Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2008 Sep 26.
Published in final edited form as: Mem Cognit. 2007 Jan;35(1):166–175. doi: 10.3758/bf03195952

The spread of the phonological neighborhood influences spoken word recognition

Michael S Vitevitch 1
PMCID: PMC2553701  NIHMSID: NIHMS66409  PMID: 17533890

Abstract

In three experiments, the processing of words that had the same overall number of neighbors but varied in the spread of the neighborhood (i.e., the number of individual phonemes that could be changed to form real words) was examined. In an auditory lexical decision task, a naming task, and a same–different task, words in which changes at only two phoneme positions formed neighbors were responded to more quickly than words in which changes at all three phoneme positions formed neighbors. Additional analyses ruled out an account based on the computationally derived uniqueness points of the words. Although previous studies (e.g., Luce & Pisoni, 1998) have shown that the number of phonological neighbors influences spoken word recognition, the present results show that the nature of the relationship of the neighbors to the target word—as measured by the spread of the neighborhood—also influences spoken word recognition. The implications of this result for models of spoken word recognition are discussed.

In research on spoken language processing, neighborhood density refers to the number of words that sound similar to a given word: Words with many neighbors, or similar words, are said to have dense neighborhoods, whereas words with few neighbors are said to have sparse neighborhoods. Several studies in English have demonstrated that neighborhood density influences various aspects of spoken language processing, including lexical acquisition (e.g., Storkel, 2002, 2004), speech production (e.g., Vitevitch, 1997, 2002b; Vitevitch & Sommers, 2003), and spoken word recognition (Luce & Pisoni, 1998; see also Vitevitch & Rodríguez, 2005, for a discussion of the influence of neighborhood density on spoken word recognition in Spanish).

In several laboratory-based spoken word recognition tasks, Luce and Pisoni (1998) demonstrated that English words with sparse neighborhoods are responded to more quickly and accurately than those with dense neighborhoods, suggesting that multiple word forms are activated and compete with each other during spoken word recognition. Words with large numbers of phonological neighbors (i.e., dense neighborhoods) are subject to greater competition and therefore recognized more slowly and less accurately than words with few phonological neighbors (i.e., sparse neighborhoods).

Vitevitch (2002c) observed a similar processing disadvantage for words with dense neighborhoods in an analysis of a corpus containing speech perception errors, known as “slips of the ear,” that were collected via naturalistic observation. An example of a slip of the ear is erroneously hearing the correctly produced question “What’s wrong with her bike?” as “What’s wrong with her back?” (Bond, 1999). In analyzing the misperceived words in Bond’s corpus, Vitevitch (2002c) found that slips of the ear tended to occur in words with dense phonological neighborhoods, further suggesting that multiple word forms are activated and compete during spoken word recognition.

The previously discussed studies clearly demonstrate that the number of phonologically related word forms that are activated influences spoken word recognition: Words with few neighbors are recognized more quickly and more accurately than words with many neighbors in English. Now, consider two words with the same number of phonological neighbors. Does some other factor, such as the distribution of the neighbors in the lexical neighborhood, influence the speed and accuracy of spoken word recognition? By way of illustration, consider the words mop (/mⱭp/) and mob (/mⱭb/). When a single phoneme substitutes any of the phonemes in the word mop, phonological neighbors are formed (e.g., hop, map, mock). However, similar substitutions in the word mob produce phonological neighbors at only two of the three phoneme positions (e.g., rob, m*b, mock); no real word in English is formed when the phoneme in the medial position of the word mob is substituted. Note that each word has the same total number of phonological neighbors, but that the number of phoneme positions in the word that produce a neighbor differs between the two words.

To investigate the possible influence of the distribution of similar-sounding neighbors in the phonological neighborhood on spoken word recognition, a phonological analogue of a metric used in studies of visual word recognition—the spread of a neighborhood, or the P-metric (Andrews, 1997; Johnson & Pugh, 1994; Pugh, Rexer, Peter, & Katz, 1994)—was manipulated in several behavioral tasks. Spread refers to the number of phoneme positions (or letter positions, as in Johnson & Pugh, 1994) in a word that can be changed to form a neighbor. In the examples above, the word mop has a P-metric value of 3 (P = 3) because changes at three phoneme positions produce phonological neighbors, whereas the word mob has a P-metric value of 2 (P = 2) because changes at two phoneme positions produce phonological neighbors. If the distribution of phonological neighbors in the lexical neighborhood influences spoken word recognition, then differences should be observed in terms of the speed and accuracy with which these two types of words (P = 2 vs. P = 3) are responded to. The same stimuli were presented in three different laboratory-based tasks to evaluate the influence of neighborhood spread on the speed and accuracy of spoken word recognition.

EXPERIMENT 1

To examine how the spread of phonological neighbors in the similarity neighborhood might affect spoken word recognition, a lexical decision task was used. In the lexical decision task, participants are presented with a stimulus item and must decide as quickly and accurately as possible if that item is a real word in English or a nonsense word. In the present experiment, the stimuli were presented auditorily rather than visually as in Johnson and Pugh (1994) and varied in phonological P rather than orthographic P. The stimuli that the participants heard consisted of three-phoneme words, with a consonant–vowel–consonant (CVC) syllable structure, that had the same number of phonological neighbors but differed in how those neighbors were spread about the neighborhood. For half of the words, P = 2, meaning that a change in either of two phoneme positions produced a neighbor; for the remaining words, P = 3, meaning that a change in any of all three phonemes in the word produced a neighbor.

Method

Participants

Forty right-handed native English speakers from the pool of introductory psychology students at the University of Kansas participated in partial fulfillment of a course requirement. None of the participants reported a history of speech or hearing problems, and none of them participated in either of the other experiments reported in the present study.

