The authors have declared that no competing interests exist.
Conceived and designed the experiments: SBS WSH. Performed the experiments: SBS. Analyzed the data: SBS WSH. Wrote the paper: SBS WSH.
The results of two experiments by Horton (2007) show that speakers name a pictured object faster when in the presence of another person with whom the speaker has previously associated that object name. The first of those two experiments (Horton, 2007, Experiment 1) is the focus of the present research. This paper presents the results of three experiments designed to replicate and extend Horton's (2007) Experiment 1. The original findings were not replicated. These findings do not support the hypothesis that partner-specific memory associations facilitate object naming.
The use of language in conversation is shaped by knowledge of what information is and is not jointly shared by the conversational partners. During conversation, information that is mutually known between the conversational partners grows as conversational partners exchange information. This jointly shared information is known as
A central question regarding the use of common ground is how this information is stored in memory. According to the classic view, personal common ground is encoded in memory through episodic, diary-like representations of joint experience
Beyond these questions about representation, another critical question is
One finding in support of the association-based view of common ground was a result reported by Horton (2007)
The original goal of the present research was to replicate and extend the same-partner benefit in object naming, in order to examine the factors that modulate the use of association-based common ground. In what follows, we describe these efforts. The original experiment by Horton (2007) had 16 participants who each participated in three within-subjects conditions during object naming: same-partner (the object label was associated with the present partner), different-partner (the object label was associated with the other partner), and new pictures
Experiment 1 was designed to replicate Horton (2007, Experiment 1)
Experiments 1, 2a and 2b were approved by the University of Illinois at Urbana-Champaign Institutional Review Board. All participants provided written consent prior to participation.
Fourteen native English speaking participants from the student community at the University of Illinois participated in this experiment, in exchange for partial course credit or $8. This experiment was designed to run at approximately 80% power to detect the partner effect based on the results of Horton (2007)
At the beginning of the experiment, participants were introduced to two different research assistants (one male, one female) who served as partners during the task. Participants were informed that the partners would be switching in and out during the different phases of the experiment. As in Horton (2007)
During the first phase of the task, participants were seated in front of a computer screen. One of the two partners (Partner A) was seated next to the participant while Partner B waited outside. On each trial the computer provided a clue number, and the participant told this number to the experimenter, who read off the associated category clue from a worksheet, e.g., “a musical instrument”. The participant then clicked the mouse and the computer displayed the label for one category exemplar with spaces for some of the letters for five seconds, e.g., “B-NJ-”. During this time, the participant's task was to figure out what the word was (e.g., “BANJO”). After five seconds, the full word was displayed on the screen for five more seconds, at which point the trial ended. This phase of the task included 36 trials, including 16 target trials and 20 filler trials. After completing the 36 trials, Partners A and B switched places, and the participant completed the same task with the same category labels but with different exemplars. For example, the exemplar for Partner B for the category “a musical instrument” would have been “H-RP” (“HARP”).
Target stimulus labels were identical to those used in Horton (2007, Experiment 1)
Item | Category cue (phase 1) | Exemplar A | Exemplar B |
1 | an article of furniture | lamp | stool |
2 | a fruit | cherry | lemon |
3 | a weapon | arrow | rifle |
4 | a vegetable | onion | celery |
5 | a kitchen utensil | ladle | blender |
6 | a bird | ostrich | penguin |
7 | a four-footed animal | zebra | giraffe |
8 | an article of clothing | belt | gloves |
9 | an insect | grasshopper | caterpillar |
10 | a vehicle | motorcycle | helicopter |
11 | a type of ship or boat | yacht | canoe |
12 | a thing that women wear | lipstick | perfume |
13 | a type of footwear | socks | slippers |
14 | a gardener's tool | pitchfork | wheelbarrow |
15 | a carpenter's tool | pliers | wrench |
16 | a musical instrument | harp | banjo |
During the second phase of the task, participants were seated at the same computer and were asked to name a series of pictures as quickly as possible with its most commonly used label. Each trial began with a 500 ms fixation cross, followed by immediate presentation of the picture that was to be named. Once the participant named the picture, the experimenter (Partner A or B) hit a key to indicate whether the name was correct, after which the next trial began.
Stimuli were of three types: Pictures with labels that were familiar from Phase 1 of the test, such as a banjo (n = 32); pictures with novel labels that were not studied in Phase 1 (n = 32), and unrelated fillers (n = 120). Labels for the familiar targets and novel pictures were identical to those used by Horton (2007)
Each participant completed half of the naming trials (16 target, 16 novel control, and 60 filler) with one partner and then the other half of the naming trials (16 target, 16 novel control, and 60 filler) with the other partner. The order of the two partners was randomized separately for each participant. In addition, the order of the stimuli and the assignment of critical stimuli to partner (e.g., whether the “banjo” would be named with Partner A or Partner B) was randomized separately for each participant. The partner who was participating with the participant sat next to them in the testing room and advanced the trials while the other partner sat behind a closed door in another room.
