Word Naming Corpora

On this page you will find the experimental details for the two word naming corpora that we have collected.  The word naming data is freely available for anyone to down load.  The goal is to make available a large dataset to address a variety of questions regarding word naming performance.  In doing so, we have only two requests:

1.  Any use of this data should be acknowedged via the appropriate citations (of course).  For the naming data from young adults, the appropriate citation is:

Spieler D. H., & Balota, D. A. (1997).  Bringing computational models of word naming down to the item level. Psychological Science, 6, 411-416.

For the older adult data, the appropriate citation is:

Balota, D. A. & Spieler, D. H. (1998).  The utility of item level analyses in model evaluation:  A reply to Seidenberg &
Plaut (1998). Psychological Science

2.  Distribution of the naming data should be exclusively from this site.  In other words, do not distribute data second hand, rather direct requests these data to this site.  This is primarily to ensure accurate citations and to ensure that the corpora are distributed in their entirety rather than in parts.
Please direct any questions or comments to either of us:

David Balota
Department of Psychology
Washington University
St. Louis, MO 63130

Daniel Spieler
Department of Psychology
Jordan Hall, Bldg. 420
Stanford University
Stanford, CA 94305-2130

Below we report the Methods for the data collection.  Similar Methods for young subjects only are reported in the Spieler and Balota (1997) paper.


Thirty one younger adults were recruited from the undergraduate student population at Washington University.  Twenty nine older adults were recruited from the Aging and Development Subject Pool in the Department of Psychology at Washington University.  All individuals were paid $20 for their participation.  The young participants had a mean age of 22.6 years (SD = 5.0), 14.8 years of education (SD = 2.0) and scored 35.1 (SD = 2.7) on the Shipley vocabulary subtest (Western Psychological Services, 1967).


An IBM compatible Compudyne 486 computer was used to control the display of stimuli and to collect response latencies to the nearest ms.  The stimuli were displayed on a NEC4G 14 inch color VGA monitor in 40 column mode in white on a blue background.  The naming latency for each word was measured using a Gerbrands Model G1341T voice operated relay interfaced with the computer.


The words consisted of 2870 single syllable words appearing in the training corpora of the PMSP model and SM89 models.  These words ranged in frequency from 68246 to 0 counts per million according to Francis & Kucera (1982).  The words ranged from two to seven letters in length.

Notice: The word-frequency norms that are listed are based on a restricted data base version of the Francis & Kucera (1982) norms.  Although the listed estimates are highly correlated with the estimates from the full data base (.97 based on log frequency values), there are some deviations.  The full Kucera & Francis (1967) listing should be available in the data base by 9/15/98.  Please contact spieler@psych.stanford.edu, if you have any questions.


Each individual participated in two separate experimental sessions.  In each session, participants named 1435 words.  Words were presented in a different random order for each participant.  At the beginning of each of the two experimental sessions, individuals were seated in front of the computer and given the instructions for the experiment.  Participants were told that they would be shown single words at the center of the computer screen and that their task was to name the words aloud as quickly and as accurately as possible.  They were told to avoid making any extraneous noises which might trigger the voicekey and they were also told not to precede any of their responses with vocalized pauses such as “um” or “err”.  Participants were told that some of the words were very common while others were quite rare.  Each trial consisted of the following sequence of events:  a) a fixation consisting of three plus signs (“+ + +”) appeared in the center of the computer screen for 400 ms, b)  the screen went blank for 200 ms, c) the word appeared at the position of the fixation and remained on the screen until 200 ms after the initial triggering of the voicekey.  After each naming response, participants pressed a button on a mouse to go on to the next word.  If there was an error or an extraneous sound triggered the voicekey, participants were told to press the right button on the mouse.  If everything appeared to have worked properly on that trial, subjects were told to press the left button on the mouse.  Pressing the mouse button initiated a 1200 ms intertrial interval.
Participants were given breaks after every 150 trials.  Two buffer trials consisting of filler words not appearing in the training corpora were inserted at the beginning of each block of trials.  In addition, at the beginning of each session, subjects were given 20 practice trials to familiarize them with the task.  Each experimental session lasted for approximately 60 minutes.


Response latencies for trials that participants marked as errors and response latencies faster than 200 ms and slower than 1500 ms were excluded from all analyses.  Also, items that fell more than 2.5 standard deviations beyond each subject’s mean response latency were also dropped from these analyses.  These criteria eliminated 4.8% of the observations in younger adults.   An identical screening method was also applied for the older adults with the exception that an upper limit on naming latency was increased to 2000 ms.  These criteria eliminated 4.9% of the naming responses in the older adults.  In addition, 10 words were eliminated because their colloquial connotations were such to make it unlikely that their naming latencies are meaningful.  Mean latencies were then computed for each item across subjects separately for each group.

Updated: July 22, 1998.