Linguistic and Social Meaning Match: An experiment on modal concord in English

Mingya Liu & Stephanie Rotter*

Abstract. Modal concord (MC) refers to the phenomenon where two modal elements
of the same flavor and force in a sentence yield an interpretation of single modality
(SM). In this paper, we report on an experimental study on MC in English, address-
ing their linguistic and social meaning. Our results show a strengthening effect by
necessity MC and a weakening effect of possibility MC in that significantly higher
speaker commitment ratings were received for necessity MC vs. SM constructions
(i.e., must certainly vs. must) with the reverse pattern for possibility modal construc-
tions (i.e., may possibly vs. may). Furthermore, MC and SM were shown to differ in
social meanings, suggesting a correlation between the meaning strength of a linguistic
expression and the social perception of the speaker.
Keywords. modal concord; social meaning; experiment; English

1. Introduction. Modality is one of the key empirical areas in linguistics, and also one of the
most complex ones. This paper deals with the phenomenon ‘modal concord’ (MC) and addresses
the linguistic and social meaning of MC constructions.

As shown in (1), a sentence can be unmodalized (1-a), contain a single modal (SM) element
such as a modal auxiliary or adverb (1-b), or include two modal elements including a modal aux-
iliary and an adverb (1-c).

(1) a. It is the case.
b. It may be the case. / It is possibly the case. (SM)
c. It may possibly be the case. (MC)

In (1-c), the two modal elements are both weak epistemic modals and their co-occurrence can
yield the same meaning as just one of them—this phenomenon is labeled as MC in the literature
(Geurts & Huitink 2006, Zeijstra 2007, Yalcin 2007), with the discussion dating back to at least
the 70s, see an earlier quote and a more recent one below.

“In most dialects of English not more than one modal verb can occur within the same
clause. But both a modal verb and a modal adverb may be combined. When this
happens a distinction is to be drawn between modally harmonic∗ and modally non-
harmonic∗ combinations. For example, ‘possibly’ and ‘may’, if each is being used
epistemically, are harmonic, in that they both express the same degree of modality,
whereas ‘certainly’ and ‘may’ are, in this sense, modally non-harmonic. It has been
pointed out by Halliday (1970a: 331) that the adverb and the modal verb may, and
normally do, “reinforce each other” in a modally harmonic combination; so that, ... or
there is a kind of concord running through the clause, which results in the double
realization of a single modality.” (Lyons 1977: 807-808).

*This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – SFB
1412, 416591334. Authors: Mingya Liu, Humboldt-Universität zu Berlin (mingya.liu@hu-berlin.de) & Stephanie
Rotter, Humboldt-Universität zu Berlin (rotterst@hu-berlin.de).

Proceedings of ELM 3: 224-235, 2025

c©2025 Mingya Liu and Stephanie Rotter
Published by the LSA with permission of the author(s) under a CC BY license. 224

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


“if two modal elements are of the same modal type (epistemic/ deontic/...) and have
similar quantificational force (universal/existential), the most salient reading is mostly
not a cumulative one but a concord reading, where the semantics seems to contain only
one modal operator.” (Zeijstra 2007: 137)

With the ‘meaning-equivalence’ assumption for MC and SM, we can make the inference that
the optional adverb possibly in (1-c) is semantically vacuous. This is debatable. For example, some
authors hold the view that multiple modal expressions have one semantic role but the adverb does
make a semantic contribution, giving rise to speaker bias (i.e., weakened or strengthened speaker
commitment to the modified proposition), see the discussion of ‘modal spread’ by Giannakidou &
Mari (2018). Despite the numerous attempts to integrate MC into semantic theories, the empir-
ical picture remains unclear: Do MC constructions really share the same semantics as their SM
counterparts?

