Edited by: M. Teresa Espinal, Autonomous University of Barcelona, Spain
Reviewed by: Rosalind Jean Thornton, Macquarie University, Australia; Jacee Cho, University of Wisconsin-Madison, United States
This article was submitted to Language Sciences, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Negative Concord (NC) constructions such as
Human languages display diversity in whether and how they instantiate negative dependencies (
In (1), the preverbal negative marker
Other languages instantiate negative dependencies through Negative Polarity Item (NPI) constructions
In (2), the term
NC constructions are often modeled as a syntactic dependency between negative elements within a clause (e.g.,
English is among the languages which instantiate both NC and NPI constructions. In vernacular English varieties, spontaneous speech reflects variation in negative contexts between these two structure types, as in the following examples from
Speakers may even employ both construction types within a single utterance, as in the following example from an Appalachian English speaker (cited in
An important and distinguishing feature of English NC is its heavy social stigma (
This paper uses experimental means to explore an alternative hypothesis, one which does not assume a direct and causal link between NC unacceptability and ungrammaticality (
This section summarizes several relevant grammatical theories and experimental and psycholinguistic studies of NC and NPI constructions. The literature is vast, and we focus on those most relevant to our experiments. We begin with the assumption that grammars are “abstract descriptions of the representations built by the cognitive system” during language comprehension and production, rather than cognitively real, static references queried by the parser (
Many recent theories of NC model it as a syntactic Agree relation between negative elements within a clause (e.g.,
Syntactic Agree approaches to modeling NC posit that negative elements are lexically endowed with an uninterpretable feature which needs to be checked in the syntax. Under an Agree approach, the NC sentence in (6) would be modeled roughly as follows (cf.
Example (8) shows how the negative noun phrase
I didn–t eat nothing. (NC)
I ate nothing.
To explain the variation seen in (9) and (10), Tubau proposes a theory in which negative noun phrases such as
While vernacular English varieties are known for instantiating NC (
I didn’t see nobody.
Nobody didn’t see me.
The results of all three studies demonstrate that SE speakers reliably prefer NC interpretations for sentences like (11), but DN interpretations for sentences like (12).
The sentences in (11) and (12) illustrate a typological divide between what
I didn’t see anybody.
*Anybody didn’t see me
Example (14) shows that NPIs are unacceptable in canonical subject position. Note that (13), which is acceptable, is equivalent in meaning and nearly identical in form to (11), while unacceptable (14) is nearly identical in form to (12)
Maria didn’t drive.
Maria didn’t drive fast.
Maria didn’t drive fast and furiously.
The sets denoted by the predicate narrow from (15) to (17), and the entailments hold in that downward direction: If Maria did not drive (the widest set), then it must also be true that she did not drive fast (a narrower set), and that she did not drive fast and furiously (the narrowest set). Note that removing the negation voids this entailment pattern:
Maria drove.
Maria drove fast.
Maria drove fast and furiously.
It can be true that Maria drove, but that she drove slowly and cautiously, which means that (18) being true does not entail that (19) and (20) are also true.
Negation’s ability to trigger downward entailing inferences, Ladusaw proposes, is the property that allows it to license NPIs, its removal leading to unacceptability:
Mary didn’t drive any cars.
*Mary drove any cars
In addition to this semantic specification, there is also thought to be a syntactic requirement that the NPI be c-commanded by its licensor (
Further research on downward entailment for NPI licensing has revealed a number of apparent exceptions to the pattern, one of which is conditionals, which we employ in our experiment. The lack of straightforward downward entailingness in these contexts has led semanticists to expand, refine, or propose alternatives to this as a licensing condition (
Some recent psycholinguistic studies of NPIs have assumed the downward entailingness theory of NPI licensing in examining speakers’ processing of NPIs. Both
The structure in (24) shows the phrase
From the perspective of this paper, the importance of
Maria didn’t eat anything for lunch today.
Maria didn’t eat a damn thing for lunch today.
If Maria eats anything for lunch today, she’ll be able to work through the afternoon.
*If Maria eats a damn thing for lunch today, she’ll be able to work through the afternoon.
Sentences (25) through (28) contain the NPIs
Both structures are noun phrases (DPs) with a negation (NEG) directly modifying an abstract SOME.
