See discussions, stats, and author profiles for this publication at: [Link]

net/publication/322070465

Semantic Ambiguity: Do Multiple Meanings Inhibit or Facilitate Word

Recognition?

Article  in  Journal of Psycholinguistic Research · June 2018

DOI: 10.1007/s10936-017-9554-3

CITATIONS READS

5 2,456

2 authors:

Juan Haro Pilar Ferré

Universitat Rovira i Virgili Universitat Rovira i Virgili
29 PUBLICATIONS   72 CITATIONS    93 PUBLICATIONS   1,200 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Syntactic ambiguity View project

Semantic similarity View project

All content following this page was uploaded by Juan Haro on 27 December 2017.

The user has requested enhancement of the downloaded file.

Semantic ambiguity: Do multiple meanings inhibit or facilitate word

recognition?

Juan Haro1 and Pilar Ferré1

1
Research Center for Behavior Assessment (CRAMC) and Department of Psychology.

Universitat Rovira i Virgili. Tarragona. Spain.

Juan Haro

Research Center for Behavior Assessment (CRAMC) and Department of Psychology. Universitat

Rovira i Virgili.

Crta. de Valls s/n, Campus Sescelades, 43007, Tarragona. Spain.

E-mail: [Link]@[Link]

Telephone Number: +34-977-558567

1
Semantic ambiguity: Do multiple meanings inhibit or facilitate word

recognition?

It is not clear whether multiple unrelated meanings inhibit or facilitate word

recognition. Some studies have found a disadvantage for words having multiple
meanings with respect to unambiguous words in lexical decision tasks (LDT),
whereas several others have shown a facilitation for such words. In the present
study, we argue that these inconsistent findings may be due to the approach
employed to select ambiguous words across studies. To address this issue, we
conducted three LDT experiments in which we varied the measure used to
classify ambiguous and unambiguous words. The results suggest that multiple
unrelated meanings facilitate word recognition. In addition, we observed that
the approach employed to select ambiguous words may affect the pattern of
experimental results. This evidence has relevant implications for theoretical
accounts of ambiguous words processing and representation.

Keywords: semantic ambiguity; ambiguity advantage; ambiguity disadvantage;

multiple meanings; word recognition

2
Introduction

Ambiguity is a characteristic feature of language: most of the words that we read or hear

every day have more than one meaning (i.e., ambiguous words; e.g., pupil, bat, newspaper,

etc.). As such, a complete theory of language comprehension must account for how

ambiguous words are processed and represented. This is why semantic ambiguity has been

extensively studied for the last 40 years by using different experimental paradigms and tasks,

such as lexical decision task (LDT), reading task, semantic categorization task, or relatedness

judgment task, among others (see Eddington & Tokowicz, 2015, and Simpson, 1984, for

reviews). The results of these studies have been inconclusive concerning the role of semantic

ambiguity in processing. The main aim of the present work was to shed further light on this

issue by examining whether multiple meanings facilitate or inhibit word recognition. To do

so, we focused on how ambiguous words are processed in LDT, which is the most used task

to study word recognition (e.g., Balota et al., 2007).

Initial studies on semantic ambiguity showed that ambiguous words were recognized

faster than unambiguous words in LDT (e.g., Borowsky & Masson, 1996; Hino & Lupker,

1996; Jastrzembski, 1981; Jastrzembski & Stanners, 1975; Kellas, Ferraro, & Simpson, 1988;

Millis & Button, 1989; but see Forster & Bednall, 1976; Gernsbacher, 1984). This so-called

ambiguity advantage received different explanations. Originally it was argued that

ambiguous words would have multiple lexical representations, each for one of their

meanings. Consequently, the chance to quickly select one of the lexical representations of an

ambiguous word in LDT would be higher than the chance to select the single representation

of an unambiguous word (Jastrzembski, 1981; Rubenstein, Garfield, & Millikan, 1970). In

contrast, subsequent accounts suggested that ambiguous words might not have multiple

lexical representations, but rather multiple semantic representations. Under this assumption,

the ambiguity advantage would arise from ambiguous words triggering a greater amount of

3
semantic activation after their presentation than unambiguous words. This would facilitate

their recognition either by an increase in the global activation at the semantic level

(Borowsky & Masson, 1996) or by a large semantic-to-orthographic feedback (Balota,

Ferraro, & Connor, 1991; Hino & Lupker, 1996).

Whatever the explanation, the mentioned accounts agree in that ambiguous words

benefit from having multiple meanings. However, several studies conducted during the last

decade have strongly challenged that the source of the ambiguity advantage is the multiplicity

of meanings (Armstrong & Plaut, 2008, 2011; Beretta, Fiorentino, & Poeppel, 2005;

Klepousniotou & Baum, 2007; Rodd, Gaskell, & Marslen-Wilson, 2002; Tamminen, Cleland,

Quinlan, & Gaskell, 2006). These studies have distinguished two types of semantic

ambiguity: (a) ambiguity between meanings, and (b) ambiguity between senses. On the one

hand, ambiguity between meanings –also called homonymy– is the best known form of

semantic ambiguity. It is observed in words referring to multiple unrelated meanings; for

example, in words like bat, which can mean either nocturnal flying mammal, or a club used

in certain games to strike the ball. On the other hand, ambiguity between senses is present in

words referring to a wide range of related meanings; for instance, in the word newspaper,

which means: (a) a publication, usually issued daily or weekly; (b) a business organization

that prints and distributes such a publication; (c) a single issue of such a publication, and (d)

the paper in which a newspaper is printed. This class of semantic ambiguity is called

polysemy, and each of the related meanings of a polysemous word is named sense. Of note, in

the following we will refer to unrelated meanings as meanings and to related meanings as

senses.

The first study that examined these two types of semantic ambiguity was that of Rodd

et al. (2002, Experiments 2 [visual] and 3 [auditory]), where number of meanings (i.e., one

vs. many) and number of senses (few vs. many) were manipulated orthogonally. Contrary to

4
previous evidence, the results showed that words with many meanings were recognized

slower than words with one meaning (although only in the Experiment 3). In contrast, words

with many senses were recognized faster than words with few senses. Thus, surprisingly,

these authors found that multiple meanings inhibited word recognition (i.e., ambiguity

disadvantage), whereas many senses facilitated it (i.e., sense advantage). Furthermore, Rodd

et al. analysed previous studies reporting a clear ambiguity advantage (e.g., Azuma & Van

Orden, 1997; Millis & Button, 1989), and found that their ambiguous words differed in

number of senses, but not in number of meanings, with respect to their unambiguous words.

According to Rodd et al., all these findings suggested that previous reports of a processing

advantage for ambiguous words should be interpreted as a benefit for words with many

senses rather than for words with many meanings. Thus, these authors concluded that the

source of the ambiguity advantage is not the multiplicity of meanings, but the multiplicity of

senses.

The results from Rodd et al. (2002) posed a challenge to previous accounts of

ambiguous word recognition. In order to provide an explanation for them, Rodd, Gaskell, and

Marslen-Wilson (2004) developed a PDP model of word recognition. In this model, each

meaning or sense of a word is represented by an attractor basin located in a semantic

network. During the processing of a word, the semantic network moves from an initial state

towards the word’s attractor basin. The word is recognized (e.g., in LDT) when the semantic

network enters the word’s attractor basin. Senses are represented by neighbouring attractor

basins, forming a single, broad and shallow attractor basin; in contrast, meanings are

represented by attractor basins located at different and distant regions of the semantic

network. Given these assumptions, the model predicts a sense advantage: the semantic

network should find the broad and shallow attractor basin of a word with multiple senses

faster than the narrow attractor basin of a word with few senses. On the other hand, a distinct

5
prediction is made for words with multiple meanings. In this case, a blend of all their

meanings would be activated (i.e., a blend state) in the early stages of the semantic network’s

settling. After that, and in order to complete word processing, the semantic network should

escape from this blend state and move towards one of the word’s attractor basins. Assuming

that moving away from the blend state involves a high processing cost, and that meanings

compete during word processing, the model predicts a disadvantage in the recognition of

words with multiple meanings in comparison to words with one meaning.