Another question that arises naturally is the use of MC in comparison to SM: Assuming MC
and SM are ‘functional equivalents’, that is, variants of a linguistic variable with the same seman-
tic interpretation (Labov 2006: 145-146), why would a speaker choose to use MC over SM? This
question is specific to MC in the modal domain, as it is available across languages and varieties
of English, in contrast to the phenomenon of ‘multiple modals’ (e.g., might could), which is much
more restricted, that is, only possible in some varieties of English (see the first sentence in Lyons’
1977 quote above, and Kortmann & Schneider 2004 on colloquial American and Appalachian
American English) and in general, in a smaller set of natural languages. Based on the sociolin-
guistic literature and more recent work on register in Lüdeling et al. (2022) and Pescuma et al.
(2023), we assume that the choice between MC and SM can be register-driven, that is, the speaker
chooses one or the other depending on the properties of the situational contexts with regard to, for
example, interlocutor relations or communicative purposes. Furthermore, the choice can be better
understood in relation to the social meaning, that is, “the set of inferences that can be drawn on
the basis of how language is used in a specific interaction” (Hall-Lew et al. 2021; 3). For example,
Glass (2015) reports on a social meaning study of three different semi-modals of universal force in
English, see (2), and argues that “you need to has a slightly different semantics than you have/got
to, and that this subtlety gives it a unique social meaning that depends on whether the speaker
is licensed to tell the hearer what’s good for him” (p.87). Similarly to that study, our study also
addresses the social meaning of modal expressions of the same—existential or universal—force:
More specifically, when a speaker utters MC in a specific context, what pragmatic inferences can
be drawn about the speaker’s backgrounds and personae?

(2) You {need to /have to/gotta} admire her for persevering. (Glass 2015: example (1))

In the following, we report on an experimental study on MC in English, which is guided by the
following questions: (RQ1) How are MC constructions interpreted? We tested MC and SM con-
structions, using speaker commitment ratings, and found that their interpretations differ. (RQ2)
How are MC constructions perceived? We tested the social meaning of MC and SM, as well as
that of possibility and necessity modals, and found that their social meanings differ. (RQ3) How
does the perception of doubling constructions differ? We will discuss the results of MC in the
current study with the results of negative concord (NC) in our previous work (Rotter & Liu 2025,

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 225

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


2024) and show that the social meanings of MC and NC differ.
The paper is structured as follows: Section 2 presents the method and the results of the ex-

periment. Section 3 discusses the results, the limitations and future outlooks of the study, and
concludes the paper.

2. Experiment.

2.1. MATERIAL AND DESIGN. The experiment was implemented with PCIbex farm and hosted
on its platform (https://farm.pcibex.net/). We used a 2x2-factorial design with the
factors FORCE (possibility vs. necessity) and NUMBER (of used modal elements, i.e., MC vs.
SM) in a Latin Square design with four lists. We used 12 critical, see (3), and 29 filler items of
similar structure. (S1) remained the same across all items of the study, see (3). (S2) contained the
FORCE and NUMBER manipulation. (Q1) was used to measure the interpretation of the sentence
in terms of speaker commitment ratings (Liu et al. 2021) and (Q2) to measure the grammaticality
ratings.

(3) (S1) Somebody says:
(S2) “I / may possiblyMC / have gotten / the wrong address.” (possibility)

“I / maySM / have gotten / the wrong address.” (possibility)
“I / must certainlyMC / have gotten / the wrong address.” (necessity)
“I / mustSM / have gotten / the wrong address.” (necessity)

(Q1) Does the person believe they have gotten the wrong address?
(Q2) Is the sentence grammatical?

2.2. PROCEDURE. The experiment consisted of four parts: (P1) consent, (P2) two practice sen-
tences to get familiarized with the experimental set-up, (P3) main study, and (P4) demographic
and language background survey.

Participants were instructed to use the space bar to continue. Each trial in (P2/3) started with
a fixation cross in the middle of the screen. Then (S1) appeared on the screen. (S2) was presented
as a self-paced reading task; the length and position of each chunk was indicated by a line on
the screen. Previous chunks disappeared so that only the current chunk was visible. Finally, the
entire sentence (S2) appeared on the screen together with the following social meaning measures:
social backgrounds of the speaker (low/high socioeconomic status and low/high education), and
persona of the speaker (in/formal, un/confident, im/polite, un/friendly, cold/warm, un/cool, and
obedient/rebellious); the measures appeared in groups of three and within each group, the order
was randomized. In the two last screens, the participants first rated the speaker commitment, that
is, interpretation (Q1-Certainly no/Certainly yes), and then the grammaticality of the sentence
(Q2-Certainly no/Certainly yes). We chose this order to allow for unbiased speaker commitment
ratings. All the measures and questions used a 7-point Likert scale in which the end- and midpoints
were labels (e.g., 1: ungrammatical – 4: undecided – 7: grammatical). Participants used the mouse
click to provide their answer.