Note now that the reversal structure in (30) has a second negation. Their proposal is that the outer negation cancels the force of the inner one, yielding a non-negative semantics. This model thus generates the correct truth conditions for sentences such as conditionals, in which NPIs are licensed. For example, in the sentence
A further benefit of the
In light of the English data examined here, a benefit of
Our experiments were designed to explore similarities and differences between overtly negative noun phrases and NPIs in direct object position under negation, a context for “strong” NPI licensing or unary NEG NPI structures, and under conditionals, a context for “weak” NPI licensing or reversal structures (
Do English speakers access parallel meanings for NPI and NC constructions under negation (i.e., contexts for unary NEG structures), despite asymmetries in the acceptability of these constructions?
Do these same English speakers readily distinguish between the meanings of NPIs and overtly negative noun phrases in “weak” (reversal) licensing contexts, which do not parallel NC?
Thirty participants (10 women, 20 men) were recruited through Amazon Mechanical Turk (AMT) for the main experiment, and a further 15 (5 women, 10 men) were recruited for the follow-up. To participate, speakers had to confirm that they were at least 18 and spoke American English natively. Completing the online survey took approximately 30 min, and participants were paid $6 for their time.
All participants had spent most or all of their lives in the US. Their answers to free response questions about cities and regions where they had lived indicated that 17 had spent the majority of their childhoods in the south (including 4 in Florida and 3 in Texas), 9 in the Midwest, 7 in the midatlantic, 6 on the west coast, 1 each in the northeast, great plains and southwest, and that 3 had spent similar amounts of time in two or more regions. Four participants reported familiarity with a language other than English, two heritage language speakers (Chinese, Spanish), and two foreign language learners (Spanish, German).
Participants were between 24 and 72 years old (main study mean = 38.5 years, follow-up study mean = 40.2 years), and the majority had completed either high school (
An additional 5 participants, 4 from the main experiment and 1 from the follow-up, completed the task and were paid, but were excluded from the final dataset for failing to achieve 80% accuracy on the catch trials [described below, see (31g)]. These participants gave ratings of 5 or higher (i.e., felicitous) to 4 or more of the 16 fillers that were designed to have infelicitous continuations, or gave ratings of 4 or lower (i.e., unacceptable) to 4 or more of the first clauses of these fillers, despite the fact that these first clauses were unremarkable English sentences. This indicated either that they were not reading carefully, or that they had misunderstood the task.
In a post-survey language questionnaire, participants were asked how likely they and their family and friends were to use NC and NPI constructions to communicate a negative meaning, on a scale from 1 (never) to 7 (always). Ratings were low for use of NC (participants’ median = 1, mean = 1.84; family and friends median = 2, mean = 2.6), and high for use of NPI constructions (participants’ median = 6, mean = 5.9; family and friends median = 6, mean = 6.0). Given the heavy social stigma associated with NC, we interpret these responses with caution, but they suggest that the group of speakers who participated in our experiments can be characterized as primarily non-NC users.
We designed two experiments to explore our research questions. The main experiment compared participants’ ratings of three noun phrase types in negative contexts (a context for “strong” NPIs, or unary NEG structures) and in conditionals (a context for “weak” NPIs or reversals). The follow-up experiment further explored the acceptability of negative noun phrases in conditionals, and participants’ interpretation of NPIs in these non-negative contexts. Both experiments included 48 critical sentences and 112 fillers. All sentences contained two clauses, the second of which described a consequence of or context for the first. See
For critical sentences in the main survey, the first clause was either conditional or negative, and the direct object was a DP of one of three types: bare plural (
We were particularly interested in the comparison between NPIs and negative noun phrases in conditional and negative contexts, since this would show us whether speakers calculate parallel truth conditions for NC and NPI constructions in “strict,” unary NEG contexts, and whether these same speakers also calculate opposite truth conditions when these noun phrase types appear the “weak” reversal context of conditionals. Constructions with a bare plural, which have the same truth conditional meaning as the NPIs in these sentences but no linguistic dependency, were employed as a control.
Critical sentences in the follow-up survey were derived from those in the main survey, by pairing the conditional first clauses with the single-negation continuations, as shown in (34). This was intended to render the negative noun phrases fully felicitous in the conditional sentences, and the NPIs and bare plurals infelicitous. Because there were only three conditions, participants saw twice as many sentences per condition as in the main study, and only half as many participants were needed to obtain the same number of observations per condition.