Stimuli

Ninety-two CVC words were used as stimuli in the experiment (see Appendix A). The stimuli were divided into two sets of 46 words each. One set contained words that formed a neighbor when a single phoneme was substituted (e.g., Landauer & Streeter, 1973; Luce & Pisoni, 1998) at any of the three phoneme positions of the word (P = 3). The other set contained words that formed a neighbor when a single phoneme substitution could be made at one of only two phoneme positions of the word to form a neighbor (P = 2). Words for which P = 1 were not examined because of the paucity of words in this category.

APPENDIX A.

Stimulus Items Used in Experiments 1–3, and Examples of Their Neighbors

Phonological P = 2
Phonological P = 3
Neighbors
Neighbors
Stimulus P1 P2 P3 Stimulus P1 P2 P3
chalk Hawk check * cheese tease choose cheap
chill Fill * chip chess guess chase check
church search * chirp chose rose cheese chore
deaf Chef * dead curb verb cub curl
dodge lodge * dot dish wish dash dip
doll * dull dock dog log dug dawn
dose * dice dove doubt shout dirt down
fetch Retch * fed dove cove dive dome
fish Wish * fib firm term fame fern
five Dive * fine foam home firm phone
foul Howl feel * fog hog fig fall
gab Cab * gag foot soot fight full
geese peace gas * gauze pause gaze gone
good Wood guide * germ term gem jerk
gouge gauge * gown gown down gun gouge
hedge wedge * head guide wide god guise
hen den * hem guise size gauze guide
jade wade * jail hive dive heave hike
joke poke jerk * hog dog hug haul
judge fudge * jug jab cab job jack
king ring * kick jerk work joke germ
league * log lease ledge hedge lodge leg
leash * lash leap lobe robe lob load
loaf * life lobe lodge dodge ledge lock
lull hull * lush lurch church leach learn
mesh * mash met mop hop map mock
mob lob * mock mouse house mace mouth
moth * mouth moss mouth south moth mouse
noise poise nose * neck wreck knock net
noun down nun * niece piece nurse need
nudge fudge * nut nurse purse noose nerve
palm psalm * pop pause cause poise pawn
path math * pad peg beg pig pen
poise noise pause * pouch vouch pitch pout
sash rash * sack sauce toss cease sought
shawl wall shell * shave wave shove shape
sheath wreath * sheep shop top ship shot
shine dine shun * shove love shave shun
thought fought * thong theme beam thumb thief
tube * tub tune toad road tide tote
vague * vogue vase van man vein vat
verb curb * verse verse terse voice verb
verge surge * verb vote boat vet vogue
wing sing * wish weave leave wove weep
womb tomb worm * wedge hedge wage well
worse curse * worth worth birth with word
Open in a new tab

Note—P1, change at first phoneme position; P2, change at second phoneme position; P3, change at third phoneme position.

*

No English word in the corpus (see Nusbaum, Pisoni, & Davis, 1984) is formed by changing the stimulus phoneme at this position.

Although the two sets of words differed in the number of phonemes that could be changed to form a neighbor, they did not differ [all Fs(1,90) < 1] in the overall number of neighbors (i.e., neighborhood density), word familiarity (Nusbaum, Pisoni, & Davis, 1984), word frequency (Kuèera & Francis, 1967), the frequency with which the neighbors occurred (i.e., neighborhood frequency; Kuèera & Francis, 1967), or phonotactic probability (Vitevitch & Luce, 1998, 1999, 2005). Note that information related to the familiarity, frequency, neighborhood density, and neighborhood frequency for each word can be obtained from a Web-based interface maintained by Mitchell Sommers at Washington University (128.252.27.56/neighborhood/Home.asp). Information related to the phonotactic probability of each word can be obtained from a Web-based interface (www.people.ku.edu/~mvitevit/PhonoProbHome.html) described in Vitevitch and Luce (2004). The mean values for these characteristics for each set of words are presented in Table 1. The same number of initial segments appeared in each condition.

Table 1.

Mean Values (and Standard Deviations) for the Lexical Characteristics of the Stimuli

P = 2
P = 3
Characteristic M SD M SD
Frequency of occurrence (log) 1.000 0.760 1.100 0.620
Familiarity* 6.860 0.280 6.880 0.200
Neighborhood density† 8.700 3.500 9.200 1.900
Neighborhood frequency 1.230 0.370 1.240 0.310
Phonotactic probability
 Sum of phones 0.116 0.050 0.113 0.040
 Sum of biphones 0.004 0.004 0.004 0.003
Open in a new tab

Note—No differences were statistically significant [all Fs(1,90) < 1].

*

Based on a 7-point scale.

†

A word was considered a neighbor if a substitution of a phoneme in the target word formed that word and it appeared in the computer-readable phonemically transcribed Webster’s Pocket Dictionary (Nusbaum et al., 1984). This method of determining neighborhood size was consistent with the method employed by Johnson and Pugh (1994), with the exception that phonemes rather than letters were substituted (i.e., the N-metric commonly attributed to Coltheart, Davelaar, Jonasson, & Besner, 1977).

In addition, onset density did not differ between the two conditions of words [F(1,90) < 1]. Onset density refers to the proportion of neighbors that share the same initial phoneme as the target word (Vitevitch, 2002a). For words for which P = 2, the mean proportion of neighbors that shared the same initial phoneme as the target word was .60, whereas for words for which P = 3 the mean proportion was .59.

Although the stimuli were presented auditorily rather than visually (cf. Johnson & Pugh, 1994), the two conditions of words did not differ in the number of letters comprising the words [F(1,90) < 1]. Words for which P = 2 had a mean of 4.5 letters per word (SD = 0.81), and words for which P = 3 had a mean of 4.4 letters per word (SD = 0.77). The two conditions of words also did not differ in the number of orthographic neighbors [F(1,90) < 1]. Words for which P = 2 had a mean of 4.9 orthographic neighbors (SD = 4.2), and words for which P = 3 had a mean of 5.5 orthographic neighbors (SD = 4.3), on the basis of calculations from the N-Watch program described by Davis (2005).