During the third phase of the task, participants sat in the testing room alone while the two partners sat in the adjacent room with the door open. The participant viewed a series of words presented on the screen. For each test word, the participant answered the question “Is this one of the words from the guessing game?” by clicking on one of two response buttons (yes or no) at the bottom of the screen. If the participant responded “yes”, the computer asked the participant “Which partner was working with you in the guessing game when you had this word?”. The participant responded by clicking one of two buttons at the bottom of the screen that displayed the two partners' names (e.g., “Sally”, “John”).
Stimuli during the partner identification phase included all 32 target labels from Phase 1 that were subsequently named in Phase 2 (e.g., “banjo”), and all 40 of the filler labels from Phase 1. Each of these items should generate a “yes” judgment for having been studied during the guessing game. In addition, participants responded to labels for 40 of the 120 filler pictures from Phase 2 (which should be responded to with a “no” judgment since they did not appear in Phase 1), and 40 new words (which should also generate a “no” judgment).
Data analysis focused on the latency to name the target pictures in Phase 2, and partner identification in Phase 3. All audio recordings were transcribed and word onsets identified by hand in Praat
Naming accuracy was defined as participants using the anticipated name for the picture, and was high, 89% for target pictures and 90% for novel control pictures. Trials excluded from the analysis included cases where the recording was lost due to equipment failure (4% of the data), cases where the participant used the wrong label (10% of the data), and cases where the speaker used the right label, but was disfluent, e.g., “uh, banjo” (6% of the data). The remaining 721 trials (80% of all possible trials) were submitted to analysis. By convention, analyses are calculated separately using participant as the random variable (
Same-partner | Different-partner | Novel controls | Different - Same | New - Old | |
Experiment 1 | 846 | 849 | 964 | 3 ms (97) | 117 ms (104) |
Experiment 2a | 1004 | 977 | 1247 | −26 ms (217) | 256 ms (241) |
Experiment 2b | 1023 | 1016 | 1241 | −7 ms (138) | 221 ms (176) |
Mean differences are shown in the right-most columns along with the standard deviation of the difference (ms) in parenthesis.
The average latency to name target pictures was 117 ms faster than novel control pictures, an effect that was significant only by participants
Participants correctly responded that they had studied the label during Phase 1 of the task 94% of the time for filler labels from Phase 1, and 90% of the time for target labels from Phase 1. By contrast, incorrect “studied before” responses were only 7% for the labels associated with Phase 2 filler pictures, and 1% for novel labels. These high levels of accuracy show that even at the end of the task, participants had good memory for which items had been studied in Phase 1.
For all items that participants correctly indicated they had studied before, they further indicated which partner they had studied it with. Partner identification for Phase 1 filler labels was high—91% (Horton (2007) similarly reports partner identification rates of 85% for Phase 1 fillers
Horton (2007) reported that the observed speed-up in naming times for same-partner trials was not significantly correlated with each participant's ability to explicitly recall the Phase 1 partner-label pairings for either the Phase-1 fillers or the Phase-1 targets
The results of this experiment are inconsistent with the findings of Horton (2007)
Methodological differences between the original experiment and Experiment 1, however, may be in play. One difference between Horton's (2007) experimental design
In what follows, we present the results of Experiments 2a–2b, which use the identical stimuli as Horton (2007, Experiment 1)
Experiment 2a was similar to Experiment 1; thus only differences in the experimental design are noted here. As before, our first step in the larger goal of investigating the role of partner-associations in perspective-taking was to replicate Horton (2007, Experiment 1)
Forty-nine native English-speaking participants from the student community at the University of Illinois participated in this experiment, in exchange for partial course credit or $8. This is three times the number of subjects run in Horton (2007, Experiment 1)
The basic procedure was identical to Experiment 1 with the exception that participants completed Phases 2 and 3 in a different room than Phase 1.
The materials, including all target and filler labels and all picture stimuli were identical to those used in Horton (2007, Experiment 1)
Naming accuracy was defined as participants using the anticipated name for the picture and was high, 92% for target pictures and 75% for novel control pictures. Trials excluded from the analysis included cases where the recording was lost due to equipment failure (0.2% of the data), cases where the participant used the wrong label for the picture (16% of the data), and cases where the speaker used the right label, but was disfluent (10% of the data). The remaining 2303 trials (73% of all possible trials) were submitted to analysis. For the analysis of the label familiarity effect, by-participant means were based on between 11–32 observations; by-item means were based on between 10–49 observations. For the analysis of the partner effect, by-participant means were based on between 7–18 observations; by-item means were based on between 17–26 observations. Average naming latencies for Experiment 2a are shown in
The average latency to name target pictures was 256 ms faster than novel control pictures,
Participants correctly responded that they had studied the label during Phase 1 of the task 88% of the time for fillers from Phase 1, and 90% of the time for targets from Phase 1; by contrast incorrect “studied before” responses were only 9% for Phase 2 fillers, and 1% for novel labels. For all items with a correct “studied before” response, accurate identification of the Phase 1 partner for Phase 1 fillers was 84%, and for Phase 1 targets it was 77%. Partner identification rates for both types of stimuli were well above a chance level of 50% (single-sample
As in Experiment 1, correlations between explicit partner identification and the difference in naming times between same-partner and different-partner trials were calculated. The correlation between the naming-time difference scores and partner memory for Phase-1
Despite using the identical stimuli and procedure, and running three times as many participants as the original experiment, the results of Experiment 2a are inconsistent with those of Horton (2007)
Experiment 2b was a conceptual replication of Horton (2007)
Participants were 48 native English speaking participants from the student community at the University of Illinois, who participated in exchange for partial course credit or $8. No participant had participated in Experiments 1 or 2a, and participants were randomly assigned to participate in Experiment 2a or 2b. Five additional participants were run but not included in the analysis due to problems with the audio recordings.