Fillers were identical for the two surveys and were designed with the same two-clause structure as critical sentences. They included a variety of features intended to blend with the critical items, including several different subordinating conjunctions, universally quantified direct objects, and a single negated auxiliary without quantified or bare plural direct objects, as shown in (35). Of the 112 fillers, 96 were designed to have felicitous continuations (35a-f), and the remaining 16 were designed to be fully acceptable but have infelicitous continuations (35g). This created a similar proportion of infelicitous continuations in the filler items as we predicted there would be in the critical items (1/7 and 1/6 respectively). The 16 “mismatch” fillers served as catch trials and allowed us to exclude participants who had misunderstood the task or were not attending it carefully (
Upon selecting the survey, AMT workers were directed to a Qualtrics survey link. They first read and acknowledged an informed consent statement, then proceeded to the survey
The survey was preceded by four practice trials with feedback, to familiarize participants with the task. Of the 4 practice trials, two had low acceptability first clauses (glaring word order errors), and two had high acceptability first clauses. This was crossed with plausibility of the consequence, to demonstrate the independence of the two judgments.
The body of the survey included the 112 fillers and 48 critical items presented in a fully random order and was followed by a short debriefing and language history questionnaire. Upon completion of the survey, participants were given a code to enter into the AMT interface in order to get their payment.
Both acceptability and felicity ratings were on a 7-point Likert scale, and were therefore analyzed using ordinal rather than linear regression techniques (
This analysis technique differs in several ways from other common approaches to analyzing Likert data. Most importantly, in contrast to linear modeling techniques, ordinal modeling does not make the assumption that participants treat the ratings as equally spaced. That is, ordinal modeling allows for the possibility that participants will, for instance, be particularly hesitant to give the minimum rating, effectively making the distance between 1 and 2 larger than the distance between 2 and 3. Second, raw ratings are entered into the model, rather than z-scored ratings. Z-scoring serves two purposes when analyzing ratings using linear models: to make the measure more continuous and therefore more appropriate for a non-ordinal analysis, and to factor out between-participant variation. The use of ordinal analysis obviates the need for continuity, and mixed model approaches, whether linear or ordinal, take between-participant variation into account using random effects.
When interpreting model output, note that estimates are threshold changes in terms of shared standard deviation. Thus, while they are not readily interpretable as predicted change in score or probability (as might be the case in a well-coded linear model), they are readily comparable to each other within a model: an estimate of 3 indicates that a factor has twice as large an influence on the thresholds as a factor with an estimate of 1.5.
For other linguistic studies applying cumulative link mixed models to Likert scale judgment data, see
To explore the relationship between participants’ acceptance of English NC and their ability to interpret it as truth conditionally equivalent to negative NPI constructions, we compared participants’ acceptability ratings of three types of direct object (overtly negative noun phrases, NPIs, bare plurals) in negative and conditional sentences. Each initial clause was followed by a second clause that, for the negative sentences, was compatible with a single negation reading, and for the conditional contexts was compatible with a no-negation reading. We predicted that participants would rate all first clauses as relatively acceptable except for the negative noun phrase in a
In the follow-up survey, we paired the single-negation compatible consequences with the conditional first clauses [see (30)] in order to confirm that negation is not uniformly less acceptable in conditionals, and that participants treat NPIs and negative noun phrases as opposites in non-negative conditional statements, in contrast to the negative contexts in the main study where we predicted they would be treated as syntactic variants.
Crucially, we predicted a disconnect between participants’ acceptability ratings for NC sentences, and their felicity ratings for the single negation continuations in the main study, which would be instantiated as low acceptability but high felicity ratings for negative sentences with negative noun phrases. High felicity ratings would indicate that participants readily achieved the intended reading of the NC construction, and would suggest that low acceptability ratings are likely more the result of social pressure than the speaker’s inability to generate the structure.
Raw acceptability and felicity judgments for the main survey on a 7-point Likert scale, with box plots showing overall quartiles and median. Values for conditional sentences are shown in light green, and for negative sentences in dark blue.
Other patterns visible in the graph include very high acceptability ratings for both bare plurals and NPIs in both conditional and negative sentences (all medians = 7), with the most consistently high acceptability ratings for bare plurals in conditional sentences, and generally high felicity ratings for consequences following bare plurals and NPIs (median = 6 for conditional-NPI, 7 elsewhere). Also note that there is more spread in the generally low ratings for negative NPs in the negative sentences (median = 3) than one might expect for something truly unacceptable. Compare, for instance, the consistent, very low felicity ratings (median = 1) for the truly infelicitous continuations, following conditional sentences with negative noun phrases. We return briefly to this variability in the discussion.