The stimuli were spoken in isolation and recorded by the author in an IAC sound-attenuated booth on high-quality audio-recording equipment. The stimuli were digitized at a sampling rate of 20 kHz using a 16-bit analog-to-digital converter. All words were edited into individual digital files and stored on a computer disk for later presentation. Stimuli in the P = 2 condition had a mean file duration of 863msec (SD = 105), and stimuli in the P = 3 condition had a mean file duration of 851msec (SD = 100); this difference was not statistically significant [F(1,90) =0.30, p > .5]. Ninety-two nonsense words were also used (see Appendix B). The method used to create nonwords in previous studies (e.g., Vitevitch, 2002a) was used in the present experiment: The last phonemes of words not found in the stimulus set were changed to create the nonwords that were used. Only the last phoneme was changed to increase the likelihood that the participants would listen to the entire stimulus item before making a response. The nonwords were recorded and treated in the same manner as the real word stimuli.

APPENDIX B.

Nonwords (Transcribed in IPA) Used in Experiment 1

bæf dεʒ læθ pin
bæv daɪt lⱭd pɪp
bæb dʌp liθ pɪ∫
bæz fod luθ pɪv
bæp faɪm laɪθ pob
bɔn hæb meθ pod
bɔp hæð meg pot
bef hεb mep pʌv
bεf hεk mig ræb
bεdʒ hɪf mʌp rⱭθ
bεp hɪb nⱭp rʌp
bεv hɪʒ nɪs rʌz
bib hɪdʒ naɪp sæz
biθ hɪ∫ pæb sεk
big kæk pæg sib
bɪk kⱭdʒ pæv ∫ɪd
bɪθ ked peg siv
bog keb pεp sʌt
bʌp kɪf peθ tæt
bʌθ kɪdʒ pid tedʒ
dⱭz kɪθ pidʒ tev
deb kɪz pɪf tɪ∫
dεdʒ kof pig taɪv
Open in a new tab

Procedure

The participants were tested in groups of 4 or fewer. Each participant was seated in a booth equipped with an iMac running PsyScope 1.2.2 (Cohen, MacWhinney, Flatt, & Provost, 1993) that controlled stimulus randomization and presentation, a set of Beyerdynamic DT-100 headphones, and a PsyScope buttonbox with a dedicated timing board. Each trial proceeded as follows: The word READY appeared in the center of the computer screen for 500 msec to indicate the beginning of the trial. The participants were then presented with one of the randomly selected stimuli at a comfortable listening level over the headphones. The left button on the response box was labeled NONWORD, and the right button (i.e., that for the dominant hands of the participants) was labeled WORD. The participants responded as quickly and accurately as possible by pushing the appropriately labeled button. Reaction time was measured from the onset of the stimulus file to the onset of the response. Prior to the experimental trials, each participant received 10 practice trials. These trials were used to familiarize the participants with the task and were not included in the final data analysis.

Results and Discussion

Separate ANOVAs were performed on response latency and accuracy rates with participants and items treated as random factors. Although there is some debate about whether or not to treat stimulus items as a random factor in statistical analyses (Cohen, 1976; Hino & Lupker, 2000; Keppel, 1976; Raaijmakers, 2003; Raaijmakers, Schrijnemakers, & Gremmen, 1999; Smith, 1976; Wike & Church, 1976), it is the current practice in psycholinguistic research to conduct both types of analyses. For consistency with this convention, both types of analyses will be reported; however, the discussion and interpretation of the results will be based only on the analyses in which participants were treated as a random factor. Also, estimates of effect size will be conducted only on the analyses in which participants were treated as a random factor.

Only correct responses to the stimulus items within 2SD s of the mean response time were used in the analyses of response latency. A significant difference in response latencies was found in the lexical decision task [F(1,39) = 39.76, p < .001] given that the participants responded more quickly to words for which P = 2 (M = 1,080msec, SD = 99) than to words for which P = 3 (M = 1,115 msec, SD = 95). The same pattern of results was obtained when stimulus items were treated as a random factor [F(1,90) =4.57, p < .05]. An estimate of effect size using Cohen’s d shows that this can be considered a medium-sized effect (d = 0.36).

No significant difference was found for the accuracy rate in the lexical decision task (both Fs < 1), suggesting that the participants did not sacrifice speed for accuracy in making their responses. The participants responded to words for which P = 2 with 90% accuracy (SD = 4.8) and to words for which P = 3 with 91% accuracy (SD = 5.0).

The results of the auditory lexical decision task showed that words for which P = 2 were responded to more quickly than words for which P = 3, even though these two sets of words had comparable numbers of neighbors overall. These results extend the work of Johnson and Pugh (1994), who examined neighborhood spread in visual word recognition, to the auditory domain. Recall that in the present experiment the number of phoneme positions, rather than the number of letter positions, was manipulated, and an auditory lexical decision task was employed rather than a visual lexical decision task. To further examine the influence of neighborhood spread on spoken word recognition, an auditory naming task was performed in Experiment 2, and an auditory same–different task was performed in Experiment 3.

EXPERIMENT 2

In the present experiment, an auditory naming task was used to further examine how the spread of phonological neighbors in the similarity neighborhood might affect spoken word recognition. In the auditory naming task, a word is presented to participants over a set of headphones, and they must simply repeat the word as quickly and accurately as possible. This task, as well as the same–different task in Experiment3, was used to better generalize the results observed in Experiment 1 (and those of Johnson & Pugh, 1994), in which the lexical decision task was employed. Because every task used in laboratory settings has advantages and disadvantages, replication across a variety of tasks increases our confidence that the observed effect was not due to the assumptions of a particular task employed in a particular experiment. Furthermore, Wike and Church (1976) recommended replication as a means of generalizing results without resorting to statistical techniques that might be inappropriate, such as analyses that treat stimulus items as a random factor.