2.3. PARTICIPANTS. We collected data from 104 native speaker of US English on the crowd-
sourcing platform Prolific (https://www.prolific.co/). The experiment took roughly
35 minutes and participants received monetary compensation. All participants provided their in-
formed consent as approved by the Ethics Committee of the Deutsche Gesellschaft für Sprach-

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 226

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


wissenschaft (DGfS) in the context of SFB 1412 ‘Register’. Based on the inclusion criteria
(see 2.4 for details), we analyzed the data of 93 participants from 37 of the 50 US states (mean
age=39.2/SD=10.9, range=[19, 63]; female N=48, male N=43, non-binary N=2).

2.4. DATA ANALYSIS. The data was processed and analyzed with the open source software ‘R’
(version 4.1.1, R Core Team 2024) in the RStudio environment (RStudio Team 2024). We defined
demographic (native English speaker and aged between 18 to 65 years) and attendance (75% of the
chunks were attended for at least 150msec) inclusion criteria: First, all participants matched the
demographic criteria. Second, the data of 11 participants were excluded based on the attendance.
Furthermore, we removed the data of 254 (10.2%) trails, in which at least one chunk was attended
for less than 150msec.

We calculated separate models for each measure with the remaining data in the cumulative link
function model framework (Howcroft & Rieser 2021, Liddell & Kruschke 2018) using the package
‘ordinal’ (Christensen 2019). The link function of each model was determined as the one with
the highest log-likelihood value among the five possible link functions (i.e., probit, logit, cauchit,
loglog, and cloglog) (Christensen 2019). Each model contained the sum-coded factors FORCE
(possibility: 0.5, necessity: −0.5) and NUMBER (MC: 0.5, SM: −0.5), as well as their two-way
interaction FORCE×NUMBER (2-int). If the interaction turned out significant, we conducted a
sub-analysis by splitting the data along the significant main factor. The sub-analysis included the
same coding of the main effects.

We used the most parsimonious model approach to obtain the random effect structures; the
used model structures are indicated in the respective result section. Log-likelihood ratio test com-
parisons of nested models (Bates et al. 2018) were used to obtain p-values, which were defined
as significant at a value below 0.05. We reduced the complexity of models if their NULL-model
did not converge. We used the model with the smaller Akaike Information Criterion (AIC) if the
model choice was between subject or item intercepts. All statistical values of means, estimates and
the like are rounded to the second decimals except for p-values smaller than 0.01.

3. Results. Below, we first present the results of the interpretation and grammaticality measures
and then the results of the social meaning measures.

3.1. INTERPRETATION AND GRAMMATICALITY. The output of the models using interpretation
(Q1) and grammaticality (Q2) ratings as dependent variable are shown in Table 2.

For the interpretation ratings, the logit-link function model including random subject inter-
cepts and slopes for FORCE, NUMBER, and their interaction fit the data best. The results showed
a significant main effect of FORCE in that possibility conditions were rated lower than necessity
conditions (β̂=−3.65, χ2(1)=95.99, p<0.001). There was a main effect of NUMBER in that MC
was rated higher than SM (β̂=0.60, χ2(1)=15.91, p<0.001). Furthermore, the 2-way interaction
turned out significant (β̂=−1.85, χ2(1)=41.51, p=<0.001), which results from a cross-over effect:

• Weakening effect in possibility modals: MC received significantly lower speaker commit-
ment ratings than SM (β̂=−0.35, χ2(1)=6.79, p<0.009)

• Strengthening effect in necessity modals: MC received significantly higher speaker commit-
ment ratings than SM (β̂=1.50, χ2(1)=36.92, p<0.001).

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 227

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


Figure 1: By-subject means (transparent dots) of the rating measures in comparison to overall
means (opaque dots with error bars). Panel A depicts the ratings of the interpretation and gram-
maticality, Panel B those of the social background measures, and Panel C those of the the persona
measures. The x-axis indicates the specific measures. The top scale shows the ratings from the
necessity conditions (abbreviated as nece), the one below those of the possibility conditions (ab-
breviated as poss). The colors indicate the factor NUMBER with MC (i.e., modal concord) in blue
and SM (i.e., single modal) in yellow. The y-axis depicts the ratings on a 7-point Likert scale. SES
abbreviates socioeconomic status.

