EMPIRICAL INVESTIGATIONS ON PRIMING OF WORD- STEM COMPLETION

Transcription

1 In: Psychology of Priming ISBN Editor: Nobuaki Hsu and Zacharias Schutt 2012 Nova Science Publishers, Inc. Chapter 1 EMPIRICAL INVESTIGATIONS ON PRIMING OF WORD- STEM COMPLETION Janet M. Gibson * Grinnell College AUTHOR NOTE This research was supported by Grinnell College faculty scholarship grants. I thank Lynn Hasher and Therese Macan who helped in arranging data collection at University of Toronto and University of Missouri, Saint Louis, and Michael Watkins for advice given at the beginning of the these investigations. I especially thank research assistants Lisa Grimm, Leticia Saintz, Stephanie Schneider, and Cassandra Damm for their help with stimuli selection, data collection, and scoring. ABSTRACT Boundary conditions of priming of word-stem completion are examined across twelve experiments. These conditions include: 1) when participants wrote as many words as they thought of to complete the stem for up to 18 seconds, enabling comparisons between priming levels for the target when written first, second, third, or within 6 s, 12 s, or 18 s; 2) over various delay intervals after presentation of 6 min through 2 days; 3) when stems had unique or multiple completions; 4) comparisons between priming of stem completion and fragment completion; 5), when the sound of the stem in isolation matched or did not match the intact target, and 6) when the stem was or was not the first syllable of the target. Methodological implications are discussed for selection of materials in priming studies utilizing word-stem completion. * Send correspondence to Janet M. Gibson, Dept. of Psychology, Grinnell College, Grinnell, IA 50112, gibsonj@grinnell.edu, (641) fax, (641) office.

2 2 Janet M. Gibson Two popular implicit memory tests are word stem completion, where initial letters are given, e.g., fra, and word fragment completion, where some letters of the word are replaced with a blank space marker, e.g., f _ a _ m e _ t, and participants complete the cue to make a word, e.g., fragment, with the first word that comes to mind. Priming is observed when the proportion of completions for previously presented targets is significantly greater than that for new, nonpresented words. This chapter reports on six experiments that compared priming of word stem completion with word fragment completion within a set of empirical conditions: when participants wrote as many completions as they could within short time limits, whether the stem or fragment could take multiple or unique completions, and when the time delay between presentation and test was short or long. The second set of six experiments explored the match of the sound of the stem in isolation to the intact target and whether or not the stem was the first syllable of the target. These studies explored how/why sound and syllable increased or decreased priming of stem completion. Together, the twelve experiments support the theory that priming of stem completion involves perceptual (visual and phonological), lexical, and semantic processes. They show that priming of stem completion is more than just a first guess phenomenon and is a robust tool to study implicit memory due to its versatility to control and manipulate the lexical qualities of targets. To begin, each set of experiments will be discussed separately to place the relevant issues in focus. COMPARISONS OF WORD STEM COMPLETION AND WORD FRAGMENT COMPLETION Initially, in the early days of implicit memory research, priming of stem completion appeared to decline faster over time delays between presentation and test than priming of fragment completion. For example, within one hour priming of stem completion was unobserved by Graf and Mandler (1984) but priming of fragment completion was observed beyond a year after presentation (Sloman, Hayman, Ohta, Law, & Tulving, 1988). Consequently, it was not uncommon in the 1980s for researchers to look at the differences in cues for the explanation. For example, Greene (1986) wrote Although the stem completion and fragment completion tasks seem formally similar, priming of these tasks does not necessarily arise from identical processes" (p. 664). However, both completion tasks were considered perceptual tasks (e.g., Roediger, Weldon, & Challis, 1989), and priming of both responded similarly to manipulations of levels of processing and modality. This conflict concerning delay seemed resolved when studies in the early 1990s found that priming of the two types of completion tasks did not differ over delay intervals when comparisons are made within experiments and not between experiments. For example, Roediger, Weldon, Stadler, and Riegler (1992, Experiment 2) examined priming of word fragment and word stem completion over 2 and 48 hours after immediate testing. In their design, the type of test and delay were manipulated between subjects (some also took an anagram task which is not of interest here): Those who took the 2 hour test did not take the 48 hour test and vice versa. They concluded that priming of the two tasks did not differ over delay. Rajaram and Roediger (1993) manipulated modality of presentation and found similar priming for word fragment and stem completion. Likewise, Craik, Moscovitch, and McDowd