Method

Participants

Thirty native English speakers from the pool of introductory psychology students at the University of Kansas participated in partial fulfillment of a course requirement. None of the participants reported a history of speech or hearing problems, and none of them had participated in either of the other experiments reported in the present study.

Stimuli

The stimuli consisted of the same 92 words manipulated for neighborhood spread that were used as stimuli in Experiment 1.

Procedure

The participants were tested 1 at a time. Each participant was seated in a booth equipped with an iMac running PsyScope 1.2.2 (Cohen etal., 1993), which controlled stimulus randomization and presentation; a set of Beyerdynamic DT-109 headphones; and a PsyScope buttonbox with a dedicated timing board. Each trial proceeded as follows: The word READY appeared in the center of the computer screen for 500 msec to indicate the beginning of the trial. The participant was then presented with one of the randomly selected stimuli at a comfortable listening level over the headphones. Response latency was measured from the onset of the stimulus file to the onset of the participant’s response. When a response was made, the word READY appeared on the screen and the next trial began. Responses were also recorded on digital audio tape for later accuracy analyses. Prior to the experimental trials, each participant received 10 practice trials. None of the items used in the practice session was used in the experiment. The practice trials were used to familiarize the participants with the task, and the data collected from them were not included in the final analysis. The participants were instructed to respond as quickly and accurately as possible.

Results and Discussion

As in Experiment 1, separate ANOVAs were performed on response latency and accuracy rates with participants and items treated as random factors. Only correct responses within 2 SDs of the mean response time were used in the analyses of response latency. An accurate response was one in which each phonological segment in the verbal response made by a participant matched the segments in a phonological transcription of the stimulus word as judged by a trained speech scientist (see Vitevitch & Luce, 2005).

A significant difference in response latencies was found [F(1,29) =126.04, p < .001] given that the participants responded more quickly to words for which P = 2 (M = 1,018msec, SD = 144) than to words for which P = 3 (M = 1,056msec, SD = 140).1 The same pattern of results was observed when stimulus items were treated as a random factor [F(1,90) =6.78, p < .01]. An estimate of effect size using Cohen’s d shows that this can be considered an effect of small to medium size (d = 0.26).

No significant differences were found for the accuracy rates in the naming task (both Fs < 1), suggesting that the participants did not sacrifice speed for accuracy in making their responses. The participants responded to words for which P = 2 with 94% accuracy (SD = 4.2) and to words for which P = 3 with 95% accuracy (SD = 5.1).

The results of the auditory naming task are consistent with the results obtained in Experiment 1 using the auditory lexical decision task: Words for which P = 2 were responded to more quickly than words for which P = 3, even though the two sets of words had comparable numbers of neighbors overall. These results further extend the work of Johnson and Pugh (1994), who examined only neighborhood spread with the lexical decision task (and only in the visual modality). An auditory same–different task was performed in Experiment 3 to further generalize the results observed in Experiments 1 and 2.

EXPERIMENT 3

In the present experiment, an auditory same–different task was used to further examine how the spread of phonological neighbors in the similarity neighborhood might affect spoken word recognition. In the auditory same–different task, participants hear two words presented close together in time and must decide as quickly and accurately as possible whether the two words were the same (e.g., dog–dog) or different (e.g., dog–doll).

Method

Participants

Thirty-eight right-handed native English speakers from the pool of introductory psychology students at the University of Kansas participated in partial fulfillment of a course requirement. None of the participants reported a history of speech or hearing problems.

Stimuli

The stimuli consisted of the same 92 words manipulated for neighborhood spread that were used as stimuli in Experiments 1 and 2, and 184 additional English words that were recorded and edited in the same fashion as the other stimuli.

Procedure

The equipment used in Experiment 1 was also used in the present experiment. Each experimental trial proceeded as follows: The word READY appeared in the center of the computer screen for 500 msec to indicate the beginning of the trial. The participants were then presented with two of the spoken stimuli at a comfortable listening level. The interstimulus interval was 50 msec. Reaction times were measured from the onset of the second sound file in the pair to the buttonpress response. The participants were instructed to respond as quickly and accurately as possible on each trial. The buttonbox had the label DIFFERENT on the left button and the label SAME on the right button (the middle response button was deactivated). Half of the trials consisted of two presentations of the stimulus items (constituting “same” trials), and half consisted of nonmatching stimuli (constituting “different” trials). For the “different” stimulus pairs (listed in Appendix C), items with the same initial phoneme and (when possible) the same vowel were paired to increase the likelihood that the participants would listen to both words in the stimulus pair and base their decisions on both words rather than adopt a strategy of simply listening for the match (or mismatch) of the initial phonemes of each word in the pair. Each participant was allowed 10 practice trials prior to the experimental trials. These trials were used to familiarize the participants with the task and were not included in the final analysis.

APPENDIX C.