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 228

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


n Med Mean SD SE Med Mean SD SE
FORCE NUMBER Interpretation Grammaticality

poss MC 542 5 5.11 0.96 0.04 6 5.06 2.05 0.09
poss SM 541 5 5.22 0.98 0.04 7 6.45 1.09 0.05
nece MC 536 7 6.40 0.90 0.04 6 4.90 2.18 0.09
nece SM 539 6 6.12 0.85 0.04 7 6.41 1.16 0.05

Socioeconomic status Education
poss MC 542 5 4.73 1.30 0.06 5 4.69 1.37 0.06
poss SM 541 5 4.87 1.11 0.05 5 4.94 1.12 0.05
nece MC 536 5 4.85 1.40 0.06 5 4.80 1.46 0.06
nece SM 539 5 4.87 1.18 0.05 5 4.89 1.19 0.05

Formality
poss MC 542 5 4.73 1.61 0.07
poss SM 541 5 4.84 1.38 0.06
nece MC 536 5 4.99 1.66 0.07
nece SM 539 5 4.76 1.40 0.06

Politeness Confidence
poss MC 542 5 5.36 1.17 0.05 4 4.03 1.61 0.07
poss SM 541 5 5.45 1.11 0.05 4 4.36 1.51 0.06
nece MC 536 5 5.39 1.16 0.05 6 5.48 1.41 0.06
nece SM 539 5 5.28 1.15 0.05 5 5.19 1.39 0.06

Friendliness Warmth
poss MC 542 5 4.92 1.17 0.05 5 4.82 1.19 0.05
poss SM 541 5 5.03 1.14 0.05 5 4.94 1.16 0.05
nece MC 536 5 4.76 1.21 0.05 4 4.62 1.25 0.05
nece SM 539 5 4.97 1.15 0.05 5 4.86 1.17 0.05

Coolness Rebelliousness
poss MC 542 4 4.29 1.33 0.06 3 3.11 1.33 0.06
poss SM 541 4 4.56 1.27 0.05 3 3.09 1.34 0.06
nece MC 536 4 4.20 1.32 0.06 3 3.04 1.34 0.06
nece SM 539 4 4.53 1.27 0.05 3 3.12 1.36 0.06

Table 1: Descriptive statistics of the interpretation (Q1), grammaticality (Q2), social background,
and persona ratings. Poss stands for possibility conditions, nece for necessity, MC for modal
concord, and SM single modal. Med abbreviates median, SD standard deviation, and SE standard
error.

For the grammaticality ratings, the logit-link function model including random subject and
item intercepts, both with slopes for FORCE and NUMBER fit the data best. The results showed
a significant main effect of NUMBER in that MC was rated less grammatical than SM (β̂=−3.48,
χ2(1)=83.01, p<0.001).

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 229

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


Fixed effects Model comparison Fixed effects Model comparison
β̂ SE χ2(1) p-value β̂ SE χ2(1) p-value

Interpretation Grammaticality
FORCE −3.65 0.30 95.99 <0.001⋆ 0.21 0.29 0.53 0.47

NUMBER 0.60 0.15 15.91 <0.001⋆ −3.48 0.36 83.01 <0.001⋆
2-int −1.85 0.28 41.51 <0.001⋆ 0.15 0.23 0.43 0.51

Table 2: Output of the model using interpretation and grammaticality ratings. 2-int abbreviates the
two way interaction FORCE×NUMBER. SE stands for the standard error.

3.2. SOCIAL BACKGROUND. The output of the models using social background ratings as de-
pendent variables are shown in Table 3.

Fixed effects Model comparison Fixed effects Model comparison
β̂ SE χ2(1) p-value β̂ SE χ2(1) p-value

Socioeconomic status Education level
FORCE −0.15 0.05 9.13 <0.003⋆ −0.11 0.05 4.84 < 0.03⋆

NUMBER −0.01 0.05 0.01 0.92 −0.06 0.05 1.25 0.26
2-int −0.26 0.10 6.58 0.01⋆ −0.31 0.10 9.62 <0.002⋆

Table 3: Output of the model using the social background ratings. 2-int abbreviates the two way
interaction FORCE×NUMBER. SE stands for the standard error.

For the SES and education level ratings, the loglog-link function model including random
subject intercepts fit the data best. The results showed a significant main effect of FORCE in that
possibility conditions received lower SES (β̂=−0.15, χ2(1)=9.13, p<0.003) and education level
(β̂=−0.11, χ2(1)=4.84, p<0.03) ratings than necessity conditions. Furthermore, the 2-way inter-
action showed a significant effect in the SES model (SES-model: β̂=−0.26, χ2(1)=6.58, p=0.01;
education level-model: β̂=−0.31, χ2(1)=9.62, p<0.002), which results from MC receiving lower
SES (β̂=−0.18, χ2(1)=5.55, p<0.02) and education level (β̂=−0.26, χ2(1)=11.92, p<0.001) rat-
ings than SM only in possibility conditions.