To explore these patterns statistically, we fit separate cumulative link mixed models for acceptability ratings and felicity ratings (see section Analyses). For both models, predictor variables were the two-level factor sentence type (conditional, contrast code -0.5 vs. negative, contrast code 0.5), and the three-level factor NP type, coded as two Helmert contrasts, the first comparing negative noun phrases to NPIs and bare plurals together (“negative-other,” negative noun phrases, 0.67 vs. NPIs and bare plurals, both -0.33), and the second comparing NPIs to bare plurals (“NPI-bare,” NPIs, -0.5 vs. bare plurals, 0.5, negative NPs, 0), as well as the interactions of sentence type and the NP type contrasts. The model included random intercepts for item and participant and the random slopes of sentence type by item and of NP type, sentence type and their interaction by participant.
Model results are shown in
Model results for acceptability ratings in the main survey.
negative-other | –2.40 | 0.20 | –11.78 | 57.85 | < 0.00001 |
NPI-bare | 0.33 | 0.16 | 2.03 | 4.05 | 0.044 |
–0.57 | 0.13 | –4.21 | 14.27 | 0.0002 | |
negative-other | –1.75 | 0.25 | –7.14 | 33.26 | < 0.00001 |
NPI-bare | –0.52 | 0.31 | –1.68 | 2.65 | 0.10 |
Planned comparisons further explored the key interaction and supported this conclusion. Three models, identical to the main model except for their contrast codes, were conducted to examine the simple main effects of both NP type contrasts in conditional and in negative sentences, and to examine the simple main effect of sentence type on the acceptability of negative noun phrases. These models revealed that negative noun phrases were less acceptable than NPIs and bare plurals in negative sentences [
For felicity ratings, the continuations of the conditional sentences with negative noun phrases were predicted to be infelicitous, which should result in a reliable interaction between sentence type and the negative-other NP type contrast. This prediction was supported by the model results, shown in
Model results for felicity ratings in the main survey.
negative-other | –1.64 | 0.13 | –12.69 | 57.46 | < 0.00001 |
NPI-bare | –0.05 | 0.10 | –0.49 | 0.25 | 0.62 |
1.51 | 0.15 | 9.91 | 51.02 | < 0.00001 | |
negative-other | 2.26 | 0.29 | 7.85 | 43.09 | < 0.00001 |
NPI-bare | –0.51 | 0.19 | –2.65 | 7.37 | 0.007 |
This primary model was again followed by further analyses to explore the interactions in the data. These revealed a reliable effect of the negative-other NP type contrast in both conditional [
Again, intriguingly and consistent with the overall interaction between sentence type and the NPI-bare contrast, there were differences between NPIs and bare plurals. There was a reliable effect of the NPI-bare contrast in negative sentences [
One possible explanation for the pattern of felicity ratings in the main study is that the consequences of conditional sentences with negative direct objects were rated as infelicitous at least partly because negation is difficult to process, and this was exacerbated by the presence of the conditional. The follow-up survey was designed to confirm first that sentences with negative noun phrases were not inherently less felicitous under conditionals, and further, to confirm that speakers understand when negative noun phrases are equivalent in meaning to NPIs and when they are not.
Raw acceptability
We again fit cumulative link mixed models of acceptability ratings and of felicity ratings, this time with a single fixed effect predictor, the three-level Helmert-coded factor NP type. The models had random intercepts for participants and items, and random slopes for NP type by participant. For acceptability ratings, the model revealed only a marginal main effect of the NPI-bare NP type contrast (model results are shown in
Model results for acceptability ratings in the follow-up survey.
negative-other | –0.31 | 0.27 | 2.08 | 1.21 | 0.27 |
NPI-bare | 0.51 | 0.25 | –1.15 | 3.53 | 0.06 |
For felicity ratings, there was no reliable difference between bare plurals and NPIs, but there was a reliable difference between negative NPs and the other NP types (model results are shown in
Model results for felicity ratings in the follow-up survey.
negative-other | 2.67 | 0.26 | 10.43 | 33.07 | < 0.00001 |
NPI-bare | –0.01 | 0.13 | –0.09 | 0.01 | 0.93 |
Our main experiment involved two comparisons, one which compared negative noun phrases, NPIs, and bare plural controls under negation, and another which compared these same elements under non-negative conditionals. We asked whether speakers would calculate parallel truth conditions for NPIs and overtly negative noun phrases under negation (a context for unary NEG structures), and whether these same speakers would calculate opposite truth conditions for these words in non-negative conditionals (a context for reversal structures). We first discuss the comparison which involved negative dependencies.