3 Word-Stem Completion 3 (1994, Experiment 2) examined modality of presentation and priming of word stem and fragment completion over a 24 hours delay. They, too, used a between-subjects design and found no differential effect of delay on priming of fragment and stem completion (and found modality had a similar effect for both tasks, consistent with the theory that these two implicit memory tasks are perceptual tasks). These studies used a design where different participants were taking the two completion tasks, but other variables, such as words used and the nature of the presentation phases, were kept constant. The conclusion from these researchers was that delay did not have differential effects on priming of these two tasks. The experiments reported in this chapter supplement the findings concerning comparisons of priming of stem and fragment completion and contribute to this literature in at least three small but important ways: One, they compare the effect of delay on priming of fragment and stem completion within subjects--the same participant completed both the fragment- and stem-completion tasks, and at all delay intervals. Counterbalancing ensured that a target was not tested more than once to avoid repetition or test contamination effects. Two, the number of completions to the cues was examined by comparing fragments with one or many solutions with stems with one or many solutions. Most studies using fragment completion used fragments with one completion (Tulving, Schacter, & Stark, 1982) whereas word stems (e.g., Graf & Mandler, 1984; Graf, Squire, & Mandler, 1984; Mandler, Graf, & Kraft, 1986) had at least 10 completions. For example, in Roediger et al. (1992), 70% of the fragments were unique and all stems had multiple completions. In Craik et al. (1994), all fragments were unique and stems had at least 4 completions. The number of completions may have an effect on priming of completion (certainly on baselines), and, more importantly, priming over delay intervals. One reason this may be so is the role activation may play in priming and the role of competitors on the level of activation of the target. When cues have many completions, activation is shared (e.g., Nelson, Bajo, & Casaneuva, 1985) and thus make it less likely for the primed target to come to mind when reading the cue. Unique cues have no competition, and the activation of the target will help that target to come to mind when cued. Graf and Mandler (1984) expressed this view when they said, The availability of alternative completions for each of the cues on the completion test transforms it into a test of competition the recently presented words will be given on the word stem completion test as long as they provide the most accessible response to the cues" (p. 562). A third contribution of the present set of experiments concerns strategies the participants may use during the time given to complete the cue. Because word fragments often are difficult (though difficulty may be manipulated by the number of letters provided in the cue), but most 3-letter stems are considered easy to complete, it is often the case that, across experiments, more time is given to participants to complete a word fragment than a word stem. More time often implies the engagement of effortful or conceptual processes. Again, Graf and Mandler (1984) noted this difference between tasks when they said, on the stem completion test participants give a response within a few seconds, whereas the word fragment test is demanding of time and effort. When no completion immediately comes to mind, participants may attempt to retrieve words they have recently seen (p. 562). The concern is that fragments have a bias toward explicit retrieval or deliberative lexical search strategies (e.g., Nelson, Keelean, & Negrao, 1989). Thus, even if explicit strategies are ruled out after no levels of processing effects are found for priming of word-fragment and stem completion, some lexical search strategy could make the two cues different in how they are

4 4 Janet M. Gibson completed and, more importantly, how the prior presentation of the target interacts with such strategies. In several experiments reported in this chapter, I examine completions written by participants over equivalent timed intervals where participants were asked to write as many completions to the cues as they could. This method allows us to know the order of output and whether the target was written within 6, 12, or 18 seconds. Analyses examine priming for the first word, second or third word, or within 6, 12, or 18 seconds. The results show that, when more than just the first word to come to mind is asked for, the advantage of presented over nonpresented words remains fairly constant, with only a small increase found for both presented and nonpresented words over time and opportunity. Thus, whether more or less time is given, and whether more than one response is thought of and expressed, priming of both fragment and stem completion look much like when the first response only is measured. Additionally, these studies will show very little effect of the number of completions to the cues on priming over time delays, and essentially that, as found with Roediger et al. (1992) and Craik et al. (1994), delay need not be more harmful to or dissociate priming of stem completion and fragment completion. THE STEM-IN-ISOLATION TO STEM-IN-TARGET RELATIONSHIP IN STEM COMPLETION When the first three letters (the stem) do not sound in isolation the way they sound when in the target, this incongruency is called dysphonemia (Brooks, Gibson, Friedman, & Yesavage, 1999). The dysphonemic effect is a reduction in priming of visual word stem completion when the stem (e.g., ANT) does not sound like its completion (e.g., ANTHEM) compared to when it does (e.g., ANTLER). Brooks et al. (1999) reported that priming of word stem completion is dependent on whether the stem in isolation sounds like those letters in the intact word. They called the match "phonemic" when the three-letter stem likely sounded the same as in the word, e.g., COL and COLONY, FIG and FIGURE, or MUS and MUSTARD, and the mismatch "dysphonemic", e.g., COLONEL, FIGHT, or MUSHROOM. Brooks et al. (1999) found that priming for dysphonemic items was much lower than that for phonemic ones. This dysphonemic effect was replicated in several experiments, with both older and younger adults, and was shown not to be tied to the selected phonemic and dysphonemic targets, as these words primed the completion of their 4-letter stems and word fragments equally well. Ryan, Ostergaard, Norton, and Johnson (2001, Experiment 2) reported a similar finding based on matching spoken pronunciation of the stem with the target: for both older and younger adults, priming was lower when the sound of the stem in isolation did not match the sound of the stem in the target. The dysphonemic effect implicates a phonological component to the priming of visual word stem completion. Other researchers have argued for such a component. For instance, Rueckl and Matthew (1999) found that homophones with different 3-letter stems primed each other's completion (e.g., the presentation of WEEK biased the completion of wea with WEAK). Research on the relationship between phonological processes and priming of visual