Experiment 2b was identical to Experiment 2a in every respect with the exception that rather than human partners, participants played the game with two different dolls, a red inflatable dinosaur, “Dr. Learnasaurus” and a Raggedy Anne doll, “Raggedy Anne”. The dolls were distinct in appearance and quite large; Dr. Learnasaurus was about 4′ tall and Raggedy Anne was about 2.5′ tall (
Naming accuracy was high, 90% for target pictures and 73% for novel control pictures. Trials excluded from the analysis included cases where the recording was lost due to equipment failure (0.2% of the data), cases where the participant used the wrong label (18% of the data), and cases where the speaker used the right label, but was disfluent (8% of the data). The remaining 2260 trials (74% of all possible trials) were submitted to analysis (
The average latency to name target pictures was 221 ms faster than novel control pictures,
Participants correctly responded that they had studied the label during Phase 1 of the task 89% of the time for fillers from Phase 1, and 93% of the time for targets from Phase 1; by contrast incorrect “studied before” responses were only 6% for Phase 2 fillers, and 0.8% for novel labels.
For items with a “studied before” response, correct identification of the partner (doll) for Phase 1 filler labels was 78%; for Phase 1 targets participants correctly identified the Phase-1 partner 73% of the time. Partner identification rates for both types of stimuli were well above a chance level of 50% (single-sample
The correlation between Phase-1 partner recall and the difference in naming times between the same-partner and different-partner conditions was calculated as before. The difference in naming times between same-partner and different-partner trials was not significantly correlated with partner memory for Phase-1 fillers (
This experiment was designed as a conceptual replication of the original study by Horton (2007)
The results of two experiments fail to replicate the primary finding of Horton (2007, Experiment 1)
The fact that none of the experiments observed a significant effect of partner associations on picture naming, despite being designed to run at high power, casts doubt on the notion that partner associations, at least as instantiated in this experimental paradigm, can reliably facilitate processes such as lexical access. In particular, Experiment 2a, which was designed as a direct replication, was run at 99% power, and provides strong evidence against the effect, if real, being as large as originally estimated.
In this context, it is important to note that the original study by Horton (2007) also reported a second experiment that was
Experiment 2 from Horton (2007) involved a within-subjects design with 24 participants, and the effect size for the critical same-partner vs. different-partner comparison can be estimated at
In what follows, we discuss the implication of the present null findings for theories of common ground. As far back as Clark and Marshall's account of definite reference and mutual knowledge
In the context of this memory-based account, Horton (2007)
The current findings, though, raise questions about the possible role that partner-specific memory associations might have in shaping speakers' access to relevant lexical information. In doing so, these results are reminiscent of other findings concerning the role of incidental environmental contexts, such as testing rooms and experimenters, on memory encoding and retrieval (see
An open question, then, is whether partner-specific memory associations could, in fact, reliably speed lexical access in a different sort of experimental paradigm. In the present research, the lack of partner context effects may be related to the fact that the partner was incidental to the task (see discussion in
Aside from issues related to the specific paradigm, there may be the temptation to view the current (null) results as reason to question broader claims about the need to better understand the basic memory processes involved in how speakers and listeners manage common ground more generally. We wish to emphasize, however, that there is a growing body of evidence that supports the notion that mechanisms of memory encoding and retrieval in conversational contexts play an important role in shaping language use. Much of this work is consistent with the claim that access to partner-specific memory representations can constrain language production and comprehension in ways relevant for common ground
Additionally, recent work with memory-impaired patients suggests that multiple memory systems may be involved in the maintenance of common ground
In conclusion, the present research fails to replicate the finding by Horton (2007, Experiment 1) that partner-specific associations influence object naming
Thanks to Daniel Simons for helpful discussions regarding this research, and to Richard Gerrig for providing many helpful comments on a previous version of this manuscript. Thanks to Laurel Brehm for making Dr. Learnasaurus available for Experiment 2b.