3.3. PERSONAE. The output of the models using personae ratings as dependent variables are
shown in Table 4.

For the formality ratings, the cloglog-link function model including random subject inter-
cepts with FORCE, NUMBER, and their 2-way interaction as well as random item intercepts fit
the data best. The results showed a significant main effect of FORCE in that possibility condi-
tions received lower formality ratings than necessity conditions (β̂=−0.18, χ2(1)=4.19, p=0.04).
Furthermore, the 2-way interaction showed a significant effect (β̂=−0.50, χ2(1)=9.12, p<0.003),
which results from MC receiving higher formality ratings than SM only in necessity conditions
(β̂=0.45, χ2(1)=8.80, p=0.003).

For the politeness ratings, the cloglog-link function model including random subject intercepts
fit the data best. The results showed no significant effects.

For the confidence ratings, the logit-link function model including random subject and item
intercepts fit the data best. The results showed a significant main effect of FORCE in that possibil-

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 230

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


Fixed effects Model comparison Fixed effects Model comparison
β̂ SE χ2(1) p-value β̂ SE χ2(1) p-value

Formality
FORCE −0.18 0.09 4.19 0.04⋆

NUMBER 0.21 0.12 3.17 0.08
2-int −0.50 0.16 9.12 <0.003⋆

Politeness Confidence
FORCE 0.08 0.06 2.06 0.15 −1.79 0.09 453.50 <0.001⋆

NUMBER 0.03 0.06 0.22 0.64 0.05 0.08 0.45 0.50
2-int −0.21 0.11 3.56 0.06 −1.03 0.16 41.28 <0.001⋆

Friendliness Warmth
FORCE 0.14 0.05 7.16 <0.008⋆ 0.16 0.05 9.47 0.002⋆

NUMBER −0.20 0.05 14.01 <0.001⋆ −0.26 0.05 23.74 <0.001⋆
2-int 0.08 0.10 0.52 0.47 0.13 0.10 1.62 0.20

Coolness Rebelliousness
FORCE 0.02 0.05 0.15 0.70 0.02 0.05 0.10 0.76

NUMBER −0.42 0.05 62.42 <0.001⋆ −0.05 0.07 0.45 0.50
2-int 0.20 0.10 3.62 0.06 0.13 0.10 1.91 0.17

Table 4: Output of the model using the social background ratings. 2-int abbreviates the two way
interaction FORCE×NUMBER. SE stands for the standard error.

ity conditions received lower ratings than necessity conditions (β̂=−1.79, χ2(1)=453.50, p<0.001).
Furthermore, the 2-way interaction turned out significant (β̂=−1.03, χ2(1)=41.28, p<0.001), which
results from a cross-over effect: MC received significantly lower confidence ratings in possibility
conditions than SM (β̂=−0.66, χ2(1)=20.25, p<0.001), while MC received higher ratings than SM
in necessity conditions (β̂=0.52, χ2(1)=16.76, p<0.001).

For the friendliness and warmth ratings, the loglog-link model with random subject inter-
cepts fit the data best. The results showed a significant main effect of FORCE in that possibil-
ity conditions received higher friendliness (β̂=0.14, χ2(1)=7.16, p<0.008) and warmth (β̂=0.16,
χ2(1)=9.47, p=0.002) ratings than necessity conditions. There was a significant main effect of
NUMBER in that MC received lower friendliness (β̂=−0.20, χ2(1)=14.01, p<0.001) and warmth
(β̂=−0.26, χ2(1)=23.74, p<0.001) ratings than SM conditions.

For the coolness ratings, the loglog-link model with random subject intercepts fit the data best.
The results showed a significant effect of NUMBER in that MC received lower coolness ratings
than SM (β̂=−0.42, χ2(1)=62.42, p<0.001).

For the rebelliousness ratings, the probit-link model with random subject intercepts and slopes
for NUMBER fit the data best. The results showed no significant effects.