Rodd et al. (2004)’s model has gathered additional support from several LDT studies

that showed a disadvantage for words with multiple meanings along with an advantage for

words with many senses (Armstrong & Plaut, 2008, 2011; Beretta et al., 2005; Tamminen et

al., 2006). However, there is also evidence incompatible with the model, as other LDT

studies have found an advantage of the same magnitude for both types of ambiguous words

(e.g., Hino, Kusunose, & Lupker, 2010; Hino, Pexman, & Lupker, 2006; Pexman, Hino, &

Lupker, 2004). Importantly, these latter results are consistent with previous reports of an

ambiguity advantage in suggesting that multiple meanings facilitate word processing (e.g.,

Borowsky & Masson, 1996; Hino & Lupker, 1996; Jastrzembski, 1981; Jastrzembski &

Stanners, 1975; Kellas, Ferraro, & Simpson, 1988; Millis & Button, 1989).

As can be seen, the above conflicting findings do not allow researchers to draw firm

conclusions about the representation and processing of ambiguous words. The main question

is whether multiple unrelated meanings (e.g., bat) inhibit or facilitate word processing. The

way in which this question is answered has a significant theoretical relevance, as each

possible answer would give support to a different account of the semantic ambiguity

phenomenon. Indeed, the ambiguity disadvantage would suggest, according to the Rodd et al.

(2004)’s account, that the different semantic representations of a word with multiple

meanings compete during word recognition. In contrast, the ambiguity advantage would be in

6
line with accounts of an enhanced semantic activation for ambiguous words, according to

which words with multiple meanings benefit from triggering a greater amount of semantic

activation than unambiguous words (e.g., Borowsky & Masson, 1996; Hino & Lupker, 1996).

In what follows we suggest that the above opposite experimental findings could be explained,

at least in part, by the different approaches employed to categorize ambiguous and

unambiguous words across studies.

Importantly, all the LDT studies that obtained an ambiguity disadvantage employed a

dictionary approach (Armstrong & Plaut, 2008, 2011; Beretta et al., 2005; Rodd et al., 2002;

Tamminen et al., 2006). This approach relies on the assumption that unrelated meanings are

listed in separate dictionary entries, so that words with more than one dictionary entry are

classified as ambiguous, whereas words with only one dictionary entry are classified as

unambiguous. Moreover, four of these dictionary-based studies used exactly the same set of

experimental stimuli (Armstrong & Plaut, 2008; Beretta et al., 2005; Rodd et al., 2002;

Tamminen et al., 2006), which were in turn those employed by Rodd et al. (2002,

Experiments 2 and 3). Finally, it should be noted that the ambiguity disadvantage has been

found only in studies conducted in English.

On the contrary, the studies that found an advantage for both types of ambiguous

words (i.e., polysemes and homonyms) relied on a subjective approach to categorize their

stimuli (e.g., Haro, Demestre, Boada, & Ferré, 2017; Hino et al., 2010; Hino et al., 2006; Lin

& Ahrens, 2010; Pexman, Hino, & Lupker, 2004), and they were conducted in different

languages, such as English (e.g., Experiment 1 in Pexman et al., 2004), Japanese (e.g., Hino

et al., 2006), Chinese (e.g., Lin & Ahrens, 2010), and Spanish (Haro et al., 2017). The

subjective measures employed in these studies were mainly two: Number-Of-Meanings

(hereafter, NOM) and Relatedness-Of-Meanings (hereafter, ROM). To obtain NOM ratings,

participants are asked to indicate if a word has one or more than one meaning, so that this

7
measure allows to determine if a word is ambiguous or not according to speakers’

knowledge. On the other hand, to collect ROM ratings, participants are required to indicate

whether the meanings of an ambiguous word are related or unrelated. As such, ROM can be

employed to categorize ambiguous words into polysemes and homonyms on the basis of the

linguistic knowledge of participants.

It is worth noting that depending on the approach employed, the same word might be

classified either as unambiguous or as ambiguous. This becomes clear if we compare how

words’ meanings are represented in dictionaries to how they are likely to be represented in

the speaker’s memory. For instance, Gernsbacher (1984) showed that college professors

could provide only 2 of the 30 dictionary definitions of the word gauge, 3 of the 15

dictionary definitions of the word fudge, and 1 of the 15 dictionary definitions of the word

cadet. This led Gernsbacher to suggest that speakers do not store in their memory all the

word definitions listed in the dictionary, but only a small sample of them. In addition, Ferraro

and Kellas (1990) showed that the correlation between subjects’ ratings of number of

meanings of English words and dictionary entries only accounted for the 12% of the total

variance. In a similar study, Lin and Ahrens (2005) compared subjective norms of word

meanings and dictionary entries in Chinese and English. They observed that, although

correlated, the number of word meanings provided by each measure was significantly

different. In addition, they pointed out that the meanings obtained through each of these

measures may also differ with regard to their content, since dictionaries are plenty of old-

fashioned definitions and do not reflect the emerging novel meanings of words.

The above considerations indicate that a dictionary approach may have some

limitations as a psychological measure of semantic ambiguity. Indeed, we believe that there

are, at least, two potential problems with this approach. The first of them occurs when a word

is classified as ambiguous according to the dictionary (i.e., because it has more than one

8
dictionary entry), but the second and subsequent entries/meanings are unknown or unfamiliar

for the majority of the speakers. This usually concerns words whose subordinate dictionary

meanings are old-fashioned, jargon or have a very low frequency. For example, considering

the Spanish Language Dictionary published by the “Real Academia Española” (RAE) (2014),

the word coleta is ambiguous because it has two distinct entries. The first entry is related to a

tied-back hairstyle (i.e., “ponytail” in English), whereas the second one refers to a mix of glue

and honey used to restore paintings. Although many speakers know the first meaning of

coleta, the second meaning is unknown or unfamiliar to almost all of them, as it is a very low

frequency jargon meaning.

The second problem arises when the word is ambiguous for nearly all the speakers,

but not according to the dictionary (i.e., it has only one dictionary entry). An example is the

Spanish word tronco, which has two unrelated meanings for the majority of the speakers:

trunk and mate. However, the RAE dictionary includes only one entry for the word tronco,

which, in addition, does not cover the mate meaning, the most recent but widely known

meaning of such word. Moreover, many dictionaries group together different meanings under

the same entry according to etymological criteria. This leads to cases where unrelated

meanings are put together within the same entry. For example, the RAE dictionary provides

only one entry for the word sirena (“siren” in English), comprising their two unrelated

meanings: sea nymph and alarm device. More importantly for the present study, several

instances of this problem can be found in the stimuli of Rodd et al. (2002). For example, the

word hang was considered as unambiguous in such study, although it has two unrelated

meanings (i.e., to hold and the idea of how to do something, i.e., knack). The same was true

for the words belt (i.e., a strip of cloth and to sing loudly) and soap (i.e., a cleaning substance

and soap opera). Probably, the reason for this was that the two meanings of these words are

9
listed under the same entry in the Wordsmyth dictionary (Parks, Ray, & Bland, 1998), the

one used by Rodd et al.

Given the differences between both approaches, it is very relevant to determine

whether the criterion used to classify words as ambiguous or unambiguous has contributed to

the inconsistent findings between dictionary-based and subjective-based studies. Thus, to

contribute to a better understanding of how ambiguous words are processed, we conducted

three LDT experiments in which we varied the measure used to categorize ambiguous and

unambiguous words. This would particularly help to identify the most suitable -and

psychologically valid- approach to classify words having one and multiple meanings, and

thereby to properly assess whether multiple meanings facilitate or inhibit word recognition.

Experiment 1

The aim of this experiment was to examine how ambiguous words are processed in LDT, by

using a dictionary approach. To do so, we manipulated orthogonally dictionary entries and

dictionary senses. Thus, words with many dictionary entries were defined as ambiguous, and

words with one dictionary entry were defined as unambiguous. In addition, words were

classified as having few or many senses on the basis of their number of dictionary senses. In

line with previous studies that used a similar dictionary approach, we predicted a lack of

advantage for words with many dictionary meanings. Indeed, they would be recognized

either as fast and accurately (e.g., Rodd et al., 2002 [Experiment 2]), or even slower and less

accurately (Armstrong & Plaut, 2008, 2011; Beretta et al., 2005; Tamminen et al., 2006) than

words with one dictionary meaning. In contrast, we expected words with many senses to be

recognized faster and more accurately than words with few senses (Rodd et al., 2002).

