (Kuqera & Francis, 1967)
.
938
M. M. JACEWICZ
The target letter for the target-absent word set was randomly determined, with the obvious constraint that it not appear in the word. All the letters of the alphabet were potential target letters. The stimulus set was created by applying black upper case transfer lettering (Chartpak Futura demi-bold 24 Pt.) to white cards. The word was centered around a central fixation point, with approximately 12 mm between the centers of each letter. The distance between the viewers' eye and the stimulus card was 85 cm. Subjects saw the unmasked word stimulus for 50 msec. The subject was seated at a Scientific Prototype two-channel tachistoscope and was told that the purpose of the study was to determine the speed with which they could recognize letters within words. The subjecc was given verbally a "target" letter and asked to indicate if that target letter was present or not in the word stimulus. A "yes" response meant the target was present; a "no" response indicated the target was not present. The stimulus presentation was always initiated by the subject at his or her own pace. Instructions to the subject stressed the importance of fixating upon a small black point and generally preparing oneself before initiating the stimulus. The dependent measure was the interval between the stimulus presentation and the voice response of "yes" or "no." Subjects wore a vibration-sensitive collar around their necks. The voice response rime was recorded with a Lafayette voice response control (Model 6602-A) and clock counter (Model 54519). On each trial ( 1 ) the experimenter informed the subject of the current target letter, ( 2 ) the experimenter reset the apparatus and gave the subject a ready signal, ( 3 ) the subject initiated the stimulus and ( 4 ) responded "yes" or "no." Each trial took about 20 sec. Each subject received numerous practice trials to ensure stable performance before response times were recorded. There were 60 recorded trials in each session. Although a 12% error rate was allowed, the average error rate was 4%. Only two of the 24 subjects exceeded an 8% error rate.
RESULTS Mean correct response times (RT) for each subject to detect the presence of a target letter within the stimulus word are recorded in Table 2. RTs for the condition in which the target letter was clearly "sounded" in the word and for the condition in which the target was "unsounded" are separaced in the rable. The difference between the two conditions for each subject is also noted. Data from subjects with average response times above and below the median value are presented separately. An analysis of variance was performed on the response times when che target letter was present in the word and the subjecc correctly detected its presence. The variables considered in the analysis were (1) whether or not the target letter was "sounded" in the stimulus word, ( 2 ) the position of the target in the stimulus word, and ( 3 ) subjects. All main effects reached significance. Targets which were sounded in the stimulus word yielded fascer response times than did targets which were not sounded ( M = 553.5, S D = 1 0 4 . 9 a n d M = 580.3, S D = 1 5 1 . 6 msec., respectively; F1,23= 7.03, MSE = 80.52, p .025). The time co find che target lecter in the word was fastest when the target was in the far left position and slowest at the far right position
W O R D CONTEXT EFFECTS O N LETTER RECOGNITION1 MARION M. JACEWICZ University of Tulsa
Stmzmasy.-24 subjects were asked to determine whether a target lener was present in a tachistoscopically exposed word. T h e presence of the target was detected faster if the cargec was clearly sounded in the word, that is, the target G in the word TTGER mas detected fascer than the target G in the word RTGHT. This effect was stiongcsr among subjects whose over-all response time was below the median value T h ~ sresult was interpreted as evidence-that (1) a word is recognized as a whole before ics component parts are analyzed, ( 2 ) the transformation of the word from visual to acoustic code occurs before the analysis of the individual letters, and ( 3 ) individual differences in efficiency of target detection depend on a subject's over-all response time.
There exists compelling recent evidence that even under highly divergent experimental conditions a word-like stimulus is recognized as a unit before its components parts are recognized. Johnson (1975) observed that subjects could identify a target word faster than they could identify a target letter within a word. Travers ( 1973) showed that backward masking or sequential oneposition letter presentation affected the accuracy with which subjects identified a word but did not affect the identification of random letter strings. Although exceptions are noted in the literature (McClelland, 1976), the bulk of these studies indicate that with word-like stimuli, the whole word may be the salient perceptual unit even when the subject's task does not necessarily require apprehension of a word. This repeated demonstration of the advantage of words over random letter strings has been explained in terms of a "unitization hypothesis." The word is recognized first, as a higher order unit, and closer analysis of its lower level components takes place secondarily, and then only if it is required by the task. Since words are clearly processed differently from random letter strings, the major problem becomes understanding the source of this difference. On the one hand, some recent researchers have focused on the visual-spatial nature of word recognition. Pollatsek, Well, and Schindler (1975) following a technique developed by Eichelman (1970) required subjects to judge whether tachistoscopically presented word pairs were same or different. They showed chat an advantage for words appeared when a judgment of difference could be based only on a physical mismatch (for example, WORD,W O R ~was judged to be different faster than OR^, O R W ~ ) .Further, they found no significant re'This research was carried out at Florida International University in Miami, Florida. Special thanks go to Patti Fisher and Gregory Maguire for assistance in data collection. Parts of this study have been presented at the Southeastern Psychological Association Meeting. New Orleans. March. 1976. Requests for reprints should be sent to Marion M. Jacewicz, Psychology Program. University of Tulsa. 600 South College. Tulsa, Oklahoma 74104.