“Different” Stimulus pairs used in Experiment 3

bad/badge fool/food match/mass safe/save
bake/base fuzz/fun maze/main sane/same
batch/bat game/gaze met/mess sang/sake
beam/beach gate/gain moan/mole sat/sag
beige/bait gum/gun mood/moon scene/seal
bell/bed hack/hash mug/mud shape/share
birch/bird head/hem net/nerve shock/shot
bowl/boil heard/heap note/nose soup/suit
cat/can hole/hope patch/pack tag/tack
chip/chin hot/hop peach/peel talk/taught
code/comb hum/hut perch/perk tall/toss
coil/coin hype/height pipe/pike term/terse
core/cone kick/kin pub/puff tide/tight
cove/coat knife/nice pun/puck toll/tone
curve/curl knit/nick rage/race ton/tough
date/dame leaf/leak rash/rat tour/took
dial/dire lean/leap reef/reek tub/tuck
dill/dim lease/leave ride/ripe weak/weep
duck/dug less/leg rip/riff well/web
dull/done life/light roach/road wine/wipe
fame/fake load/loan roam/rope wing/whip
feet/feel make/mate run/rug wise/wife
fig/fin man/map sack/sad yell/yawn
Open in a new tab

Results and Discussion

As in Experiments 1 and 2, separate ANOVAs were performed on response latency and accuracy rates with participants and stimulus items treated as random factors. Only correct responses within 2SD s of the mean response time were used in the analyses of response latency. A significant difference in response latencies was found [F(1,37) =35.288, p < .001] given that the participants responded “same” more quickly to words for which P = 2 (M = 819 msec, SD = 87) than to words for which P = 3 (M = 859 msec, SD = 92). The same pattern of results was obtained when stimulus items were treated as a random factor [F(1,90) =7.43, p < .01]. An estimate of effect size using Cohen’s d shows that this can be considered a medium-sized effect (d = 0.44).

No significant differences were found for the accuracy rates in the auditory same–different matching task (both Fs < 1), suggesting that the participants did not sacrifice speed for accuracy in making their responses. The participants responded to both types of words with 96% accuracy (SD = 4 in both cases).

The results of the auditory same–different task in the present experiment are consistent with the results of Experiments1 and 2: Words for which P = 2 were responded to more quickly than words for which P = 3, even though the two sets of words had comparable numbers of neighbors overall. The results of the present set of experiments further generalize the work of Johnson and Pugh (1994), who examined only neighborhood spread with the lexical decision task, and only in the visual modality.

GENERAL DISCUSSION

Previous studies demonstrated that the number of words in the phonological neighborhood influences the speed and accuracy of spoken word recognition. In English, words with few neighbors (i.e., those with sparse phonological neighborhoods) are recognized more quickly and accurately than words with many neighbors (i.e., those with dense phonological neighborhoods) (Luce & Pisoni, 1998; Vitevitch, 2002a; cf. Vitevitch & Rodríguez, 2005). The results of our Experiments1–3 clearly demonstrate that the spread of the neighborhood also influences spoken word recognition. Specifically, words with two phoneme positions that can be changed to form a neighbor (P = 2) were responded to more quickly than words with three phoneme positions that can be changed to form a neighbor (P = 3), despite their having comparable numbers of neighbors overall. Although current models of spoken word recognition can account for processing differences that result from different numbers of competitors (see, e.g., Auer & Luce, 2005; Luce & Pisoni, 1998; McClelland & Elman, 1986; Norris, 1994), it is not clear whether or not each of these models can account for the results of the present set of experiments, in which words with equal numbers of neighbors were differentially responded to as a function of the spread of the neighborhood.

We shall first consider cohort theory because Johnson and Pugh accounted for their findings with a model of visual word recognition based on the assumptions of the cohort theory of spoken word recognition proposed by Marslen-Wilson and Welsh (1978). Recall that Marslen-Wilson and Welsh suggested that acoustic–phonetic information activates a set of lexical candidates (i.e., the cohort) that is consistent with the input. As more of the word is heard, additional acoustic–phonetic information accumulates. Candidates that are no longer consistent with the additional input drop out of the cohort. Once sufficient information has accrued to distinguish the input from all other words in the cohort of partially activated candidates, word recognition is said to occur. Using a gating task, in which listeners attempt to identify the stimulus word as increasingly larger portions of the word are presented auditorily, Grosjean (1980) found that words in which this recognition point occurred early were correctly identified sooner (i.e., with fewer “gates”) than words in which the recognition point occurred later. Thus, one might hypothesize that, in the present set of experiments, words for which P = 2 had earlier recognition points than words for which P = 3, thereby accounting for the difference in response times in all three experiments.

To examine the possibility that in the present set of experiments words for which P = 2 had earlier recognition points than words for which P = 3, the recognition points, or the computationally derived uniqueness points, of the stimulus items were examined. Note that use of the terms isolation point, uniqueness point, and recognition point is not consistent in the literature (cf. Bölte & Uhe, 2004; Grosjean, 1996; Radeau & Morais, 1990). In the present context, the term uniqueness point will be used to refer to the point in a word at which it becomes unique from all other words in the lexicon, as assessed via computational search through a corpus of English words (i.e., the same corpus used to estimate phonological neighborhood density in the present study). Uniqueness points differ from recognition or isolation points, which are empirically derived via the gating task (see, e.g., Grosjean, 1980, 1996). Note furthermore that there is some debate about the psychological validity of such constructs in the processing of fluent speech (cf. Bölte & Uhe, 2004, and Radeau, Morais, Mousty, & Bertelson, 2000).

In an analysis of computationally derived uniqueness points, Luce (1986) found that the uniqueness point for monosyllabic words in English—such as those used in the present set of experiments—typically occurred after the end of the word. That is, the sound sequences that comprise many monosyllabic words are also part of longer words (e.g., car–card, cat–cattle–catalog), which means that listeners need to hear the beginning of the next word before they can be sure they have reached the end of the present monosyllabic word and correctly recognize it.