5 Word-Stem Completion 5 word stem completion should help us to better understand how the prior presentation of words facilitates performance of word-stem completion. The purpose of six experiments presented in this chapter was to further examine the dysphonemic effect under several experimental manipulations and examining whether the stem as the initial syllable of the target causes or modulates the effect. In the present series of experiments, words are classified in the dysphonemic condition within a variety of criteria, some of which include: a) the third letter was the beginning of a 2-letter phoneme (e.g., s for sh, t for th, or g for gh, as in MUSHROOM, MOTHER, or LIGHT), b) the third letter in the isolated stem evoked the hard g or c but the target used the soft letter (e.g., LEGEND and SECEDE), c) a silent third letter (e.g., SUBTLE), or d) a change in the whole stem pronunciation (e.g., SERGEANT and CELLO). Note that we did not use the criteria of the dominant long/short vowel in spoken pronunciation of the isolated stem to classify stems. That is, CATER and CATCH were both considered potential phonemic completions to the stem CAT, or LAKE and LACK to the stem LAC, because choice of short or long vowel may be arbitrary (or biased by recently processed words) for the isolated stem and perhaps not consistent within a participant on different occasions. Technically, our definition was that dysphonemic stems could not sound like the target s initial letters, but the phonemic ones could match. Because no one ever would pronounce the T in the isolated stem CAT as a TH, these type mismatches were labeled dysphonemic, but whenever phonemic pronunciations were theoretically possible (e.g., KATE or KAT for CAT), we placed that item in the phonemic category. We recognized that we could not control all contexts and accents, and thus could not predict for some words (such as FORWARD) whether participants will pronounce the target FOHWARD or FOREWARD ) each time they read the word; dominant, common, or dictionary pronunciations do not always match a speaker s pronunciation. Most importantly, the potential vowel mismatches in the phonemic category did not prevent us from testing and rejecting the null hypothesis (pure/potential match = mismatch). Despite the phonemic label on targets-and-stems that might not share the initial short or long vowel, our findings clearly show in all six experiments that targets we classified as having phonemic stems significantly prime stem completion more than dysphonemic primes. In Brooks et al. s (1999) study, many dysphonemic stems broke the intact syllable whereas phonemic stems maintained it. Thus, finding that targets were used less likely as completions to stems when the sound did not match between target and stem may be due entirely or to a large extent by the lexical property of the stem. Ryan et al. (2001) argued in their paper that participants were silently reading the stem and using its sound to search for a completion. It is likely that the sound of the stem is encapsulated as a syllable, which by definition is a break in pronunciation of a word, and is used as a lexicon search unit for finding a completion. Ryan et al. asked their pilot participants to pronounce the stem as a word, which is also a monosyllabic unit. They used their responses to match stem and target on the basis of dominant pronunciation. It follows, then, that even if sound information is not used when looking at stems, a lexical search for a completion could use the stem-as-asyllable-search unit, and thus would be more likely to find completions that have the stem as the first syllable. Thus, a sound or a syllable match would favor targets classified as phonemic to be completed more often than dysphonemic ones. I explored this confound of syllable with sound in six experiments in the present chapter to help understand why the dysphonemic effect occurs. These experiments sought evidence for the role of sound or syllable to priming

6 6 Janet M. Gibson of stem completion and at the same time investigated some effects of traditional variables (e.g., modality of presentation) in hopes of providing insight into the cause of the dysphonemic effect beyond its definition. COMPARISONS OF STEM COMPLETION AND FRAGMENT COMPLETION ((EXPERIMENTS 1-6) The first six experiments presented below make a comparison of priming of completion of fragments and stems. In Experiment 1, a levels of processing (LOP) manipulation was used to show that no LOP effect is observed with these two tests and allays fears of explicit memory contamination with fragments as compared to stems in the paradigm I use here (i.e., write as many words as one could with up to 18 s per cue). Measures of up to 18 seconds were taken with the multiple stems to examine whether priming is a "first" effect and remains constant across output order. The second experiment compared multiple fragments and multiple stems over a 15-minute delay with a 12-second interval for several completions written to each cue. The third experiment compared unique and multiple stems to determine whether the number of completions to stems matters to the level of priming or to interactions with delay for stem completion. Experiments 4 and 5 compared unique fragments, multiple fragments, and multiple stems over three delay intervals (6 min, 30 min, and 48 hours), with Experiment 4 using 18 seconds for writing completions per cue and Experiment 5 using just 8 seconds. Finally, the sixth experiment compared unique fragments and multiple stems when the same target was used as completions to both type of cue (in previous experiments, words were counterbalanced across delay and presentation but not cue). Experiment 1 The first experiment examined the effect of a 15-minute delay on the priming of word fragment and word stem completion under deep and shallow conditions. Participants were instructed to complete cues by writing as many words as they could within 18 seconds. The response sheets and instructions were designed so that it would be known when the word was written by the participant either first, second, third, or fourth, or within 6 seconds, 12 seconds, or 18 seconds. Method Participants Thirty-two undergraduates served as participants. Materials The materials consisted of letter words, selected from the Puzzle Solver's Handbook (1970), a word locator for solving crossword puzzles. Selection was based on the following criteria: a) 2 of the six letters in the word uniquely cued it so that no other word in the pool could complete the 2-letter fragment, and b) the initial 3 letters of each target were