Finally, we did several additional analyses by considering a subjective measure of ambiguity

(NOM) in order to know whether the pattern of results varied depending on the criterion used

10
to classify the stimuli.

Method

Participants

Thirty-nine Spanish speakers (34 women and 5 men, mean age 20.4 years, SD = 2.1)

participated in the experiment. They were undergraduate students who received academic

credits for their participation. All of them had either normal or corrected-to-normal vision.

Design and materials

Following previous studies (Beretta et al., 2005; Rodd et al., 2002), we used a 2x2 factorial

design, with the first factor ambiguity (unambiguous words vs ambiguous words) and the

second, number of senses (words with few senses vs words with many senses). We selected

72 Spanish words from the RAE dictionary to be included in the factorial design. Regarding

the ambiguity factor, words were categorized as unambiguous if they had only one entry in

the RAE dictionary and as ambiguous if they had more than one entry. Overall, the average

number of dictionary entries was 1 (SD = 0) and 2.14 (SD = 0.49) for unambiguous and

ambiguous words, respectively, t(70) = 14.03, p < .001. Number of senses for each word was

computed as the total number of definitions listed considering all the dictionary entries of that

word. Words having 10 or less dictionary senses were classified as words with few senses,

whereas those having more than 10 senses were classified as words with many senses. Words

with few senses had 6.33 senses (SD = 2.46) on average, and words with many senses had

17.44 senses (SD = 6.25) on average, t(70) = 9.93, p < .001.

Each cell of the factorial design consisted of 18 words. Number of dictionary senses

was matched between ambiguous and unambiguous words, t(70) = 0.48, p = .63, and number

of dictionary entries was matched between words with few senses and words with many

senses, t(70) = 0.28, p = .86. In addition, many lexical and semantic variables were matched

11
as close as possible across experimental conditions (all ps > .2, see Table 1 for more details).

These variables were obtained from distinct sources. On the one hand, word length, number

of syllables, logarithm word frequency (log word frequency), mean Levenshtein distance of

the 20 closest words (old20), number of neighbours, number of higher frequency neighbours,

bigram frequency, trigram frequency, and logarithm contextual diversity (log contextual

diversity) were obtained from the EsPal subtitle tokens’ database (Duchon, Perea, Sebastián-

Gallés, Martí, & Carreiras, 2013). On the other hand, familiarity, concreteness, subjective age

of acquisition, arousal and emotional valence ratings were taken from the database of Haro,

Ferré, Boada, and Demestre (2017).

12
Table 1. Characteristics of the stimulus used in the experiments (standard deviations are shown in parentheses).
DIC SEN NOM ROM1 ROM2 FRE CTD FAM AoA LNG SYL CON VAL ARO OLD NEI NHF BFQ TFQ
Experiment 1
Unambiguous, 1 6.28 1.6 1.35 0.93 5.51 6.58 5.67 2.39 4.66 4.98 4.32 1.62 7.06 1.11 5224 675
- -
few senses (0) (2.63) (0.29) (0.53) (0.44) (0.73) (2.2) (1.41) (0.61) (0.91) (0.95) (0.81) (0.41) (7.65) (1.94) (2885) (633)
Unambiguous, 1 18.33 1.71 1.25 0.88 5.68 6.25 5.5 2.33 4.6 4.96 4.27 1.55 8.72 1.61 5696 611
- -
many senses (0) (6.94) (0.21) (0.52) (0.4) (0.74) (1.99) (1.29) (0.69) (0.92) (1.05) (0.78) (0.41) (9.32) (2.3) (4185) (793)
Ambiguous, 2.17 6.39 1.54 1.11 0.74 5.34 6.74 5.56 2.5 4.8 5.23 4.43 1.62 7.78 1.56 4855 674
- -
few senses (0.51) (2.35) (0.27) (0.67) (0.5) (0.96) (1.72) (1.54) (0.62) (1.17) (1.52) (1.34) (0.48) (8.36) (3.29) (3843) (731)
Ambiguous, 2.11 16.56 1.79 1.35 0.96 5.45 5.97 5.67 2.56 4.51 5.31 4.53 1.44 8.89 0.89 6203 986
- -
many senses (0.47) (5.52) (0.15) (0.67) (0.53) (1.09) (1.95) (1.33) (0.51) (0.82) (1.2) (1.01) (0.32) (8.26) (1.49) (4295) (1298)
Experiment 2
4.63 1.13 0.99 0.63 5.36 6.43 6.69 2.78 5.03 5.12 4.53 1.97 4.38 0.94 5218 489
Unambiguous - - -
(1.76) (0.1) (0.59) (0.44) (1.04) (2.16) (2.21) (0.87) (0.93) (1.3) (0.96) (0.57) (6.61) (1.88) (3420) (471)
4.91 1.78 1.04 0.7 5.32 6.73 6.41 2.69 4.84 5.38 4.63 1.83 4.13 0.63 5507 705
Ambiguous - - -
(1.59) (0.08) (0.5) (0.38) (0.78) (1.81) (1.32) (0.59) (0.65) (1.01) (1.12) (0.42) (5) (1.34) (3543) (707)
Experiment 3
5.63 1.13 1.10 0.71 5.48 6.39 6.37 2.78 5.07 5.17 4.51 1.93 5.07 1.07 4300 433
- - -
Unambiguous (2.75) (0.10) (0.66) (0.49) (1.07) (2.34) (2.51) (0.97) (1.02) (1.29) (0.95) (0.69) (6.69) (2.04) (3284) (430)
6.19 1.76 1.76 2.06 0.97 0.63 5.31 6.94 6.41 2.78 4.76 4.87 4.55 1.78 6.48 1.37 5708 608
-
Homonyms (2.34) (0.12) (0.53) (0.50) (0.52) (0.40) (0.73) (2.26) (1.74) (0.64) (0.80) (0.95) (0.85) (0.52) (8.65) (2.39) (3696) (522)

Note. DIC = number of dictionary entries; SEN = number of dictionary senses; NOM = subjective Number-Of-Meanings ratings; FRE = log word frequency; CTD = log

contextual diversity; FAM = familiarity; AoA = subjective age-of-acquisition; LNG = word length; SYL = number of syllables; CON = concreteness; VAL = emotional

valence; ARO = emotional arousal; OLD = old20; NEI = number of substitution neighbors; NHF = number of higher frequency substitution neighbors; BFQ = mean bigram

frequency; TFQ = mean trigram frequency.

13
 Finally, following the procedure of Rodd et al. (2002), 72 pseudohomophones (i.e.,

nonwords that have a similar pronunciation to words, such as “sircle”, pronounced like

“circle”) matched to words in length were used as nonwords in the LDT. The reason for using

pseudohomophones is that they make the distinction between non-words and real words more

difficult. This may cause a greater activation of the meanings and senses of the words

resulting in larger ambiguity effects (Eddington & Tokowicz, 2015). We also decided to use

pseudohomophones for the sake of comparability with the study of Rodd et al. (2002), as they

employed this type of non-words.

Procedure

Participants completed a lexical decision task consisting of 144 experimental trials. Each trial

started with a fixation point (i.e., “+”) appearing in the middle of the screen for 500 ms. Next,

a string of letters (a word or a nonword) replaced the fixation point, and then participants had

to decide whether the string was a Spanish word or not. They were instructed to press the

“yes” button of a keypad with the preferred hand if the string of letters was a word, and to

press the “no” button of the keypad with the non-preferred hand if it was not a word. The

string of letters remained on the screen until participant’s response or timeout (i.e., 2000 ms).

After responding, a feedback message (i.e., “ERROR” or “CORRECT”) was displayed for

750 ms. The interval time between trials was 500 ms. The DMDX software (Forster &

Forster, 2003) was used for displaying the stimuli and recording the responses. The order of

the experimental trials was randomized for each participant. Prior to the beginning of the

experiment, a practice block consisting of 10 trials (5 words and 5 nonwords) was presented.