In the stimuli used in the present experiments, an analysis of the uniqueness points of the stimulus items showed that words for which P = 2 had a mean uniqueness point at 3.6 phonemes (SD = 0.6) and words for which P = 3 had a mean uniqueness point at 3.7 phonemes (SD = 0.5); this difference was not statistically significant [F(1,90) =1.95, p = .17]. Recall that the stimuli used in the present experiments consisted of monosyllabic words that were three phonemes long. Uniqueness points greater than three indicate that the three-phoneme-long monosyllabic stimulus items did not diverge from other words in the lexicon until after the offset of the word, which is consistent with the results reported by Luce (1986) for monosyllabic words. Furthermore, words for which P = 2 did not diverge from other words in the lexicon sooner than did words for which P = 3, suggesting that differences in the uniqueness points of the stimulus words cannot account for the results of the present set of experiments. Although Johnson and Pugh (1994) interpreted their results in terms of a cohort-based model, it is unlikely that such an account can explain the results of the present set of experiments.

Rather than being a proxy measure for the uniqueness point,2 the spread of the neighborhood, or P-metric, seems to measure some other lexical construct. That is, P measures the distribution of phonological neighbors in the similarity neighborhood. As was demonstrated in Experiments1–3, spoken word recognition is significantly affected by the distribution of phonological neighbors in the similarity neighborhood. Specifically, words with neighbors that are “packed” into fewer regions of the neighborhood are responded to more quickly than words with neighbors spread throughout the neighborhood. Given the emphasis that cohort theory places on the initial portion of word forms and the clear evidence (provided in the present set of experiments) that neighbors located in other parts of the word influence processing, it is unlikely that cohort theory can account for the present results.

Given that TRACE (McClelland & Elman, 1986)—a computational model that accounts for numerous effects observed in studies of spoken word recognition—incorporates several assumptions of cohort theory into its design, it is logical to consider this model next. As McClelland and Elman discussed, potential lexical candidates in TRACE, as in cohort theory, are activated as the acoustic–phonetic input accrues over time. Thus, as in cohort theory, the initial portion of the word is important for activating potential lexical candidates in TRACE. As described above, relying on the initial portion of the word proved problematic for cohort theory in accounting for the present results, and, thus, one might expect that TRACE would also fail to account for these results.

In contrast to the earlier cohort theory, however, other parts of the word can also partially activate lexical candidates, enabling TRACE to correctly retrieve a lexical item despite a distortion in the beginning of the word (e.g., recognizing dwibble as the word dribble). Indeed, Allopenna, Magnuson, and Tanenhaus (1998) used an eyetracking task to provide evidence that the initial parts of a word (i.e., the cohort) and the rhyme portion of a word activated lexical competitors. Furthermore, the time course and probabilities of eye movements obtained by Allopenna etal. closely corresponded to the response probabilities derived from simulations of TRACE. Given the fact that other portions of the acoustic–phonetic input can continuously activate lexical candidates in TRACE, it is possible that this model might be able to account for the effects observed in the present set of experiments—that is, TRACE might be able to account not only for the influence of the number of lexical competitors on processing, but also for the influence that the location of those neighbors in the neighborhood has on processing that has been demonstrated in the present study. The previous statement should be interpreted cautiously, however, given the inherent difficulty of predicting exactly how complex computational models might perform without examining an actual simulation (Lewandowsky, 1993).

In response to the interactive nature of TRACE, Norris (1994) developed Shortlist, a feedforward model of spoken word recognition (see also MERGE, the feed-forward model of speech recognition; Norris, McQueen, & Cutler, 2000). Although Shortlist differs from TRACE with regard to the existence of feedback between levels, the models are similar in that initial and subsequent input influence lexical retrieval in both models. Indeed, Norris demonstrated that Shortlist correctly activates the word cigarette (/sɪgǝrεt/) even when it is presented with input that contains a mispronunciation in the initial portion of the word (e.g.,/∫ɪgǝrεt/). Thus, despite the noninteractive architecture of Shortlist, it too might be able to account for the present set of results. Again, however, caution should be exercised when the computational simulation is not actually performed.

Luce and Pisoni (1998) described another model of spoken word recognition—the neighborhood activation model (NAM)—which accounts for the influence of the intelligibility of the stimulus words, the frequency of occurrence of the stimulus words, and the number of lexical competitors (as well as the frequency of occurrence of the competitors) on processing. In assessing the confusability of the stimulus word and its competitors, NAM places equal weight on each phoneme (regardless of whether it is a consonant or a vowel) and on the position of each phoneme (regardless of whether the phoneme occurs in the onset, the nucleus, or the coda position) in a word. Given that all phoneme positions are treated equally in NAM, it is unclear whether NAM would be able to account for the results of the present experiment (or for those of Vitevitch, 2002a), which demonstrate that some phoneme positions do influence spoken word recognition more than others. As the present experiments demonstrate, phoneme positions that form a neighbor influence spoken word recognition differently than do those that do not form a neighbor.

Although the original NAM might have problems accounting for the results of the present experiments, a more recent connectionist instantiation of NAM, dubbed PARSYN (Auer & Luce, 2005), might be able to account for the present findings (as well as for those of Vitevitch, 2002a). In PARSYN, paradigmatic and syntagmatic representations are activated (hence the name) as a spoken word is presented. Paradigmatic states refer to the number of alternatives active at a given point in time, whereas syntagmatic states refer to patterns that occur over time. In the case of the word cat, the paradigmatic representations activated would include the initial phoneme/k/as well as other related phonemes, such as/b/(a stop that differs from/k/in place of articulation and voicing) and/g/(a stop that differs from/k/in voicing). Syntagmatic states that would be highly activated in the case of/kæt/would include representations of the pattern of sounds/kæ/and/æt/, whereas related but less common sequences of segments (such as/ki/or/æv/) would be less active. By considering the dynamic interaction of paradigmatic and syntagmatic states, PARSYN can account for many aspects of spoken word recognition (Auer & Luce, 2005; see also Luce, Goldinger, Auer, & Vitevitch, 2000). Given that PARSYN takes the number of competitors (i.e., paradigmatic information) as well as the distribution of those representations over time (i.e., syntagmatic information, which would convey some information about phoneme position), it is possible that PARSYN could account for the results observed in the present set of experiments. However, as was stated in the discussions of TRACE and Shortlist, we must be cautious in predicting exactly how a complex computational model might perform without examining an actual simulation (Lewandowsky, 1993).