7 Word-Stem Completion 7 unique, so that none of the selected targets began with the same 3 letters. Of these selected words, 192 were chosen as test items, and 20 were used as primacy and recency items on the ratings lists but were not tested. A fragment was constructed by adding a third letter to the two already uniquely specifying cues; the choice was made by selecting the letter thought to provide most information about the fragment's identity. Design The pool of 192 targets was divided into 4 sets (A, B, C, and D) of 48 words each. Each set was then divided into 2 lists of 24 words each (e.g., A1 and A2). Each participant received 4 lists for the rating task, either A1, B1, C1, and D1 or A2, B2, C2, and D2. Five words that were not tested were assigned to each set to begin and end the two presentation lists for that set. To form the tests, 12 words from both presentation lists for each set were randomly selected to form 24-item tests; thus each test contained 12 presented and 12 nonpresented words. Two such tests were made for each set, one given immediately and one given after the delay. Each test was composed of either all fragments or all stems. Two fragment and 2 stem tests were given immediately and at delay. For immediate testing, each type of test was preceded once by a presentation list that was rated for liking of meaning and once by a presentation list that was rated for liking of sound. For the delay testing, the four tests were given in the same order as their immediate counterparts. Counterbalancing across participants allowed for every word to be either presented or not presented, rated for participants' liking of its meaning or its sound, tested immediately or after a delay. Procedure The first task in all booklets was a rating task. Participants were told to rate each word on a scale of 1 to 3 for whether they liked the sound of the word (shallow) or the meaning of the word (deep), depending on condition (1 = disliked, 2 = neutral, and 3 = liked). Three seconds were allowed for each rating. Participants were told to proceed down the page, moving a blank sheet that covered the page with each beep to reveal the next word The next task was one of either fragment completion or stem completion. No reference was made to the fact that previously rated words were included in the test. For the unique fragment test, participants were told that if they thought of more than one word that completed the fragment they should write only the first one that came to mind. They were given 6 seconds per fragment. If they had not written a word when time was up, they were to leave the line blank, move the sheet down to uncover the next fragment, not to go back and complete earlier fragments, and not to worry if they later realized that a word they had written did not actually complete a fragment, as the experimenter would simply ignore it. A practice page containing two fragments was provided just prior to the participants' first encounter with the fragment task in the booklet. The instructions for the word stem completion task were to write as many words as they could think of that began with the 3-letter string on the left margin. Three lines per stem were provided for this task, and participants were given 6 seconds per line to write as many words as they could. It was pointed out that participants should write the words in the order in which they came to mind, and to write left to right on each line so that the experimenter would know which word was written first, second, and so on. Practice was provided on two stems prior to the participants' first encounter with the stem task in the booklet, and examples of possible completions were provided.

8 8 Janet M. Gibson For all tasks, performance was monitored to ensure compliance with task instructions. The booklet contained 12 tasks in all--4 rating lists, 4 word stem tests, and 4 fragment tests. The first half of the booklet contained 4 rating list-test combinations, and the last half of the booklet contained 4 tests. The session lasted minutes, and afterwards, participants were debriefed. Results The number of times the fragment or stem was completed with a target was totaled for all conditions for each participant. The stem data were scored according to each of four criteria: whether the target word was written (a) first, (b) within the first 6 seconds (line 1), (c) within the first 12 seconds (lines 1 or 2), or (d) within the allotted 18 seconds (at all). The mean proportions (and SD) for fragment and stem completion are provided in Table 1 for presented and nonpresented targets. As desired, baseline performances on the stem task surround the baselines for fragments, and the two types of cues can be considered to be of similar difficulty. Priming was measured as the difference in completion for presented and nonpresented words. As can be seen in Table 1, more completions occurred for presented words, and they are greater for immediate testing than for delayed testing. Also, comparison of the differences for words rated for sound and meaning shows the rating tasks had little effect on the degree of priming of word fragment or stem completion. For all analyses reported in this chapter, a p <.05 is implied for significant effects. Table 1. Mean (and SD) Proportion of Targets Completing the Fragment and Stem Cues, Experiment 1 SOUND Fragments Stems First Line 1 Line 1+2 At All 6s 12s 18s Imm. Presented.38 (.07).25 (.07).27 (.07).31 (.07).32 (.07) Nonpresented.15 (.08).13 (.05).14 (.05).17 (.04).18 (.05) Difference.23 (.11).12 (.08).13 (.08).14 (.08).14 (.07) Delay Presented.27 (.08).21 (.07).23 (.08).26 (.08).28 (.08) Nonpresented.16 (.08).13 (.06).15 (.07).18 (.07).19 (.07) Difference.11 (.09).07 (.08).09 (.08).09 (.09).09 (.08) Difference.12 (.15).04 (.12).04 (.12).05 (.12).05 (.12) MEANING Imm. Presented.37 (.07).22 (.08).26 (.08).30 (.07).31 (.07) Difference.23 (.10).10 (.09).13 (.09).14 (.09).13 (.09) Delay Presented.26 (.10).22 (.09).24 (.09).28 (.09).29 (.08) Nonpresented.16 (.09).13 (.05).14 (.05).17 (.07).18 (.07) Difference.10 (.10).09 (.08).10 (.08).11 (.08).11 (.07) Difference.13 (.14).02 (.11).02 (.12).03 (.11).02 (.11) Note: Timing of Stem Lines is cumulative.