Data analyses

Data on reaction times (RT) and error rates (%E) were analyzed separately using linear

mixed-effect models (e.g., Baayen, 2008; Baayen, Davidson, & Bates, 2008). For this we

used the lme4 package of R (Bates, Maechler, Bolker, & Walker, 2014). In each analysis, the

14
word’s ambiguity (unambiguous and ambiguous words), the number of senses of the word

(few senses or many senses), and the ambiguity x number of senses interaction were included

as fixed effects, and participants and words as random effects (adjusting for the intercept). To

determine the significance of fixed effects, log-likelihood ratio tests were used (R function

Anova). We assessed the contribution of each fixed effect by comparing a model that

included the effect of interest with another model that did not include it. We also checked

whether the fit of the model was significantly improved by adding any of the following

additional variables: log word frequency, word length, familiarity, number of neighbours, and

concreteness. Thus, we first created a model including the variables of interest and their

interaction (i.e., ambiguity and number of senses), random effects (participants and words),

and the additional variables. The model was then progressively simplified by excluding one

additional variable in each step, this being the one making the lowest contribution at that

point. As such, the remaining model only included those additional variables that

significantly increased its fit. Of note, the p-values for pairwise comparisons were estimated

using the lmerTest package (Kuznetsova, Brockhoff, & Christensen, 2014).

Results and Discussion

We removed the data of one participant with more than 15% of errors. Reaction times that

exceeded 2 SD of each participant’s mean were also removed (5.0% of the whole). In

addition, we excluded from the analyses one unambiguous word with many senses due to a

high percentage of errors (>70%). The mean of reaction times (RT) for correct responses and

the mean of error rates (%E) across experimental conditions (averaged across participants)

are shown in Table 2. The best fitting model for RTs included the additional variables log

word frequency and familiarity, which significantly increased model's fit (all ps < .05). On

15
the other hand, the best fitting model for %E included the additional variable log word

frequency (p < .05).

Table 2. Mean RT (in ms), and percentage of error rates (%E) in Experiment 1 (averaged

across participants; standard deviations in parentheses)

Ambiguity Senses Mean RT %E

Unambiguous Few 595 (77) 1.81 (3.03)

Unambiguous Many 591 (81) 3.92 (4.48)

Ambiguous Few 634 (87) 12.24 (9.01)

Ambiguous Many 601 (73) 4.41 (4.48)

Nonwords 673 (92) 9.4 (6.43)

Concerning ambiguity, there was a significant effect on RTs, χ²(1) = 6.01, p = .014.

Ambiguous words were recognized slower than unambiguous words, ß = 23.36, SE = 9.92, t

= 2.35, p = .022. Likewise, ambiguous words were recognized less accurately than

unambiguous words, χ²(1) = 8.34, p = .004, ß = .07, SE = .02, t = 2.87, p = .005. With respect

to number of senses, no differences were found between words with many senses and words

with few senses, either on RTs, χ²(1) = 2.56, p = .11, or on %E, χ²(1) = 0.99, p = .32.

Regarding the interaction between ambiguity and number of senses, it was significant in %E,

χ²(1) = 4.33, p = .04. Post-hoc comparisons showed that this interaction was produced

because number of senses reduced the %E of ambiguous words, ß = -.06, SE = .04, t = -1.49,

p = .15, but increased that of unambiguous words, ß = .03, SE = .01, t = 1.93, p = .06

(although the comparisons were not significant).

16
 The results of this experiment showed a disadvantage for ambiguous words: words

with more than one dictionary entry were recognized more slowly and less accurately than

words with one entry. It should be noted, however, that this disadvantage seemed to be due

mainly to ambiguous words with few senses, in that this was the condition with the slowest

RTs and the highest %E, where a considerable difference was seen compared to the other

conditions (although this difference failed to reach statistical significance). On the one hand,

these results are incompatible with reports of an ambiguity advantage from LDT studies that

relied on subjective norms (e.g., Hino et al., 2010; Hino et al., 2006; Pexman et al., 2004).

Conversely, they are in line with several English LDT studies that employed a dictionary

approach for selecting ambiguous and unambiguous words (Armstrong & Plaut, 2008, 2011;

Beretta et al., 2005; Rodd et al., 2002; Tamminen et al., 2006) and that mainly relied on the

same stimulus set. The data from the present experiment replicate these latter findings by

using the same approach for selecting ambiguous words, and, importantly, they extend them

to another language and to another set of words. Hence, these results support that multiple

unrelated meanings inhibit word recognition.

However, before drawing definite conclusions, we wanted to explore if the results of

this experiment were affected by the type of criterion used to classify the

ambiguous/unambiguous words. To that end, we obtained NOM ratings for the experimental

words from Haro et al. (2017). As described earlier, NOM ratings are a subjective measure of

semantic ambiguity. To collect them, participants are usually required to think about all the

meanings of a string of letters and then choose the appropriate value in a three-point scale:

the word has no meaning (0), the word has one meaning (1), or the word has more than one

meaning (2). Words with values close to 2 are considered as ambiguous, and words with

values close to 1 are considered as unambiguous (e.g., Hino et al., 2010; Hino et al., 2006;

Pexman et al., 2004).

17
 The analysis of our experimental stimuli showed, on the one hand, that that there was

no difference in NOM ratings for words with many dictionary entries (NOM = 1.66) and

words with one dictionary entry (NOM = 1.65), t(69) = 0.24, p = .81 (see Table 1 for the

NOM ratings of each condition). This may suggest that words with one/many dictionary

entries would not correspond with words having one/many meanings according to the

speakers’ linguistic knowledge. On the other hand, we observed that NOM ratings and

number of senses were significantly correlated, r(71) = .29, p = .01. Words with many senses

had higher NOM values (NOM = 1.75) than words with few senses (NOM = 1.57), t(69) =

3.22, p = .002, from which it may be inferred that participants considered words with many

senses as having more meanings than words with few senses.

After having characterized our stimuli in terms of a subjective measure, we analysed

the relationship between that measure and participants’ performance in the LDT. We found

that NOM ratings were negatively correlated with both RTs, r(71) = .-34, p = .004, and %E,

r(71) = .-39, p = .001. That is, words with higher NOM ratings were responded faster and

more accurately than words with lower NOM ratings. In addition, we reclassified the words

of Experiment 1 based on a median split of their NOM values, finding that words with high

NOM ratings were associated with faster RTs (597 ms) and less errors (3.61%) than words

with low NOM ratings (mean RTs = 624 ms; mean %E = 8%) (both ps < .05). Importantly, it

should be noted that number of dictionary senses did not differ between words with low

NOM (mean senses = 10.66) and words with high NOM (mean senses = 13.06), t(69) = 1.38,

p = .17. Thus, although NOM ratings and number of dictionary senses of the words were

correlated, the NOM advantage may not be entirely explained by dictionary senses.

In sum, these results indicate that the direction of the ambiguity effect varied

depending on the measure employed to classify the same stimuli. That is, we obtained an

ambiguity disadvantage when the stimuli were classified according to the number of

18
dictionary entries, whereas an ambiguity advantage emerged when the classification was

based on NOM ratings. Apart from the methodological and theoretical implications of these

findings, they suggest that the subjective number of meanings could have contributed to the

ambiguity disadvantage found with the dictionary criterion. In particular, this latter result

might be explained by a misdistribution of NOM between words with one/many dictionary

entries. Indeed, although words with one/many dictionary entries did not differ in NOM

ratings, it should be noted that words with many dictionary entries and few senses had the

lowest NOM ratings (mean = 1.54, see table 1). In addition, this was the condition showing

the slowest RTs in the experiment. Interestingly, on examining the words of this condition we

observed that several of them were unambiguous words according to NOM ratings. In

addition, many words of the one dictionary entry and few senses ought to be considered

ambiguous words, based on NOM ratings. Hence, this misdistribution of NOM ratings may

have had a notable impact in these two conditions, increasing the RTs in the many dictionary

entries and few senses condition, and reducing the RTs in the one dictionary entry and few

senses condition. This would also serve to explain the reversed ambiguity effect when using

dictionary entries as a measure of semantic ambiguity.

Taking the above into consideration, it seems that NOM have a relevant role in

determining ambiguity effects. However, this conclusion is based on a post-hoc

reclassification of the stimuli. In order to investigate in more detail this issue, we conducted a

further experiment in which we manipulated NOM ratings. This led us to assess if words

having multiple meanings according to the speakers’ knowledge inhibit or facilitate word

recognition.