4. Discussion and Conclusion. The results of the study are summarized in Table 5.
First, MC was rated as less grammatical (with the mean ratings 5.06/4.90 for possibility/necessity

MC) than SM. We take this to be in line with the more restricted distribution of MC.
In relation to (RQ1), our study shows that MC conveys a weakened speaker commitment

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 231

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


FORCE NUMBER 2-int FORCE×NUMBER
Interpretation poss < ness MC > SM poss: MC < SM ness: MC > SM

Grammaticality – MC < SM – –
Socioeconomic status poss < ness – poss: MC < SM –

Education level poss < ness – poss: MC < SM –
Formality poss < ness – – ness: MC > SM
Politeness – – – –

Confidence poss < ness – poss: MC < SM ness: MC > SM
Friendliness poss > ness MC < SM – –

Warmth poss > ness MC < SM – –
Coolness – MC < SM – –

Rebelliousness – – – –

Table 5: Summary of the results. The abbreviation ‘poss’ stands for possibility and ‘ness’ for
necessity conditions. MC stands for modal concord and SM for single modal. 2-int abbreviates
the two-way interaction FORCE×NUMBER. The symbol < indicates that the left entity is smaller
than the right one; > represents the reverse. The symbol – indicates the lack of a significant effect.

than SM in possibility modals and a strengthened speaker commitment in necessity modals. This
finding challenges the semantic equivalence assumption for MC and SM as well as the claim
that modal adverbs in MC constructions have no semantic contributions. One way out would
be to analyze weakening or strengthening as a pragmatic effect of, say, “reinforcing” (Halliday
1970, see the quote above from Lyons 1977) of the respective modal force (i.e., existential or
universal): in simple words, doubling does not add extra contributions to the semantics but give
rise to enriched meanings about speaker assumptions. Alternatively, such effects can be taken as
arguments for the ‘modal spread’ analysis of Giannakidou & Mari (2018), by which the weakening
and strengthening effect can be analyzed as semantic contributions of modal adverbs. While we
do not provide any formal analysis for this in the paper, we see several arguments in favor of
the ‘modal spread’ analysis. Firstly, if we take modal adverbs such as certainly in must certainly
to have no semantics in MC constructions, we need to stipulate a different lexical entry for the
same adverb in other contexts, for example, where it co-occurs with a modal verb of a different
force. Our preliminary corpus study shows that (4-a) is a naturally-occurring combination, which is
“modally non-harmonic” (see the quote above from Lyons 1977)—in contrast, the well-formedness
of (4-b) is questionable, a topic we leave for future work. Given the well-formedness of (4-a) as
well as cases where modal adverbs occur without verbs, it strikes us as unattractive to analyze
modal adverbs as semantically vacuous for MC constructions.

(4) a. It may certainly be the case.
b. ?/*It must possibly the base.

In relation to (RQ2), (i) MC was rated as less friendly, warm, and cool than SM. This is something
which we neither predicted nor have a good explanation for, and thus needs to be further inves-
tigated in future studies. Crucially, certain measures showed a significant NUMBER*FORCE
interaction in that (ii) in possibility modals, MC was rated as significantly lower than SM in SES,

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 232

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


education and confidence levels and (iii) in necessity modals, MC was rated as more formal and
more confident than SM. Relating these results with the results of the interpretation, it seems that
weaker statements give rise to more negative perceptions and stronger ones to more positive per-
ceptions, indicating a positive correlation between meaning strength and social perception.

In relation to (RQ3), we only compare MC with NC here, see a brief grammatical comparison
of these two doubling phenomena in Zeijstra (2007). Sociolinguistics and recent research on so-
cial meaning have both documented work on NC, which is more prevalent in different varieties of
English and socially stigmatized (Eckert 1989, 2019, Moore 2021, Rotter & Liu 2025). In compar-
ison, there is little work on the social meaning of MC, which unlike NC, has a wide distribution in
English(es) and is not socially stigmatized. In recent work, Rotter & Liu (2025, 2024) conducted a
series of studies on the register-sensitivity and the social meaning of NC compared to negative po-
larity items (NPIs) in American English, see (5). The findings (in the data by mostly self-reported
non-dialect speakers of US English) are that (i) NC use is register-sensitive in that it was rated
less appropriate in formal than in informal contexts, (ii) comprehenders interpreted NC similarly
to NPIs, but (iii) associated NC use with lower socioeconomic and education levels, as well as less
formal but more rebellious personae of the speaker. The current study, in combination with our
previous work, provides experimental evidence that the doubling phenomena of MC and NC have
very distinct social meanings.