9 Word-Stem Completion 9 Fragments A 2 priming (presented or not) x 2 time (immediate or delayed) x 2 rating (meaning or sound) ANOVA on the word fragment data showed a strong effect for priming, F(1, 31) = , MSe = 3.66, but no effect of rating task, F(1, 31) =.02, MSe = There was a significant priming x time interaction, F(1, 31) = 54.97, MSe = 2.59, due to a decline in priming over the interval. The facilitation of completion for presented words was reliable at immediate testing, F(1, 31) = , MSe = 3.47, as well as at delayed testing, F(1, 31) = 71.00, MSe = Stems The stem data were analyzed according to each of four criteria: the "first" criterion whereby a stem was scored correct only if written first regardless of which line it appeared; the "line 1" criterion whereby a stem was scored correct if the target was written on line 1, the "lines 1 and 2" criterion, whereby a stem was scored correct if the target was written on lines 1 or 2, and the "at all" criterion whereby a stem was scored correct if the target was written at all. To anticipate, giving participants more time or chances to write the target did not greatly increase the likelihood that the target would be written for either presented or nonpresented words. Consequently, the same pattern of results was obtained for each criterion. A 2 priming (presented or not) x 2 time (immediate or delayed) x rating (meaning or sound) ANOVA on the stem data showed a strong effect of priming, F(1, 31) = , MSe = 2.58; , MSe = 2.35; , MSe = 3.14; and , MSe = 2.54 for the first, line 1, lines 1 and 2, and at all criteria. Rating for meaning or for sound made little difference to the degree of priming, F's (1, 31) =.02, MSe = 1.58;.25, MSe = 1.25; 1.35, MSe = 1.53; and.31, MSe = 0.81, respectively, nor did it interact with other factors. The decline in priming over time was also reliable, F's (1, 31) = 6.92, MSe = 1.36; 6.88, MSe = 1.36; 6.09, MSe = 1.67; and 7.38, MSe = 1.54, respectively. The priming effect was reliable at immediate testing, F's (1, 31) = , MSe = 2.19; , MSe = 1.85; , MSe = 2.14; and , MSe = 2.01, and at delayed testing, F's (1, 31) = 66.49, MSe = 8.15; 88.57, MSe = 1.86; 76.57, MSe = 2.33; and 84.85, MSe = 2.07, respectively. Fragments and Stems Turning to the comparison of fragment completion with each criterion of stem completion, the first question of interest is whether the facilitation resulting from prior presentation differed for the two types of completion cues. The priming x cue interaction was reliable, F's (1, 31) = 36.22, MSe = 2.51; 22.06, MSe = 2.56; 12.06, MSe = 3.00; and 17.31, MSe = 2.71, for comparing fragment completion with stem completion for the criteria of first, line 1, lines 1 & 2, and at all, respectively. This interaction did not involve rating task, F's (1, 31) =.00, MSe = 1.3;.24, MS e = 1.16; 1.01, MSe = 1.31; and.26, MSe = 0.92, p >.10, respectively. The difference in priming of fragment and stem completion did interact with the time of testing, F's (1, 31) = 20.35, MSe = 1.94; 18.82, MSe = 2.09; 17.85, MSe = 2.14; and 18.47, MSe = 1.99, respectively. This interaction was due to a difference in priming of the two cues on the immediate test but not on the delayed test: Priming of fragment completion was greater than that of stem completion on the immediate test, F's (1, 31) = 49.60, MSe =

10 10 Janet M. Gibson 2.52; 36.07, MSe = 2.57; 23.57, MSe = 3.16; and 28.61, MSe = 2.95, for the respective comparisons, but not for the delayed tests, F's (1, 31) = 2.77, MSe =1.93;.37, MSe = 2.05;.13, MSe = 1.90; and.17, MSe = 1.82, p >.10, respectively. Discussion Priming was observed on both fragment completion and stem completion tests, and at both immediate and delayed testing. Levels of processing had no discernible effect on the degree of priming manifested in word fragment or word stem completion, a finding that helps allay fears of strong explicit memory contributions. Delay, on the other hand, did affect priming of both word fragment and word stem completion: The amount of facilitation observed on the immediate test was reliably greater than the amount observed on the delay test. The decline in priming over the 15-minute delay was greater for fragment completion than for stem completion. At the 15-minute delay interval, however, the degree of priming of fragment completion did not differ from that of stem completion. Lastly, the degree of priming observed was not affected by whether participants wrote the first word or as many words as they could within the short time limits. Experiment 2 Because the levels of processing manipulation did not have an effect on priming of fragment or stem completion it is not examined in the remaining five experiments. The number of completions to the two types of cues differed in Experiment 1 and might play a role in the difference observed. In Experiment 2, both fragments and stems had more than 1 completion. Both fragments and stems were presented to participants on the same test, mixed randomly. Participants were encouraged to write more than one answer to each type of test cue within 12 seconds, and a 15-minute delay was examined. Method Participants Forty undergraduates participated. Materials The 219 words were randomly selected from the Toronto Word Pool (Friendly, Franklin, Hoffman, & Rubin, 1982)--a set of 1060 common bisyllabic words--with the restrictions that, of those selected, no two words could begin with the same three letters, and no two words could have identical fragments. Fragments were constructed by replacing the first letter and all vowels with blanks and could be completed by other words besides the target. Stems were constructed by using the first three letters of a word. Of these selected words, 192 were selected as targets, 20 were used as primacy and recency items on the presentation lists and were not tested, and 7 were used as practice items for the rating and word completion tasks. An interval timer attached to a speaker was used to limit time spent on each item.