Experiment 2

19
In this experiment, we employed NOM ratings as a subjective measure of semantic

ambiguity. We compared words with high NOM ratings (i.e., ambiguous words) to words

with low NOM ratings (i.e., unambiguous words). In line with several LDT studies that

employed this measure (Hino et al., 2010; Hino et al., 2006; Pexman et al., 2004), and given

the negative correlation between NOM and RTs found in Experiment 1, we expected a

facilitation for ambiguous words with respect to unambiguous words. However, it is worth

noting that neither of these NOM-based studies took into account the number of senses of

their experimental stimuli. This is critical, since Rodd et al. (2002) demonstrated that past

reports of an ambiguity advantage from studies relying on subjective norms might be due to

ambiguous words having many senses rather than many meanings. In addition, it has been

shown that words with many senses are recognized faster than words with few senses, (e.g.,

Jager & Cleland, 2016). For that reason, in Experiment 2 we equated number of senses

between conditions, as a way to avoid the possible experimental confounding between

meanings and senses. Of note, we employed number of dictionary senses as a measure of the

number of senses known by the speakers, given the high correlation between both measures

(Fraga, Padrón, Perea, & Comesaña, 2017).

Method

Participants

Thirty-eight Spanish speakers (34 women and 4 men, mean age 21.8 years, SD = 4.3)

participated in the experiment. They were undergraduate students who received academic

credits for their participation. All of them had either normal or corrected-to-normal vision.

Design and materials

Experimental stimuli consisted of 32 ambiguous words and 32 unambiguous words. Unlike

Experiment 1, we relied on subjective measures to define ambiguity. We used two indices of

20
semantic ambiguity taken from the Haro et al. (2017)’s database. The first of them was

number-of-meanings (NOM), which was described in the additional analyses included in

Experiment 1. To obtain the second measure of ambiguity, Haro et al. used a free association

task, by requiring participants to generate word associates for a set of words. Then, on the

basis of these associates, four judges classified the words as unambiguous or ambiguous.

Words were classified as unambiguous if all their associates were related to the same

meaning (e.g., major was rated as unambiguous because participants produced the following

associates for it: council, town, politician, and president), and as ambiguous if their

associates were related to distinct meanings or senses (e.g., mouse was rated as ambiguous

because participants produced the following associates for it: computer, cheese, mickey, cat,

hamster, and keyboard).

Words with NOM values lower than 1.40 were classified as unambiguous, and words

with NOM values higher or equal than 1.65 were classified as ambiguous. The average NOM

rating was 1.13 (SD = 0.10) for unambiguous words and 1.78 (SD = 0.08) for ambiguous

words, t(62) = 27.74, p < .001. In addition, when considering the associates’ measure, all the

unambiguous words had an “unambiguous” rating and all the ambiguous words had an

“ambiguous” rating. As stated before, number of dictionary senses (measured as in

Experiment 1) was matched between conditions, t(62) = 0.67, p = .51. In addition, we

matched the same lexical and semantic variables between ambiguous and unambiguous

words as in Experiment 1 (all ps > .15). Full details of the experimental words are shown in

Table 1. Finally, a set of 64 pseudohomophones matched in length to words was included as

nonwords in the LDT.

Procedure

The experimental procedure was the same as in Experiment 1.

Data analyses

21
The analyses were the same as in Experiment 1, with the exception that the variable of

interest in this experiment was word’s ambiguity (unambiguous and ambiguous words).

Results and Discussion

We removed the data of two participants with more than 15% of errors. RTs that exceeded 2

SD of each participant’s mean were also rejected (4.4% of the whole). In addition, we

excluded from the analysis one unambiguous word due to a high percentage of errors

(>70%). The mean of RTs for correct responses and the mean of %E across experimental

conditions (averaged across participants) are shown in Table 3. The best fitting model for

RTs included the additional variable log word frequency, which significantly increased

model's fit (p < .05). On the other hand, the best fitting model for %E included the additional

variable concreteness (p < .05).

Table 3. Mean RT (in ms), and percentage of error rates (%E) in Experiment 2 (averaged

across participants; standard deviations in parentheses)

Ambiguity Mean RT %E

Unambiguous 675 (107) 6.21 (4.71)

Ambiguous 627 (102) 1.71 (2.71)

Nonwords 724 (123) 6.81 (3.79)

There was a significant facilitative effect for ambiguous words both in RTs, χ²(1) =

13.31, p < .001, ß = -50.96, SE = 13.48, t = 3.78, p < .001, and in %E, χ²(1) = 12.15, p < .001,

ß = -.06, SE = .02, t = 3.58, p < .001.

22
 The results showed a clear ambiguity advantage: ambiguous words were recognized

faster and more accurately than unambiguous words, even when number of dictionary senses

was matched between conditions. These results are at odds with the ambiguity disadvantage

reported when a dictionary criterion is used (e.g., Armstrong & Plaut, 2008; Beretta et al.,

2005; Rodd et al., 2002; Tamminen et al., 2006, and Experiment 1 of this study). In contrast,

this finding is in line with several LDT studies that have relied on subjective norms (e.g.,

Hino et al., 2010; Hino et al., 2006; Lin & Ahrens, 2010; Pexman et al., 2004). It also agrees

with the results from the additional analyses of Experiment 1, which suggested that NOM

facilitates word processing, and that the observed disadvantage for words with many

dictionary entries might be due to a misdistribution of NOM across conditions.

In light of the above, it could be concluded that the advantage for words with high

NOM ratings suggests that multiple meanings facilitate word recognition. This conclusion

would be based on the assumption that these words had multiple meanings according to

participants. However, there are some reasons to question this assumption. First, Rodd et al

(2002) demonstrated that when participants are asked to provide word meanings, they also

take into account related word senses. Second, NOM ratings and dictionary senses are highly

correlated (e.g., r = .50 in the study of Haro et al., 2017). Finally, a close look at the stimuli

of Experiment 2 reveals that there are several examples of ambiguous words with related

senses (e.g., mouse or ecstasy). Consequently, it is possible that unrelated meanings and

related senses were somewhat confounded in Experiment 2. As such, the advantage for words

of high NOM could be due to either multiple unrelated meanings or multiple related senses.

To overcome this limitation and to elucidate whether multiple unrelated meanings facilitate

word recognition, we conducted a new experiment in which we used a subjective measure of

relatedness between meanings.

23
Experiment 3

In this experiment, we compared words having multiple unrelated meanings (homonyms) to

unambiguous words. Homonyms were selected according to ROM ratings, a measure

obtained by asking participants to indicate if the distinct meanings of an ambiguous word are

related or unrelated. Thus, if multiple unrelated meanings facilitate word processing,

homonyms should show an advantage over unambiguous words.

Method

Participants

Forty-four Spanish speakers (37 women; mean age 22.09 years, SD = 3.54) participated in

the experiment. They were undergraduate students who received academic credits for their

participation. All of them had either normal or corrected-to-normal vision.

Design and materials

We selected 27 unambiguous words and 27 homonyms for the present experiment.

Homonyms were chosen according to NOM and ROM ratings (ROM1 and ROM2) obtained

from Haro et al. (2017)’s database. The criteria for the selection of homonyms were that they

should have NOM ratings above 1.5, as well as ROM1 and ROM2 ratings below 3. On the

other hand, the criterion to select unambiguous words was that they should have NOM

ratings below 1.4. Unambiguous words and homonyms differed in NOM ratings (1.13 and

1.76, respectively), t(52) = 20.93, p < .001. However, they did not differ in number of

dictionary senses, t(52) = 1.35, p = .18.

We matched the same lexical and semantic variables as in Experiments 1 and 2

between unambiguous words and homonyms (all ps > .15). Full details of the experimental

words are shown in Table 1. Finally, a set of 54 pseudohomophones matched in length to

words was included as nonwords in the LDT.

24
Procedure

The experimental procedure was the same as in Experiments 1 and 2.

Data analyses

The analyses were the same as in Experiments 1 and 2, with the exception that the variable of

interest in this experiment was homonymy (homonyms and unambiguous words).