(5) a. I didn’t see nothing. (NC construction)
b. I didn’t see anything. (NPI construction)

Before we finish the paper, we would like to briefly address the limitations and future outlooks
of our study. First, our study compared MC with SM constructions with a single modal auxiliary.
The design can be extended to the comparison between MC, SM-verb and SM-adverb (e.g., must
certainly vs. must vs. certainly and may possibly vs. may vs. possibly). Secondly, we used isolated
sentences in the study, which was our first step towards understanding the social meaning of MC,
but it is evident that the set of inferences to be drawn from its use can differ greatly depending on
the specific interaction. The next step would therefore be to incorporate context and further explore
the question of whether MC is register-sensitive, and what social meanings arise in relation to its
use in different situational contexts.

Data repository. The data set for this study can be found in the online repository: https:
//osf.io/47rkg/.

Authorship contribution statement. Mingya Liu: Conceptualization, Formal analysis, Method-
ology, Writing - original draft, Supervision, Project administration, Funding acquisition. Stephanie
Rotter: Methodology, Data collection, Data analysis and visualization, Writing - original draft.

References

Bates, Douglas, Reinhold Kliegl, Shravan Vasishth & Harald Baayen. 2018. Parsimonious mixed
models. arXiv 1506. https://doi.org/10.48550/arXiv.1506.04967.

Christensen, Rune Haubo Bojesen. 2019. Ordinal: Regression models for ordinal data CRAN.
R-project.org/package=ordinal. R package version 2023.12-4.

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 233

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


Eckert, Penelope. 1989. Jocks and burnouts: Social categories and identity in the high school.
New York: Teachers College Press.

Eckert, Penelope. 2019. The limits of meaning: Social indexicality, variation, and the cline of
interiority. Language 95(4). 751–776. https://doi.org/10.1353/lan.0.0239.

Geurts, Bart & Janneke Huitink. 2006. Modal concord. In Paul Dekker & Hedde Zeijlstra (eds.),
Proceedings of the ESSLLI 2006 workshop concord phenomena at the syntax-semantics inter-
face, 15–20. Malaga: University of Malaga.

Giannakidou, Anastasia & Alda Mari. 2018. The semantic roots of positive polarity: Epistemic
modal verbs and adverbs in English, Greek and Italian. Linguistics and Philosophy 41(6).
623–664. https://doi.org/10.1007/s10988-018-9235-1.

Glass, Lelia. 2015. Strong necessity modals: Four socio-pragmatic corpus studies. In Paul Dekker
& Hedde Zeijlstra (eds.), University of Pennsylvania working papers in linguistics, vol. 21(2),
79–88.

Hall-Lew, Lauren, Emma Moore & Robert J. Podesva. 2021. Social meaning and linguistic vari-
ation: Theoretical foundations. In Lauren Hall-Lew, Emma Moore & Robert J. Podesva
(eds.), Social meaning and linguistic variation, 1–24. Cambridge: Cambridge University
Press. https://doi.org/10.1017/9781108578684.001.

Halliday, Michael A. K. 1970. Functional diversity in language as seen from a consideration of
mood and modality in English. Foundations of Language 6. 322–361.

Howcroft, David M. & Verena Rieser. 2021. What happens if you treat ordinal ratings as in-
terval data? Human evaluations in NLP are even more under-powered than you think. In
Marie-Francine Moens, Xuanjing Huang, Lucia Specia & Scott Wen-tau Yih (eds.), Proceed-
ings of the 2021 conference on empirical methods in natural language processing, 8932–
8939. Online and Punta Cana, Dominican Republic: Association for Computational Lin-
guistics. https://doi.org/10.18653/v1/2021.emnlp-main.703. https://aclanthology.
org/2021.emnlp-main.703. Accessed 2022-01-03.

Kortmann, Bernd & Edgar W. Schneider. 2004. A multimedia reference tool. volume 1:
Phonology. volume 2: Morphology and syntax. Berlin, Boston: De Gruyter Mouton.
https://doi.org/10.1515/9783110197181.

Labov, William. 2006. The Social Stratification of English in New York City. Cambridge: Cam-
bridge University Press 2nd edn. https://doi.org/10.1017/CBO9780511618208.

Liddell, Torrin M. & John K. Kruschke. 2018. Analyzing ordinal data with metric models:
What could possibly go wrong? Journal of Experimental Social Psychology 79. 328–348.
https://doi.org/10.1016/j.jesp.2018.08.009.