936
M. M. JACEWICZ
duction in the advantage of words over non-words when the word pairs were homophones than when they were non-homophones. Both findings indicate the decision concerning the match between word pairs probably proceeds through a visual rather than verbal comparison. Similarly, Baron and Thurstone (1973, Experiment 6) reported results inviting a similar conclusion from a study designed to show whether or not pronounceability was a source of the word advantage. Two sets of stimuli were used. The homophonic set included word pairs which sounded alike but were spelled differently, e.g., FORE, FOUR. The set of pronounceable controls included word pairs (SORE, SOUR) which sounded differently and were spelled differently. Subjects had to determine which of the words in the pair was presented. One would expect if word-sound information was important in word recognition, subjects would be more accurate in recognizing words from the set of pronounceable controls than in the set of homophones. In fact, there was no difference in subjects' accuracy between the two conditions. These findings minimize the importance of the phonetic properties of words in tachistoscopic word-recognition process. On the other hand, other evidence allows the conclusion that word recognition must involve an early phonetic component. Mezrich (1973) showed that the advantage in recognizing letters within words over recognizing single letters was actually reversed when the subject was required to vocalize the recognized word or letter. Hawkins, Reicher, Rogers, and Peterson (1976) required subjects to determine which of two alternative words had appeared. When the subject was set to rely on phonetic information in making this discrimination, performance was facilitated when a phonetic distinction between the alternatives was present, e.g., SOLD,COLD, and deteriorated when a phonetic cue was not available, e.g., SENT, CENT Krueger (1975) provided indirect evidence that verbal processes may underlie the superiority of a word. He showed that subjects could determine whether a target letter appeared in a word faster and with fewer errors than they could determine the presence of the target in a non-word only when the word or non-word was presented in the right visual field. Recent models of word and letter perception (Smith & Spoehr, 1974) allow For both spatial-visual and phonetic categorization but at different phases of the recognition process. The translation from a visual to a phonetic version seems to occur relatively late in the recognition process (e.g., Thorson, Hochhaus, & Stanners, 1976) and some theorists (Smith & Spoehr, 1974, p. 263) suggest it may not even be necessary in many tasks. The present experiment was designed to provide ( 1 ) a direct test of the unitization hypothesis that the word is the most compelling perceptual unit with visually. presented word stimuli and ( 2 ) evidence that the word unit is represented phonetically quite early in the recognition process even when this representation is not necessary to the task at hand. If a word is the salient
WORD
CONTEXT
IN
LETTER
RECOGNITION
937
perceptual unit with tachistoscopically presented word stimuli and if that word is available in a phonetic form, the subject should be able to recognize a target letter within that word more easily if the target letter is sounded in the word than when the target letter is unsounded. That is, it should be faster to find the letter G within the word TIGER than within the word RIGHT.