Comparison of the three argument types in syntactically negative contexts revealed an asymmetry which can inform our understanding of the relationship between NC and NPI constructions: Participants’ acceptability ratings of socially stigmatized NC constructions were low, and their felicity ratings of consequences which correspond to the NC interpretation for these same constructions were high. Furthermore, while NC and NPI constructions were rated on opposite sides of the scale in acceptability, with NC on the low side and NPIs on the high side (and similar to bare plural controls), the consequences of all construction types were given relatively high felicity ratings.
Regarding the asymmetry between NC acceptability and felicity, we note that this finding both supports and complements previous work which compared NC with DN, its truth conditional opposite (
In the current study, there were no DN interpretations elicited from speakers during the course of the experiment, and given that participants reliably judged the single negation consequence of NC constructions as felicitous (which would have been infelicitous on a DN reading), we can assume that, at least for the most part, participants did not generate DN meanings for the items with two syntactic negations. Other differences between this study and those previous studies include the fact that participants judged NPI and bare plural sentences as well as NC, and that their judgments were made on the basis of the NC interpretation’s felicitousness as determined by a following consequence, as opposed to a preceding context. In the context of previous studies in which speakers reliably prefer NC over DN interpretations in a range of measures, the fact that a distinct design led to similar results thus further confirms the robustness of speakers’ readiness to interpret NC constructions as singly negative, and provides complementary support for the hypothesis that speakers who do not accept NC nevertheless have grammatical knowledge of it.
Regarding the interactional aspect of the asymmetry in negative contexts, in which NC acceptability and felicity were at opposite sides of the scale, while NPI (and bare plural) acceptability and felicity were on the same side, this shows that participants readily accessed the same truth-conditional meaning for all three NP types under negation, despite reliable differences in their acceptability. It should be noted that there was in fact a small but statistically reliable difference between NPI and bare plural felicity in negative contexts on the one hand, and NC felicity on the other. We believe this difference is best explained as a carryover effect of the strong unacceptability of NC. This is particularly likely since, as explained in the methods section, participants still had the critical sentence in view when judging the consequence.
The interaction between NC and NPI constructions in negative contexts also illustrates a more general methodological point, namely, that examining acceptability in isolation from meaning can obscure speaker knowledge of a construction type (especially where that construction type is socially stigmatized). In this case, the social stigma associated with English NC appears to be a primary force shaping speakers’ acceptability ratings. Yet despite the strength of this social stigma, participants drew a clear distinction between the acceptability of NC and its meaning in context. NC thus provides an example of a construction type for which binary or overall acceptability and interpretation are unrelated. We extend this to suggest that NC also provides an example of a construction type for which overall acceptability and
To conclude this subsection, we note that there was substantially more spread in the negative sentence-negative noun phrase (i.e., NC) acceptability ratings than what might be expected for something that is outright ungrammatical (e.g., sentences with glaring word order violations such as
Items where the NPIs, overtly negative noun phrases, and bare plurals appeared under conditional contexts displayed two clear additional asymmetries beyond the ones found in negative contexts. In the main experiment, the clearest asymmetry was again interactional in nature, between the NPIs and bare plurals on the one hand, and the overtly negative noun phrases on the other. These were all relatively acceptable, with mean scores well above the middle of the scale, but in the main study, the contexts were designed to make the NPIs and bare plurals felicitous and the overtly negative noun phrases infelicitous. Unsurprisingly, participants responded in reliable fashion to this experimental manipulation, rating consequences following
The follow-up experiment was designed to inform the results of the main experiment, and to provide a more complete picture of speakers’ understanding of where contexts for NPIs and overtly negative noun phrases do and do not overlap. Reversing the truth conditions for the consequence from the main experiment, we expected that the non-negative NPI (a reversal structure), and not the (unary NEG) negative noun phrase, would be infelicitous. Participants again behaved as predicted, rating consequences of NPI and bare plural
Before turning to theoretical implications, we note an additional asymmetry that our experiment was not explicitly designed to reveal: Though acceptable overall, overtly negative noun phrases were slightly less acceptable than NPIs and bare plurals in the main experiment conditional contexts. One potential explanation for this is that negation makes things more difficult to process (e.g.,