Results and Discussion

We removed the data of three participants with more than 15% of errors. RTs that exceeded 2

SD of each participant’s mean were also rejected (4.6% of the whole). The mean of RTs for

correct responses and the mean of %E across experimental conditions (averaged across

participants) are shown in Table 4. The best fitting model for RTs included the additional

variables log word frequency and familiarity, which significantly increased model's fit (p <

.05), whereas the best fitting model for %E included the additional variable familiarity (p <

.05).

Table 4. Mean RT (in ms), and percentage of error rates (%E) in Experiment 3 (averaged

across participants; standard deviations in parentheses)

Ambiguity Mean RT %E

Unambiguous 646 (133) 5.86 (4.17)

Homonyms 617 (116) 4.42 (4.66)

Nonwords 706 (143) 8.22 (5.29)

25
 Homonyms were recognized faster than unambiguous words, χ²(1) = 9.61, p = .002, ß

= -45.70, SE = 14.53, t = 3.15, p = .003. On the other hand, the difference between

homonyms and unambiguous words in %E did not reach significance, χ²(1) = 0.73, p = .39.

The results of the present experiment showed a facilitation in RTs for homonyms with

respect to unambiguous words. This homonymy advantage agrees with past findings (e.g.,

Hino et al., 2010; Hino et al., 2006; Pexman et al., 2004), and supports the hypothesis that

multiple unrelated meanings facilitate word recognition.

General discussion

In the present study we aimed to assess whether multiple meanings facilitate or inhibit word

recognition. To address this issue, we compared ambiguous words to unambiguous words in

three LDT experiments. The results obtained suggest that multiple unrelated meanings

facilitate word recognition. We also found that the results obtained in the LDT experiments

depended on the approach used to select these words. Namely, when number of dictionary

entries was used (Experiment 1), a disadvantage for ambiguous words (i.e., words with more

than one dictionary entry) appeared. Such a result is in line with previous LDT studies that

employed a similar approach to categorize their stimuli (Armstrong & Plaut, 2008, 2011;

Beretta et al., 2005; Rodd et al., 2002; Tamminen et al., 2006). In contrast, when the

classification was made according to subjective ratings of semantic ambiguity (i.e., NOM and

ROM ratings, Experiments 2 and 3), a facilitation for ambiguous words was obtained, in

agreement with early reports of an ambiguity advantage (e.g., Borowsky & Masson, 1996;

Hino & Lupker, 1996; Jastrzembski, 1981; Jastrzembski & Stanners, 1975; Kellas et al.,

1988; Millis & Button, 1989), and with the findings of more recent studies that have relied on

subjective norms (e.g., Haro et al., 2017; Hino et al., 2010; Hino, Lupker, & Pexman, 2002;

Hino et al., 2006; Lin & Ahrens, 2010; Pexman et al., 2004). It is very important to note that

26
this opposite pattern of results was observed even when the same set of words was classified

according to either dictionary measures or subjective ratings (i.e., NOM, Experiment 1).

In our opinion, there are two main points that deserve to be discussed in detail. The

first one is related to the criterion to select ambiguous/unambiguous words (i.e., dictionary

approach vs subjective ratings). The second one is the advantage in processing found for

words with unrelated meanings. Regarding the first point, our findings suggest that

researchers should be cautious when defining semantic ambiguity, since the type of definition

employed (and the way to operationalize the distinction between ambiguous and

unambiguous words) may influence the experimental results obtained. In turn, it would have

consequences for our understanding and theoretical explanations of how ambiguous words

are processed. There are several objective and subjective measures to operationalize semantic

ambiguity, and each of them has some advantages and limitations. Dictionaries provide

objective (i.e., standardized) measures of ambiguity. This is a clear advantage, as it allows

comparability across studies. Concerning dictionary senses, it seems to be a reliable index of

the number of word senses known by the speakers, as revealed by the high correlation

between both measures (Fraga et al., 2017). This makes dictionaries a practical tool for

researchers interested in analysing the effects of number of senses on word processing.

However, there are limitations in the measure of number of meanings (i.e., entries) provided

by dictionaries. As previously stated, dictionaries list some word meanings that are unknown

to the majority of the speakers (e.g., jargon, old-fashioned and low-frequency meanings)

while they do not include novel unrelated word meanings that are commonly used in

everyday life. This may cause that words having one/many dictionary entries do not

correspond to words having one/multiple unrelated meanings according to the speakers’

knowledge, as we have observed in Experiment 1. In fact, the results of Experiment 1 showed

that the pattern of results can be reversed depending on the criterion used to classify the

27
words as unambiguous or ambiguous. We reasoned that the ambiguity disadvantage found in

that experiment could be accounted for by the misdistribution of subjectively unambiguous

and ambiguous words into words with one/more than one dictionary entries. The results of

the present study do not allow us to draw firm conclusions on the extent to which the

ambiguity disadvantage found in past studies could be explained by the same reason, as our

Spanish stimuli are not directly comparable to the English stimuli used in those studies.

However, to explore this point further, we decided to examine the results of those studies that

used a dictionary measure (i.e., a total of 11 experiments, Armstrong & Plaut, 2008, 2011;

Beretta et al., 2005; Rodd et al., 2002). On the one hand, there was an interaction between

meanings and senses in several of these experiments (e.g. Armstrong & Plaut, 2008,

Experiment 2). On the other, we observed that words with many dictionary entries and few

senses had -numerically- the slowest RTs of all 11 experiments. By contrast, in only one

experiment did words with many dictionary entries and many senses show slower RTs than

words with one dictionary entry (Exp. 3 from Armstrong & Plaut, 2008). Such evidence

suggest that the ambiguity disadvantage found in these studies was due mainly to words with

many meanings and few senses, as occurred in our Experiment 1. Thus, it could be that, as in

our Experiment 1, subjective NOM ratings were misdistributed across conditions in these

studies, having the highest impact in the condition of many meanings and few senses, and

thus leading to a large inhibition for such words. Further research is needed on this issue.

The above considerations suggest that the use of objective measures has important

limitations. Given that, an alternative for the study of the processing of ambiguity is the use

of subjective measures. Nevertheless, such measures are not exempt either of some

methodological concerns. Indeed, in line with what Rodd et al (2002) pointed out, we have

observed that high NOM ratings can be assigned to both homonymous and polysemous

words (Experiment 2). Therefore, although NOM seems to be a useful measure to

28
discriminate ambiguous and unambiguous words, it is not adequate as a way to separate

related senses and unrelated meanings. Researchers interested in such distinction have to rely

on additional measures, such as ROM ratings. Consequently, it becomes evident that a proper

categorization of ambiguous words should involve the use of several measures of semantic

ambiguity.

The second point worth to be discussed here is the facilitation found in the present

study for words with multiple unrelated meanings, as it has relevant theoretical implications.

Indeed, this finding provides support for the enhanced semantic activation accounts of the

phenomenon. One of them is that of Hino and Lupker (1996), which is based on the

following assumptions. According to this account, ambiguous words are represented by one

orthographic representation linked to multiple semantic representations, one for each

meaning. Furthermore, when a word is presented, activation would flow bidirectionally

between the orthographic and semantic levels. Finally, lexical decisions are supposed to be

made on the basis of the activation at the orthographic level. Given these assumptions, the

ambiguity advantage is quite straightforward: Ambiguous words would benefit from a larger

semantic-to-orthographic feedback than unambiguous words, facilitating their orthographic

processing and thus speeding up their recognition in LDT.

According to enhanced semantic activation accounts, the ambiguity advantage can be

related to the semantic richness effects reported in the literature (see Pexman, Siakaluk, &

Yap, 2013, for a review). Research on semantic richness has been mainly devoted to explore

which dimensions of meaning affect word recognition and to study how orthography and

semantics interact during word processing. The main finding of this line of research is that

words associated with more semantic information are processed faster and more accurately.

Indeed, it has been found that words having more semantic features elicit faster responses in

several experimental tasks, such as LDT, naming and semantic categorization, than words

29
with a lower number of semantic features (Yap, Pexman, Wellsby, Hargreaves, Huff, 2012).