Previous research on spoken word recognition (as well as speech production and word learning) has focused much attention on the influence that the number of phonological neighbors has on processing. The present set of studies (see also Vitevitch, 2002a) demonstrates that the distribution of neighbors in the neighborhood also influences processing. Models of spoken word recognition must account not only for the influence of the number of competitors on processing, but—in the absence of a difference in the number of competitors—also for the influence of the location of competitors on processing. Thus, the number of neighbors, as well as the relationship among the neighbors, appears to provide an important, but different, kind of constraint on spoken word recognition.

Acknowledgments

This research was supported in part by Grants NIDCD R03 DC 04259 and NIDCD R01 DC 006472 from the National Institutes of Health to the University of Kansas through the Schiefelbusch Institute for Life Span Studies; National Institute of Child Health and Human Development Grant P30 HD002528 from the Mental Retardation and Developmental Disabilities Research Center; and Grant NIDCD P30 DC 005803 from the Center for Biobehavioral Neurosciences in Communication Disorders. I thank Shinying Chu, David Levine, and Thu Vo for their assistance in data collection, and Steven B. Chin, Luis Hernandez, Lorin Lachs, David Pisoni, and Holly Storkel for helpful discussions.

Footnotes

M.S. Vitevitch, mvitevit@ku.edu

1

These results replicate the findings of an auditory naming task described in Vitevitch (1998) with a set of stimuli that were also manipulated in terms of neighborhood spread, but which were not as well controlled as the present stimuli.

2

For the stimulus words in the present set of experiments, the correlation between P and uniqueness point was not significant [r = .15, Z(92) = 1.4, p = .17]. Furthermore, r2 = .02, meaning that 2% of the variability in P was accounted for by the uniqueness point.