11 Word-Stem Completion 11 Design The 192 words were divided into 4 sets of 48 words each. Twenty-four of each set composed one presentation list that a given participant saw, and all 48 words were tested, half immediately and half after a delay. In each test, participants saw 12 words that they had rated and 12 that they had not. Within these groups of 12, 6 were cued with a stem and 6 with a fragment. Five non-tested words were assigned to each presentation set, and 2 were placed in the beginning and 3 at the end. Balancing across participants, a given word was either presented or not presented, tested immediately or at delay, and was cued by a fragment or a stem. The design was a 2 priming (presented or nonpresented) x 2 time (immediate test or delay test) x 2 cue (stems or fragments) within-factors design. Procedure The participants were tested up to 6 at a time in a session that lasted about 55 minutes. The first task was the rating task; participants rated for their liking of the meaning of each word. The second task was the word completion task. For both types of cues, they were given 2 lines to write responses, and 6 seconds per line. When they heard the beep, they uncovered the cue, wrote words appropriate for that cue, and remained on that line until the next beep, then moved to the next line, and continued writing words appropriate to the cue. No reference was made concerning any relationship between the words in the rating list and the cues. Results Statistics were calculated for the cues completed with targets "first" and also "at all" within the 12 seconds. The means (and SDs) for the conditions are presented in Table 2. In general, cues were completed with targets more often when the word had been presented and tested immediately. Table 2. Mean (and SD) Proportion of Targets Completing Fragment and Stem Cues, Experiment 2 Immediate 15 m Delay Multiple Fragments Multiple Stems First At All First At All 12s 12s Presented.58 (.14).61 (.16).49 (.19).63 (.20) Nonpresented.33 (.09).35 (.10).23 (.12).35 (.16) Difference.25 (.18).26 (.18).26 (.26).28 (.30) Presented.42 (.13).44 (.13).38 (.15).53 (.19) Nonpresented.31 (.11).34 (.13).27 (.10).39 (.14) Difference.10 (.16).11 (.18).11 (.21).14 (.28) Difference.15 (.27).15 (.30).15 (.41).14 (.54) Note: Timing of Stem Lines is cumulative.

12 12 Janet M. Gibson Multiple Fragments Fragments of presented words were completed by the target reliably more than were fragments of nonpresented words, F(1, 39) = , MSe = 5.89 for first, and F(1, 39) = , MSe = 6.80 for at all. This was true both at immediate testing, F(1, 39) = 77.79, MSe = 8.94 for first, and F(1, 39) = 77.32, MSe = 9.78 for at all, and at delayed testing, F(1, 39) = 15.25, MSe = 7.72 for first and F(1, 39) = 13.72, MSe = 9.66 for at all, but the priming effect was reliably smaller after the delay, F 's (1, 39) = (MSe = 10.77, for first, and F(1, 39) = 10.10, MSe = for at all. Multiple Stems Stems of presented words were completed more than were stems of nonpresented words, F(1, 39) = , MSe = 7.76 for first and F(1, 39) = , MSe = 6.29 for at all. This was true at immediate testing, F(1, 39) = 39.70, MSe = and F(1, 39) = 34.57, MSe = for at all, and at delayed testing, F(1, 39) = 11.67, MSe = 12.27, and F(1, 39) = 9.25, MSe = 22.84, for first and at all, respectively. The decline in priming over time was significant for stems completed first with the target, F(1, 39) = 5.21, MSe = 23.87, but not for at all, F(1, 39) = 2.78, p >.10. Fragments and Stems Of primary interest is the comparison of the effect of priming, time, and priming x time interaction between fragment and stem completion. Four comparisons were possible, depending on which level (first or at all) was used, and the four comparisons yielded basically the same results. For readability, only the first scoring statistics are reported here. There was a main effect of priming, F(1, 39) = , MSe = 8.75, and the effect of cue was also reliable, F(1, 39) = 20.71, MSe = The time x priming interaction was reliable, F(1, 39) = 40.06, MSe = More importantly, there was no cue x priming interaction, F(1, 39) =.31, MSe = 4.90: The priming effect arising from prior presentation for the two cues did not differ at immediate testing, F(1, 39) =.05, MSe = 19.46, or at delayed testing, F 's (1, 39) =.04, MSe = Unlike the reliable cue x time x priming interaction found in Experiment 1, the decline in priming over time also did not differ between the two cues, F(1, 39) =.00, MSe = Discussion The results from Experiment 2 indicate that priming of word fragment completion and word stem completion did not differ from each other, and, contrary to what was found in Experiment 1, the decline in priming over time did not differ for the two types of cues. Two differences in methodology between Experiments 1 and 2 may account for the difference. One, in Experiment 1, fragments and stems were blocked by test whereas in Experiment 2 they were mixed. Two, fragments in Experiment 1 were unique to a pool whereas they had several completions in Experiment 2.

13 Word-Stem Completion 13 Experiment 3 The third experiment addressed the question of whether stems having only one completion would show different effects of delay relative to stems having more than 10 completions. Unique stems are not used often in the priming literature, but it may be helpful to know whether the number of completions to stems matters to priming observed in priming of stem completion. The effect of a 15-minute delay on priming of completion was examined. Method Participants Thirty-two undergraduates, 16 in each stem group, served as participants. Materials A search through a pocket dictionary (Webster's II New Riverside Dictionary, 1984) resulted in finding about 190 words that had no other entry beginning with the word's initial 3 letters. Of these, 128 were chosen as suitable for testing, and an additional 20 words were chosen for buffer items on the presentation lists. For the multiple completion group, 148 words that had 10 entries in the dictionary and did not share their stems with each other were chosen. (Most of these words came from Experiments 1 and 2, and the rest were randomly chosen from the dictionary.) Design For each type of stem, the 128 targets were divided into 4 presentation sets of 32 words each. Sixteen of each set of 32 words comprised one presentation list that a given participant saw, and all 32 words were tested, half immediately and half at delay. Five additional words were assigned to each presentation set--two at the beginning of each presentation list and three at the end; these 5 buffer items were similar to the presentation words in terms of the number of completions to their stems, but they were not tested. Counterbalancing across participants, a given word was either presented or not presented and was either tested immediately or after a delay. The type of stem (having unique or multiple completions) was a between subjects variable. Procedure Participants were tested individually or up to 6 at a time. The procedure was much the same as in Experiments 1 and 2. All words were rated for how much participants liked the sound of the word. The group that received unique stems was told to complete the three letters with one word, and it was emphasized that not too many words could complete the stems. The group that received the multiple stems was told to write as many words as they could think of in the 6 seconds. The delay interval was approximately 15 minutes between presentation and test. Results The means (and SDs) for both types of stems, as well as the differences between presented and nonpresented words, are shown in Table 3. Cues for both unique stems and