One explanation for the fact that participants gave similar felicity judgments for the NC and NPI constructions in negative or “strict” contexts is that their grammars represent NC and NPI constructions as syntactic variants with the same underlying form. This explanation finds its theoretical basis in
The process governing the spell out of the lower negation in unary NEG structures may be grammatical in nature, where SE grammars have a constraint that prohibits them from pronouncing the lower negation which is absent from vernacular varieties, or it may be a purely socially governed phenomenon which over time has been conventionalized in SE, with the effect of masking a direct underlying grammatical connection between these two construction types. Whether the differences between these two surface forms are derived by grammatical or social pressures, a plausible explanation for our results is that speakers generated the same negative dependency in both the NC and the NPI constructions in negative contexts, and this was reflected in their felicity judgments. Concurrently, their clear intuitions about the opposite meanings of negative noun phrases and NPIs in conditionals, a “weak” licensing context, supports the hypothesis that they also have two distinct underlying representations for NPIs, a unary NEG structure and a semantically non-negative reversal, and they select the item analogous to the reversal structure for these conditional contexts.
We can also view our results in light of
With regard to purely semantic theories of NPI licensing, in addition to finding experimental evidence for a parallel to the calculation of downward entailing inferences (
The experiments we reported here revealed asymmetries in the acceptability and felicity of NC and NPI constructions. We have provided evidence that speakers understand when the truth conditions for NC and NPI constructions overlap, and when they do not. The results have both methodological and theoretical implications. On the methodological side, they demonstrate a clear case where there is no straightforward causal link between acceptability and grammaticality, and concurrently how judgments of meaning can inform theories of grammar in cases where acceptability judgments fail. On the theoretical side, they show how the set of facts that grammatical theories should be capable of modeling within a single system includes NC and NPI constructions, and in the context of previous studies, also DN. We further discussed how the system in
The datasets generated for this study are available on request to the corresponding author.
Ethical review and approval were obtained for the study on human participants in accordance with the local legislation and institutional requirements, and the study was determined to be exempt from continuing review. Written informed consent to participate in this study was not required in accordance with the national legislation and institutional requirements.
Both authors made substantial contributions to the development of this work, including experimental design, data collection and analysis, drafting and revising the manuscript, and agreed to be accountable for all aspects of the work.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The Supplementary Material for this article can be found online at:
Typical of Romance languages like Catalan, French, Italian, Spanish, and Brazilian Portuguese (e.g.,
See
See
Following AAPCAppE citation conventions, tokens are followed by the corpus, subcollection, and speaker initials, along with a numerical token identifier.
We follow
In a study that compares children and adults,
Didn’t nobody see me. ‘Nobody saw me.’
The specifics are beyond the scope of this paper to discuss, but the general conclusion is that SE speakers prefer NC interpretations for sentences in which the negated auxiliary precedes (and c-commands) the negative noun phrase.
Vernacular Englishes also optionally instantiate the “strict” NC pattern (
Sentence (14) is argued to be unacceptable because it does not meet the c-command requirement for NPIs and their licensors. We discuss this further below.
The term “downward entailing” is used synonymously with “monotone decreasing” (
For reasons of space and lack of immediate relevance we set aside here and throughout instances of “free choice Maria drove any car she wanted.
For a semantic account of free choice
See
A detailed summary of
Capital letters denote emphasis, which according to
See also
See
The # symbol marks infelicity. We include it here for expository purposes. It was not included in the actual experiment.
This survey was conducted under the supervision of the Penn State IRB, which deemed it to be minimal risk and therefore exempt from requirements for written documentation of informed consent. Participants indicated their understanding of the consent document and willingness to participate by simply continuing with the survey.
Previous studies have similarly elicited two separate responses of a different nature for a single item. See for example,
We are currently collecting data in a parallel eye-tracking study. Tracking participants’ eye-movements as they read these same stimuli will allow us to more directly investigate just how readily this interpretation is generated.