METHOD Twenty-four Florida International University undergraduate students (13 men and 11 women) participated in the study as a course requirement. The subjects were between the ages of 18 and 44 yr., with an average age of 27. Six of the 24 were bilingual in Spanish and English. Four of the 24 were left-handed. All had vision corrected to normal. Each subject participated in a single experimental session which lasted approximately half an hour. The stimuli were 60 five-letter words, each on a single card. These words along with cheir frequencies per million words (Kuqera & Francis, 1967) are listed in Table 1. Half the words had at least one letter which is not sounded at all or which is given "unusual" pronuncistion. These letters are the potential target letters. The words were chosen such that the potential target letters occurred equally ofcen in each of the five positions within the word. The other half of the stimulus words were chosen so that the potential target letter was clearly pronounced within the word. Any target letter occurred only once within the word. The design was a 2 x 2 x 5 factorial, with three repeated measures. Half of the 60 stimulus words in the stimulus set were randomly designated as containing a target letter for any one subject (target-present condition). The other half did not contain target letters (target-absent condition). It is important to realize that any one stimulus word might serve in the target-present condition for one subject and in the target-absent condition for another subject. Half the set of target-present words had the target clearly pronounceable in the word, e.g., target G in the word 1TGER. In the other half of the target-present words, the target was either not pronounced (target G in RIGHT) or given "unusual" pronunciation because of spelling rules (target I in JUICE). TABLE 1 POSITION OF
TARGET IN
WORDS IN WHICHTARGET L
2
1
E ~ IS R PRONOUNCED OR NOT
Postion of Target in Word 3
4
5
Target Letter Clearly PronouncedC
36 442
KINDS WATER 104 EIGHT 224 WOMAN 130 WORKS 7 KINGS
10 SWEAR 938 WHERE 3 SCARE 4 PETAL 95 FUNDS 67 ASIDE
7 TIGER 170 TRIED 22 FAULT 58 SMILE 34 CRIME 148 STAND
5
FLAGS 7 BARGE 48 LIMIT 3 STRUT N O SHALE 28 POUND
Target Letter Not Pronounced or Given "Unusual" Pronunciation* 13 DOUGH 76 KNIFE 7 SWORD 613 RIGHT 22 RHYME 11 JUICE 7 COUGH 106 WRITE 15 CRUDE 16 TRAINS 59 EXIST 106 SCENB 129 WRONG
9 -'I11 1 GNARL
173 HEART 76 QUIET 4 ISLES
10 GRIEF
52 RAISE 110 HEAVY
8 FRAUD 2714 WOULD 332Ll~HT
134 MAYBE 11 POSSE 3 PIZZA 20 PLAT0 1 POLKA NO ZIPPO
13 BLADE 41 ROUGH 2 COCOA 4 WEIGH 14 ARROW ~~OEARTH
*Numbers are the number of, times the word appears per million words (Kuqera & Francis, 1967)
.
938
M. M. JACEWICZ
The target letter for the target-absent word set was randomly determined, with the obvious constraint that it not appear in the word. All the letters of the alphabet were potential target letters. The stimulus set was created by applying black upper case transfer lettering (Chartpak Futura demi-bold 24 Pt.) to white cards. The word was centered around a central fixation point, with approximately 12 mm between the centers of each letter. The distance between the viewers' eye and the stimulus card was 85 cm. Subjects saw the unmasked word stimulus for 50 msec. The subject was seated at a Scientific Prototype two-channel tachistoscope and was told that the purpose of the study was to determine the speed with which they could recognize letters within words. The subjecc was given verbally a "target" letter and asked to indicate if that target letter was present or not in the word stimulus. A "yes" response meant the target was present; a "no" response indicated the target was not present. The stimulus presentation was always initiated by the subject at his or her own pace. Instructions to the subject stressed the importance of fixating upon a small black point and generally preparing oneself before initiating the stimulus. The dependent measure was the interval between the stimulus presentation and the voice response of "yes" or "no." Subjects wore a vibration-sensitive collar around their necks. The voice response rime was recorded with a Lafayette voice response control (Model 6602-A) and clock counter (Model 54519). On each trial ( 1 ) the experimenter informed the subject of the current target letter, ( 2 ) the experimenter reset the apparatus and gave the subject a ready signal, ( 3 ) the subject initiated the stimulus and ( 4 ) responded "yes" or "no." Each trial took about 20 sec. Each subject received numerous practice trials to ensure stable performance before response times were recorded. There were 60 recorded trials in each session. Although a 12% error rate was allowed, the average error rate was 4%. Only two of the 24 subjects exceeded an 8% error rate.
RESULTS Mean correct response times (RT) for each subject to detect the presence of a target letter within the stimulus word are recorded in Table 2. RTs for the condition in which the target letter was clearly "sounded" in the word and for the condition in which the target was "unsounded" are separaced in the rable. The difference between the two conditions for each subject is also noted. Data from subjects with average response times above and below the median value are presented separately. An analysis of variance was performed on the response times when che target letter was present in the word and the subjecc correctly detected its presence. The variables considered in the analysis were (1) whether or not the target letter was "sounded" in the stimulus word, ( 2 ) the position of the target in the stimulus word, and ( 3 ) subjects. All main effects reached significance. Targets which were sounded in the stimulus word yielded fascer response times than did targets which were not sounded ( M = 553.5, S D = 1 0 4 . 9 a n d M = 580.3, S D = 1 5 1 . 6 msec., respectively; F1,23= 7.03, MSE = 80.52, p .025). The time co find che target lecter in the word was fastest when the target was in the far left position and slowest at the far right position