References

  1. Allopenna PD, Magnuson JS, Tanenhaus MK. Tracking the time course of spoken word recognition using eye movements: Evidence for continuous mapping models. Journal of Memory & Language. 1998;38:419–439. [Google Scholar]
  2. Andrews S. The effect of orthographic similarity on lexical retrieval: Resolving neighborhood conflicts. Psychonomic Bulletin & Review. 1997;4:439–461. [Google Scholar]
  3. Auer ET, Jr, Luce PA. Probabilistic phonotactics in spoken word recognition. In: Pisoni DB, Remez RE, editors. The handbook of speech perception. Oxford: Blackwell; 2005. pp. 610–630. [Google Scholar]
  4. Bölte J, Uhe M. When is all understood and done? The psychological reality of the recognition point. Brain & Language. 2004;88:133–147. doi: 10.1016/s0093-934x(03)00294-3. [DOI] [PubMed] [Google Scholar]
  5. Bond ZS. Slips of the ear: Errors in the perception of casual conversation. San Diego: Academic Press; 1999. [Google Scholar]
  6. Cohen J. Random means random. Journal of Verbal Learning & Verbal Behavior. 1976;15:261–262. [Google Scholar]
  7. Cohen J, MacWhinney B, Flatt M, Provost J. PsyScope: An interactive graphic system for designing and controlling experiments in the psychology laboratory using Macintosh computers. Behavior Research Methods, Instruments, & Computers. 1993;25:257–271. [Google Scholar]
  8. Coltheart M, Davelaar E, Jonasson JT, Besner D. Access to the internal lexicon. In: Dornic S, editor. Attention and performance VI. New York: Academic Press; 1977. pp. 535–556. [Google Scholar]
  9. Davis CJ. N-Watch: A program for deriving neighborhood size and other psycholinguistic statistics. Behavior Research Methods. 2005;37:65–70. doi: 10.3758/bf03206399. [DOI] [PubMed] [Google Scholar]
  10. Grosjean F. Spoken word recognition processes and the gating paradigm. Perception & Psychophysics. 1980;28:267–283. doi: 10.3758/bf03204386. [DOI] [PubMed] [Google Scholar]
  11. Grosjean F. Gating. Language & Cognitive Processes. 1996;11:597–604. [Google Scholar]
  12. Hino Y, Lupker SJ. Effects of word frequency and spelling-to-sound regularity in naming with and without preceding lexical decision. Journal of Experimental Psychology: Human Perception & Performance. 2000;26:166–183. doi: 10.1037//0096-1523.26.1.166. [DOI] [PubMed] [Google Scholar]
  13. Johnson NF, Pugh KR. A cohort model of visual word recognition. Cognitive Psychology. 1994;26:240–346. doi: 10.1006/cogp.1994.1008. [DOI] [PubMed] [Google Scholar]
  14. Keppel G. Words as random variables. Journal of Verbal Learning & Verbal Behavior. 1976;15:263–265. [Google Scholar]
  15. Kuèera H, Francis WN. Computational analysis of present-day American English. Providence, RI: Brown University Press; 1967. [Google Scholar]
  16. Landauer TK, Streeter LA. Structural differences between common and rare words: Failure of equivalence assumptions for theories of word recognition. Journal of Verbal Learning & Verbal Behavior. 1973;12:119–131. [Google Scholar]
  17. Lewandowsky S. The rewards and hazards of computer simulations. Psychological Science. 1993;4:236–243. [Google Scholar]
  18. Luce PA. A computational analysis of uniqueness points in auditory word recognition. Perception & Psychophysics. 1986;39:155–158. doi: 10.3758/bf03212485. [DOI] [PubMed] [Google Scholar]
  19. Luce PA, Goldinger SD, Auer ET, Jr, Vitevitch MS. Phonetic priming, neighborhood activation, and PARSYN. Perception & Psychophysics. 2000;62:615–625. doi: 10.3758/bf03212113. [DOI] [PubMed] [Google Scholar]
  20. Luce PA, Pisoni DB. Recognizing spoken words: The neighborhood activation model. Ear & Hearing. 1998;19:1–36. doi: 10.1097/00003446-199802000-00001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Marslen-Wilson WD, Welsh A. Processing interactions and lexical access during word recognition in continuous speech. Cognitive Psychology. 1978;10:29–63. [Google Scholar]
  22. Mcclelland JL, Elman JL. The TRACE model of speech perception. Cognitive Psychology. 1986;18:1–86. doi: 10.1016/0010-0285(86)90015-0. [DOI] [PubMed] [Google Scholar]
  23. Norris D. Shortlist: A connectionist model of continuous speech recognition. Cognition. 1994;52:189–234. [Google Scholar]
  24. Norris D, McQueen JM, Cutler A. Merging information in speech recognition: Feedback is never necessary. Behavioral & Brain Sciences. 2000;23:299–370. doi: 10.1017/s0140525x00003241. [DOI] [PubMed] [Google Scholar]
  25. Nusbaum HC, Pisoni DB, Davis CK. Sizing up the Hoosier mental lexicon: Measuring the familiarity of 20,000 words. Bloomington: Indiana University, Psychology Department, Speech Research Laboratory; 1984. (Research on Speech Perception, Progress Report No. 10) [Google Scholar]
  26. Pugh KR, Rexer K, Peter M, Katz L. Neighborhood effects in visual word recognition: Effects of letter delay and nonword context difficulty. Journal of Experimental Psychology: Learning, Memory, & Cognition. 1994;20:639–648. doi: 10.1037//0278-7393.20.3.639. [DOI] [PubMed] [Google Scholar]
  27. Raaijmakers JGW. A further look at the “language-as-fixed-effect fallacy. Canadian Journal of Experimental Psychology. 2003;57:141–151. doi: 10.1037/h0087421. [DOI] [PubMed] [Google Scholar]
  28. Raaijmakers JGW, Schrijnemakers JMC, Gremmen F. How to deal with “The language-as-fixed-effect fallacy”: Common misconceptions and alternative solutions. Journal of Memory & Language. 1999;41:416–426. [Google Scholar]
  29. Radeau M, Morais J. The uniqueness point effect in the shadowing of spoken words. Speech Communication. 1990;9:155–164. [Google Scholar]
  30. Radeau M, Morais J, Mousty P, Bertelson P. The effect of speaking rate on the role of the uniqueness point in spoken word recognition. Journal of Memory & Language. 2000;42:406–422. [Google Scholar]
  31. Smith JEK. The assuming-will-make-it-so fallacy. Journal of Verbal Learning & Verbal Behavior. 1976;15:262–263. [Google Scholar]
  32. Storkel HL. Restructuring of similarity neighbourhoods in the developing mental lexicon. Journal of Child Language. 2002;29:251–274. doi: 10.1017/s0305000902005032. [DOI] [PubMed] [Google Scholar]
  33. Storkel HL. Do children acquire dense neighborhoods? An investigation of similarity neighborhoods in lexical acquisition. Applied Psycholinguistics. 2004;25:201–221. [Google Scholar]
  34. Vitevitch MS. The neighborhood characteristics of malapropisms. Language & Speech. 1997;40:211–228. doi: 10.1177/002383099704000301. [DOI] [PubMed] [Google Scholar]
  35. Vitevitch MS. All neighborhoods are not created equal: The phonological P-metric and spoken word recognition. Bloomington: Indiana University, Psychology Department, Speech Research Laboratory; 1998. (Research on Speech Perception, Progress Report No. 22) [Google Scholar]
  36. Vitevitch MS. Influence of onset density on spoken-word recognition. Journal of Experimental Psychology: Human Perception & Performance. 2002a;28:270–278. doi: 10.1037//0096-1523.28.2.270. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Vitevitch MS. The influence of phonological similarity neighborhoods on speech production. Journal of Experimental Psychology: Learning, Memory, & Cognition. 2002b;28:735–747. doi: 10.1037//0278-7393.28.4.735. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Vitevitch MS. Naturalistic and experimental analyses of word frequency and neighborhood density effects in slips of the ear. Language & Speech. 2002c;45:407–434. doi: 10.1177/00238309020450040501. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Vitevitch MS, Luce PA. When words compete: Levels of processing in perception of spoken words. Psychological Science. 1998;9:325–329. [Google Scholar]
  40. Vitevitch MS, Luce PA. Probabilistic phonotactics and neighborhood activation in spoken word recognition. Journal of Memory & Language. 1999;40:374–408. [Google Scholar]
  41. Vitevitch MS, Luce PA. A Web-based interface to calculate phonotactic probability for words and nonwords in English. Behavior Research Methods, Instruments, & Computers. 2004;36:481–487. doi: 10.3758/bf03195594. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Vitevitch MS, Luce PA. Increases in phonotactic probability facilitate spoken nonword repetition. Journal of Memory & Language. 2005;52:193–204. [Google Scholar]
  43. Vitevitch MS, Rodríguez E. Neighborhood density effects in spoken word recognition in Spanish. Journal of Multilingual Communication Disorders. 2005;3:64–73. doi: 10.1080/14769670400027332. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Vitevitch MS, Sommers MS. The facilitative influence of phonological similarity and neighborhood frequency in speech production in younger and older adults. Memory & Cognition. 2003;31:491–504. doi: 10.3758/bf03196091. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Wike EL, Church JD. Comments on Clark’s “The language-as-fixed-effect fallacy. Journal of Verbal Learning & Verbal Behavior. 1976;15:249–255. [Google Scholar]

RESOURCES