14 14 Janet M. Gibson multiple stems were completed with the target words more often when the word had been presented and tested immediately. It should be noted that the baselines for the two groups differ. Unequal baselines probably do not reflect a difference in task difficulty but, rather, reflect the fact that the probability of completing a stem with the target is much higher when no other word can be written in its place. Although the critical comparison is between the stem completion groups, the statistics for each group will be discussed first. As in the earlier experiments, the multiple stem completion group was scored both when the target was written first and also when written at all (within 6 seconds). Table 3. Mean (and SD) Proportion of Targets Completing Unique and Multiple Stems, Experiment 3 Unique Multiple (first) Multiple (at all) Immediate 6 s 6 s Presented.79 (.12).50 (.11).59 (.10) Nonpresented.43 (.13).13 (.08).21 (.09) Difference.36 (.10).37 (.13).38 (.13) 15 m. Delay Presented.66 (.19).31 (.09).38 (.10) Nonpresented.44 (.16).14 (.06).18 (.05) Difference.22 (.11).17 (.12).20 (.11) Difference of Differences.14 (.17).20 (.18).18 (.16) Note: Timing of Stem at all is cumulative. Unique Stems Stems of presented words were completed more than stems of nonpresented words, F 's (1, 15) = , MSe = This advantage was found in immediate testing, F 's (1, 15) = MSe = 4.99 and in delayed testing, F(1, 15) = 60.00, MSe = There was a significant decline in priming over the delay, F(1, 15) = 11.05, MSe = Multiple Stems There was little difference between the first and at all measures, and therefore only the first are reported here. Stems of presented words were completed more than stems of nonpresented words, F(1, 15) , MSe = This advantage was found for immediate testing, F(1, 15) = , MSe = 8.77, and for delayed testing, F(1, 15) = 29.16, MSe = There was a significant decline in priming over the delay, F(1, 15) = 20.00, MSe = 8.13.

15 Word-Stem Completion 15 Unique and Multiple Stems The type of cue did not interact with priming, F(1, 30) =.66, MSe = 6.28, or delay, F(1, 30) =.59, MSe = 11.14, nor did the decline in priming over time interact with the type of cue, F(1, 30) =.84, MSe = Discussion The results of Experiment 3 indicate that the decline in priming of stem completion over the delay did not differ significantly for the two types of stem completion groups. Thus, the number of completions (unique vs. multiple) likely plays a minimal role in priming of stem completion. Researchers may want to consider the number of completions for controlling the level of nonpresented words (i.e., baselines). Experiment 4 The fourth experiment compared the effect of longer delays on priming of fragments with one completion, fragments with more than 10 completions, and stems with more than 10 completions. Stems having one completion were not used here because they did not differ from stems with more than 10 completions in Experiment 3. Three time intervals were used: 6 minutes, 30 minutes, and 48 hours following list presentation. The type of cue was a between-subjects factor, meaning that all three word completion tests a participant received contained the same type of cue, thus replicating the test conditions previously used by Roediger et al. (1992) and Craik et al. (1994) where participants received only one type of word completion test in previous studies, but delay was manipulated withinsubjects. Method Participants Seventy-two undergraduates served as participants, 24 in each of three conditions. Two participants did not return to the second session and were replaced. Materials All words were chosen from those listed in Francis and Kucera (1982) word frequency database, a pool of more than 50,000 words. 1 Word length was restricted to 5 to 7 letters. For the multiple fragment condition, all fragments containing 3 letters that could be completed by 8 to 15 words from the pool were listed, and 180 fragments/targets were selected as randomly as possible. Restrictions on randomness were that no target could complete another target's fragment, fragments containing successive letters, especially the initial letters, were avoided, and the frequency of the target words on average were matched to the frequency of the other two conditions. For the unique fragment condition, 180 words and fragments were selected from among those listed in Gibson and Watkins (1988), a listing of unique fragments developed from the Francis and Kucera (1982) pool. No words from this pool could complete