In addition, words with more semantic neighbours or with denser semantic neighbourhoods

are facilitated in naming and lexical decision tasks (Yap et al., 2012). Furthermore, there is an

advantage in LDT for words having more and stronger visual associations (e.g., Hargreaves

& Pexman, 2012). Following this line, the number of unrelated meanings or related senses

that a word has could be regarded as another semantic richness dimension, as having more

meanings or senses lead to faster and more accurately responses in LDT.

Apart from semantic enhanced accounts, another possible explanation for the

advantage is that, for a word with multiple meanings, there are more chances for it to be

recognized as a word1. For example, if a word has four meanings, and the participant does not

know one of these, there are still three meanings available to her/him as a means of deciding

that the item is a word. However, if a word has only one meaning and this is not known to the

participant, she/he will necessarily decide that the item is a non-word. This would be seen

especially in the accuracy measures, and it does, since we found a better accuracy for

ambiguous words in the present study. But it would also speed up RTs for words with

multiple meanings, because with greater opportunities for encountering any of the

meanings of a word, we might also predict that this would happen faster than when the sole

meaning for a word has to be found.

Importantly, the ambiguity advantage it is incompatible with the predictions of Rodd

et al. (2004), as their model proposes a disadvantage for words with multiple meanings.

Nevertheless, we think that Rodd et al.’s model could account for the ambiguity advantage if

some changes were made on it. In what follows we are going to develop this point.

As previously described, the model of Rodd et al. assumes that the disadvantage for

ambiguous words is produced, in part, because a blend of all the meanings of these words

1. We thank an anonymous reviewer for this suggestion.

30
would be activated in the early stages of the semantic network’s settling. Given its unspecific

semantic nature, this state would not be enough to recognize an ambiguous word in LDT.

Thus, in order to make a lexical decision, the semantic network should escape from this state

and settle into a stable semantic pattern representing only one of the meanings of the

ambiguous word. As a result of the need to move away from such state, and assuming that the

multiple semantic representations of an ambiguous word compete during this process, slower

RTs would be expected for ambiguous words in comparison to unambiguous words.

Nevertheless, we would like to point out that some data from simulation studies suggest that

this blend state may be sufficient to recognize a word in LDT. Part from this evidence comes

from the modelling work of Joordens and Besner (1994). These authors tried to simulate the

ambiguity advantage using the distributed memory model of Masson (1991), which included

two word processing modules, one representing the orthography of the word, and the other

representing its meaning. The model was trained with stimuli having a one-to-many mapping

between their orthographic and semantic representations (i.e., representing ambiguous

words), and stimuli having a one-to-one relation (i.e., representing unambiguous words).

After the training phase, the model was presented with ambiguous and unambiguous words,

and the performance of the network was assessed. The results of the test phase showed that

the model failed in most of the cases to retrieve only one of the meanings of ambiguous

words. In those cases, the model settled into a pattern of semantic activation containing

information from the two meanings, that is, a blend state. Interestingly, the model activated

faster the blend state of an ambiguous word than the stable semantic pattern of an

unambiguous word.

Although blend states were initially considered errors of the model (i.e., an incorrect

response in LDT), this assumption was later discussed in several commentaries (e.g., Besner

& Joordens, 1995; Masson & Borowsky, 1995; Rueckl, 1995). Overall, these commentaries

31
suggested the possibility that, as blend states are thought to include information from

multiple meanings, they would provide a strong familiarity cue to discriminate between

words and nonwords. Accordingly, blend states would be enough to make lexical decisions.

If we consider that blend states were reached faster by the network, the ambiguity advantage

might be explained by a PDP model as that employed by Joordens and Besner (1994).

Importantly, this possibility was later confirmed in the simulation study of Borowsky and

Masson (1996), which used the same PDP model as that of Joordens and Besner. The authors

assessed the level of familiarity during word recognition by using a feature of Hopfield

networks called energy. The energy of the network was measured as the sum of activation of

the orthographic and the semantic levels during the processing of an input, and such measure

indicated the distance towards a stable pattern of activation. When the energy of the network

reached a given criterion, a positive lexical decision was thought to occur. The training and

test phases were similar to those conducted by Joordens and Besner. As a result, the authors

observed that when the model was presented with an ambiguous word, the meaning units

settled faster into a blend state in which both meanings of the ambiguous word were

activated. As these states were similar to two stable patterns of semantic activation, they

elicited a larger level of semantic activation than that produced by unambiguous words.

Therefore, ambiguous words reached faster the energy criterion for a positive response than

unambiguous words. In this way, the PDP model of Borowsky and Masson (1996) was able

to account for the ambiguity advantage.

In sum, all this evidence lead us to think that the model of Rodd et al. (2002) could

explain the ambiguity advantage if it assumes that the blend state is enough to make a lexical

decision. However, it is important to note that this blend of meanings would not be sufficient

to perform tasks requiring meaning specification, and thus no advantage would be expected

for ambiguous words in such tasks. In this line, it has been found that ambiguous words are

32
associated with slower responses in semantic categorization tasks (Hino et al., 2002). To

account for this effect, Hino et al. (2002) pointed out that in order to determine if an

ambiguous word belongs to a given category (e.g., is novel a living object?), each of its

meanings must be activated, evaluated, and compared with the specific semantic category. As

a result, more time would be consumed for these words in comparison to unambiguous

words. A similar explanation can be given for the finding that participants tend to fixate

longer on ambiguous words than on unambiguous words during text comprehension (e.g.,

Duffy, Morris, & Rayner, 1988). This is because the correct interpretation of a sentence

containing an ambiguous word requires the disambiguation of its meaning. In contrast, this

disambiguation process does not seem necessary to perform a LDT.

To conclude, the findings of the present study support the idea that multiple unrelated

meanings facilitate word recognition. This pattern of results can be explained by semantic

enhanced accounts, which assume that ambiguous words benefit during recognition either by

an increase in the global activation at the semantic level (Borowsky & Masson, 1996) or by a

large semantic-to-orthographic feedback. On the contrary, these results are, at a first glance,

incompatible with the model of Rodd et al. (2004). Nevertheless, we have argued that this

model could account for the ambiguity advantage by making an additional assumption:

namely, that a blend state is enough to make lexical decisions. On the other hand, the

evidence presented in this study suggest that the approach employed to select ambiguous

words may affect the experimental results obtained. As such, researchers should consider

carefully the advantages and disadvantages of the available ambiguity measures before

selecting their experimental items.

33
References

Armstrong, B. C., & Plaut, D. C. (2008). Settling dynamics in distributed networks explain

task differences in semantic ambiguity effects: Computational and behavioral

evidence. In Proceedings of the 30th Annual Conference of the Cognitive Science

Society (pp. 273-278).

Armstrong, B. C., & Plaut, D. C. (2011). Inducing homonymy effects via stimulus quality

and (not) nonword difficulty: Implications for models of semantic ambiguity and

word recognition. In Proceedings of the 33rd Annual Conference of the Cognitive

Science Society (pp. 2223-2228).

Azuma, T., & Van Orden, G. C. (1997). Why SAFE is better than FAST: The relatedness of a

word's meanings affects lexical decision times. Journal of Memory and

Language, 36(4), 484-504.

Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R.

Cambridge: Cambridge University Press.

Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed

random effects for subjects and items. Journal of Memory and Language, 59, 390–412.

Balota, D. A., Ferraro, F. R., & Connor, L. T. (1991). On the early influence of meaning in

word recognition: A review of the literature. In P. J. Schwanenflugel (Ed.), The

psychology of word meanings (pp. 187-222). Hillsdale, NJ: Erlbaum.

Balota, D. A., Yap, M. J., Hutchison, K. A., Cortese, M. J., Kessler, B., Loftis, B., …

Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39(3),

445–459. doi:10.3758/bf03193014

Bates D., Maechler M., Bolker B., Walker S. (2014). lme4: Linear Mixed-Effects Models

Using Eigen and S4. R package version 1, 1–10. Available online at: [Link]

[Link]/package=lme4

34
Beretta, A., Fiorentino, R., & Poeppel, D. (2005). The effects of homonymy and polysemy on

lexical access: An MEG study. Cognitive Brain Research, 24(1), 57-65.

Besner, D., & Joordens, S. (1995). Wrestling with ambiguity—further reflections: Reply to

Masson and Borowsky (1995) and Rueckl (1995). Journal of Experimental

Psychology: Learning, Memory, and Cognition, 21, 515–301.