Liu, Mingya, Stephanie Rotter & Anasatasia Giannakidou. 2021. Bias and modality in condition-
als: Experimental evidence and theoretical implications. Journal of Psycholinguistic Research
50. 1369–1399. https://doi.org/10.1007/s10936-021-09813-z.

Lyons, John. 1977. Semantics, vol. 2. Cambridge: Cambridge University Press.
https://doi.org/10.1017/CBO9780511620614.

Lüdeling, Anke, Artemis Alexiadou, Aria Adli, Karin Donhauser, Malte Dreyer, Markus Egg,
Anna Helene Feulner, Natalia Gagarina, Wolfgang Hock, Stefanie Jannedy, Frank Kam-
merzell, Pia Knoeferle, Thomas Krause, Manfred Krifka, Silvia Kutscher, Beate Lütke,

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 234

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/


Thomas McFadden, Roland Meyer, Christine Mooshammer, Stefan Müller, Katja Maquate,
Muriel Norde, Uli Sauerland, Stephanie Solt, Luka Szucsich, Elisabeth Verhoeven, Richard
Waltereit, Anne Wolfsgruber & Lars Erik Zeige. 2022. Register: Language Users’
Knowledge of Situational- Functional Variation– Frame Text of the First Phase Proposal
for the CRC 1412. Register Aspects of Language in Situation (REALIS) 1(1). 1–57.
https://doi.org/10.18452/24901.

Moore, Emma. 2021. The social meaning of syntax. In Lauren Hall-Lew, Emma
Moore & Robert J.Editors Podesva (eds.), Social meaning and linguistic varia-
tion: Theorizing the third wave, 54–79. Cambridge: Cambridge University Press.
https://doi.org/10.1017/9781108578684.003.

Pescuma, Valentina N., Diana Serova, Julia Lukassek, Antje Sauermann, Roland Schäfer, Aria
Adli, Felix Bildhauer, Markus Egg, Kristina Hülk, Aine Ito, Stefanie Jannedy, Valia Kordoni,
Milena Kühnast, Silvia Kutscher, Robert Lange, Nico Lehmann, Mingya Liu, Beate Lütke,
Katja Maquate, Christine Mooshammer, Vahid Mortezapour, Stefan Müller, Muriel Norde,
Elizabeth Pankratz, Angela G. Patarroyo, Ana-Maria Plesca, Camilo Rodrı́guez-Ronderos,
Stephanie Rotter, Uli Sauerland, Gohar Schnelle, Britta Schulte, Gediminas Schüppenhauer,
Bianca Maria Sell, Stephanie Solt, Megumi Terada, Dimitra Tsiapou, Elisabeth Verho-
even, Melanie Weirich, Heike Wiese, Kathy Zaruba, Lars Erik Zeige, Anke Lüdeling &
Pia Knoeferle. 2023. Situating language register across the ages, languages, modalities,
and cultural aspects: Evidence from complementary methods. Frontiers in Psychology 13.
https://doi.org/10.3389/fpsyg.2022.964658.

R Core Team. 2024. R: A language and environment for statistical computing. R Foundation for
Statistical Computing Vienna, Austria. https://www.R-project.org/.

Rotter, Stephanie & Mingya Liu. 2024. A register approach to negative concord vs. negative
polarity items in English. Linguistics https://doi.org/10.1515/ling-2023-0016.

Rotter, Stephanie & Mingya Liu. 2025. Less formal and more rebellious — an experiment on the
social meaning of negative concord in American English. In Chiara Gianollo & Johan van der
Auwera (eds.), Negative concord: A hundred years on, vol. 385, 303–332. Berlin, Bosten: De
Gruyter Mouton.

RStudio Team. 2024. Rstudio: Integrated development environment for r. RStudio, PBC Boston,
MA. http://www.rstudio.com/.

Yalcin, Seth. 2007. Epistemic modals. Mind 116(464). 983–1026.
https://doi.org/10.1093/mind/fzm983.

Zeijstra, Hedde. 2007. Modal concord. Proceedings of Semantics and Linguistic Theory (17).
317–332. https://doi.org/10.3765/salt.v17i0.2961.

Proceedings of ELM 3: 224-235, 2025

Mingya Liu and Stephanie Rotter:
Linguistic and Social Meaning Match: An experiment on modal concord in English. 235

https://journals.linguisticsociety.org/proceedings/index.php/ELM/issue/archive
https://www.elm-conference.net/