16 16 Janet M. Gibson these fragments other than the targets. Selection was limited, therefore, only in word length and in word frequency, which was, on average, matched to the multiple fragment and stem conditions. For the stem condition, 180 targets were randomly selected from a listing of words from the Francis and Kucera pool having 8-15 words sharing initial 3 letters. Selection was as random as possible given that the target needed to be 5-7 letters in length and on average match the frequency of the other two conditions. A timer was used to pace participants. Design Type of cue (unique fragment, multiple fragment, multiple stem) was a between-subjects factor, and delay interval (6 minutes, 30 minutes, 48 hours) and priming status (presented, nonpresented) were within-subjects factors. Within each type-of-cue condition, 180 targets were tested, 60 at each delay interval. Two presentation lists of 90 targets each were constructed. These presentation lists, regardless of condition, were preceded by 2 nontested words and followed by 3 nontested words. Half the items of each test were words from the presentation list the participant received, and half were not. Counterbalancing across participants ensured that every target within a condition was presented or not presented and tested at each delay interval equally often. Procedure The presentation task asked participants to rate words for how much they liked the sound of the words. The participants in the multiple fragment and multiple stem completion conditions were told that if they thought of more than one answer, they were to write the first one that came to mind within 8 seconds. No reference was made that words from the ratings list could complete the test items. The first test took 8 minutes. Following this test, participants in all conditions took a filler task lasting approximately 20 minutes. This filler task was an auditory word fragment completion test that contained words that could not complete fragments or stems in the experiment proper. After the auditory task, participants took a second word completion test of their cue condition. The time interval between the end of the presentation list and the beginning of this second completion test ranged from 28 to 33 minutes. The variance resulted largely in group questions during the instruction phases of the completion tasks and the filler task. Participants returned 48 hours later for the third completion test. Results Table 4 displays the mean proportion (and SDs) of items correctly completed in all conditions. Across all types of cues and times of testing, it can be seen that previously presented targets facilitated completion.

17 Word-Stem Completion 17 Table 4. Mean (and SD) Proportion of Targets Completing Fragments and Stems, Experiment 4 6 minutes Unique Fragments Multiple Fragments 8s 8s 8s Stems First Stems At All Presented.23 (.14).14 (.06).32 (.09).38 (.12) Nonpresented.15 (.10).05 (.04).16 (.05).20 (.08) Difference.09 (.12).08 (.08).16 (.10).18 (.11) 30 minutes Presented.21 (.10).13 (.08).28 (.09).36 (.13) Nonpresented.13 (.08).05 (.05).14 (.06).19 (.09) Difference.08 (.12).08 (.09).14 (.10).17 (.10) 48 hours Presented.20 (.12).11 (.07).23 (.06).30 (.07) Nonpresented.12 (.10).07 (.05).15 (.07).21 (.09) Difference.08 (.11).04 (.06).08 (.08).10 (.10) Note: At All output is anywhere in output within 8 seconds. Unique Fragments A 2 priming x 3 time ANOVA revealed an effect for priming, F(1, 23) = 24.96, MSe = Priming was observed at 6 minutes: t(23) = 3.61, at 30 minutes: t(23) = 3.48, and at 48 hours: t(23) = There was no effect of time, F(2, 46) = 1.78, MSe =.478, p >.17, nor did priming significantly decline over time, F(2, 46) =.05. Multiple Fragments A 2 priming x 3 time ANOVA revealed an effect for priming, F(1, 23) = 61.55, MSe = Priming was observed at 6 minutes, t(23) = 4.77, 30 minutes, t(23) = 4.30, and 48 hours, t(23) = As in the case with unique fragments, there was no effect of time, F(2, 46) =.16, nor a priming x time interaction, F(2,46) = Multiple Stems A 2 priming x 3 time ANOVA revealed an effect for priming, F(1, 23) = , MSe = Priming was measured at 6 minutes, t(23) = 7.54, 30 minutes, t(23) = 7.21, and 48 hours, t(23) = There was an effect of time, F(2, 23) = 5.53, MSe = 5.03, and priming declined over the delay, F(2, 46) = 4.64, MSe = Whereas the decline from 6 to 30 minutes was not significant, t(23) =.51, priming did decline from 6 minutes to 48 hours, t(23) = 2.77, as it did from to 30 minutes to 48 hours, t(23) = 2.47.

18 18 Janet M. Gibson Fragments and Stems A 3 cue x 3 time x 2 priming ANOVA revealed significant effects for cue, F(2, 69) = 35.14, MSe = 14.64, time, F(2, 138) = 6.15, MSe = 4.25, priming, F(1, 69) = , MSe = 5.17, and a cue x priming interaction, F(2, 69) = 6.51, MSe = No other interactions were significant. Separate analyses compared unique fragments with stems, and multiple fragments with stems. Priming of stem completion was significantly greater than priming of unique fragment completion at the 6 minute test, t(46) = 2.23 but not at the 30 minute or 48 hour tests. Priming of stem completion was also significantly greater than priming of multiple fragment completion at all three time intervals: 6 minutes, t(46) = 2.78, 30 minutes, t(46) = 2.42, and 48 hours, t(46) = Finally, the degree of priming did not differ at each time interval for the two fragment conditions. Discussion Based on a failure to find a 3-way interaction, the decline in priming did not differ depending on type of cue. Although the pattern was similar to that suggested in the early literature, these analyses indicated that priming of stem completion did not decline at a significantly faster rate than priming of fragment completion. Multiple fragments did not differ from either unique fragments or stems with respect to a decline in priming over a 48 hour delay. Multiple fragments differed from stems only in magnitude of priming at each time interval. Also, we see that the level of priming of stem completion at 48 hours equaled that of unique fragment completion at 6 minutes. Finding reliable priming at 48 hours is consistent with the findings of Roediger et al. (1992) and Craik et al. (1994). Experiment 5 Experiment 5 examined the effect of delay on priming of completion of the three types of completion cues and used a within-subjects design; priming of word completion was tested at two time intervals following list presentation, 6 minutes and 48 hours. Each participant received three rating lists and three completion tests, one with unique fragments, one with multiple fragments, and one with multiple stems. Method Participants Sixty undergraduates participated for extra credit. Materials The words and fragments used were taken from Experiment 4, but because this was a completely within-subjects design, 60 words were deleted and 88 added to the pool in order to