Borowsky, R., & Masson, M. E. (1996). Semantic ambiguity effects in word

identification. Journal of Experimental Psychology: Learning, Memory, and

Cognition, 22(1), 63-85.

Duchon, A., Perea, M., Sebastián-Gallés, N., Martí, A., & Carreiras, M. (2013). EsPal: One-

stop shopping for Spanish word properties. Behavior research methods, 45(4), 1246-

1258.

Duffy, S. A., Morris, R. K., & Rayner, K. (1988). Lexical ambiguity and fixation times in

reading. Journal of memory and language, 27(4), 429-446.

Eddington, C. M., & Tokowicz, N. (2015). How meaning similarity influences ambiguous

word processing: the current state of the literature. Psychonomic bulletin &

review, 22(1), 13-37.

Ferraro, F. R., & Kellas, G. (1990). Normative data for number of word meanings. Behavior

Research Methods, Instruments, & Computers, 22(6), 491–498.

doi:10.3758/bf03204432

Forster, K. I., & Bednall, E. S. (1976). Terminating and exhaustive search in lexical

access. Memory & Cognition, 4(1), 53-61.

Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond

accuracy. Behavior Research Methods, Instruments, & Computers, 35(1), 116-124.

Fraga, I., Padrón, I., Perea, M., Comesaña, M. (2017). I saw this somewhere else: The

Spanish Ambiguous Words (SAW) database. Lingua, 185, 1-10.

35
Gernsbacher, M. A. (1984). Resolving 20 years of inconsistent interactions between lexical

familiarity and orthography, concreteness, and polysemy. Journal of Experimental

Psychology: General, 113(2), 256-281.

González-Nosti, M., Barbón, A., Rodríguez-Ferreiro, J., & Cuetos, F. (2014). Effects of the

psycholinguistic variables on the lexical decision task in Spanish: A study with 2,765

words. Behavior Research Methods, 46(2), 517-525.

Hargreaves, I. S., & Pexman, P. M. (2012). Does richness lose its luster? Effects of extensive

practice on semantic richness in visual word recognition. Frontiers in human

neuroscience, 6(234), 1-11.

Haro, J., Demestre, J., Boada, R., & Ferré, P. (2017). ERP and behavioral effects of semantic

ambiguity in a lexical decision task. Journal of Neurolinguistics, 44, 190-202.

Haro, J., Ferré, P., Boada, R., & Demestre, J. (2017). Semantic ambiguity norms for 530

Spanish words. Applied Psycholinguistics, 38(2), 457-475.

Hino, Y., Kusunose, Y., & Lupker, S. J. (2010). The relatedness-of-meaning effect for

ambiguous words in lexical-decision tasks: When does relatedness matter? Canadian

Journal of Experimental Psychology, 64(3), 180-196.

Hino, Y., & Lupker, S. J. (1996). Effects of polysemy in lexical decision and naming: An

alternative to lexical access accounts. Journal of Experimental Psychology: Human

Perception and Performance, 22(6), 1331.

Hino, Y., Lupker, S. J., & Pexman, P. M. (2002). Ambiguity and synonymy effects in lexical

decision, naming, and semantic categorization tasks: interactions between

orthography, phonology, and semantics. Journal of Experimental Psychology:

Learning, Memory, and Cognition, 28(4), 686-713.

36
Hino, Y., Pexman, P. M., & Lupker, S. J. (2006). Ambiguity and relatedness effects in

semantic tasks: Are they due to semantic coding? Journal of Memory and

Language, 55(2), 247-273.

Jager, B., & Cleland, A. A. (2016). Polysemy advantage with abstract but not concrete

words. Journal of psycholinguistic research, 45, 143-156.

Jastrzembski, J. E. (1981). Multiple meanings, number of related meanings, frequency of

occurrence, and the lexicon. Cognitive psychology, 13(2), 278-305.

Jastrzembski, J. E., & Stanners, R. F. (1975). Multiple word meanings and lexical search

speed. Journal of verbal learning and verbal behavior, 14(5), 534-537.

Joordens, S., & Besner, D. (1994). When banking on meaning is not (yet) money in the bank:

Explorations in connectionist modeling. Journal of Experimental Psychology:

Learning, Memory, and Cognition, 20(5), 1051-1062.

Kellas, G., Ferraro, F. R., & Simpson, G. B. (1988). Lexical ambiguity and the timecourse of

attentional allocation in word recognition. Journal of Experimental Psychology:

Human Perception and Performance, 14(4), 601-609.

Klepousniotou, E., & Baum, S. R. (2007). Disambiguating the ambiguity advantage effect in

word recognition: An advantage for polysemous but not homonymous words. Journal

of Neurolinguistics, 20(1), 1-24.

Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2014). lmerTest: Tests for random

and fixed effects for linear mixed effect models: R package version 2.0-6.

[Link]

Lin, C-J. C., & Ahrens, K. (2005). How many meanings does a word have? Meaning

estimation in Chinese and English. In J. W. Minett & W. S.-Y. Wang (Eds.),

Language acquisition, change and emergence: Essays in evolutionary linguistics (pp.

437–464). Hong Kong: City University of Hong Kong Press.

37
Lin, C-J. C., & Ahrens, K. (2010). Ambiguity advantage revisited: Two meanings are better

than one when accessing Chinese nouns. Journal of psycholinguistic research, 39(1),

1-19.

Masson, M. E., & Borowsky, R. (1995). Unsettling questions about semantic ambiguity in

connectionist models: Comment on Joordens and Besner (1994). Journal of

Experimental Psychology: Learning, Memory, and Cognition, 21, 509–514.

Masson, M. E. (1991). A distributed memory model of context effects in word identification.

In D. Besner & G. W. Humphreys (Eds.), Basic processes in reading: Visual word

recognition, (pp. 233-263). Hillsdale, NJ: Erlbaum.

Millis, M. L., & Button, S. B. (1989). The effect of polysemy on lexical decision time: Now

you see it, now you don’t. Memory & Cognition, 17(2), 141-147.

Parks, R., Ray, J., & Bland, S (1998). Wordsmyth English dictionary—Thesaurus.

[ONLINE]. Available: http:// [Link]/ [2016, May].

Pexman, P. M., Hino, Y., & Lupker, S. J. (2004). Semantic ambiguity and the process of

generating meaning from print. Journal of Experimental Psychology: Learning,

Memory, and Cognition, 30(6), 1252-1270.

Pexman, P. M., Siakaluk, P. D., & Yap, M. J. (2013). Introduction to the research topic

meaning in mind: Semantic richness effects in language processing. Frontiers in

human neuroscience, 7, 1-3.

Real Academia Española. (2014). Diccionario de la lengua española (23rd ed.). Madrid,

Spain.

Rodd, J. M., Gaskell, M. G., & Marslen-Wilson, W. D. (2002). Making sense of semantic

ambiguity: Semantic competition in lexical access. Journal of Memory and

Language, 46(2), 245-266.

38
 Rodd, J. M., Gaskell, M. G., & Marslen-Wilson, W. D. (2004). Modelling the effects of

semantic ambiguity in word recognition. Cognitive Science, 28(1), 89-104.

Rueckl, J. G. (1995). Ambiguity and connectionist networks: Still settling into a solution:

Comment on Joordens and Besner (1994). Journal of Experimental Psychology:

Learning, Memory, and Cognition, 21, 501–508.

Rubenstein, H., Garfield, L., & Millikan, J. A. (1970). Homographic entries in the internal

lexicon. Journal of verbal learning and verbal behavior, 9(5), 487-494.

Simpson, G. B. (1984). Lexical ambiguity and its role in models of word

recognition. Psychological bulletin, 96(2), 316-340.

Tamminen, J., Cleland, A. A., Quinlan, P. T., & Gaskell, M. G. (2006). Processing semantic

ambiguity: Different loci for meanings and senses. In Proceedings of the twenty-

eighth annual conference of the cognitive science society (pp. 2222-2227).

Yap, M. J., Pexman, P. M., Wellsby, M., Hargreaves, I. S., & Huff, M. J. (2012). An

abundance of riches: cross-task comparisons of semantic richness effects in visual

word recognition. Frontiers in Human Neuroscience, 6(72), 1-10.

39

View publication stats