regarded as one of the most important sources of evidence in linguistics, essentially forming its

empirical base (Wexler et al., 1975; Carr, 1990; Schütze, 1996/2016; Baggio et al., 2012). Both

the techniques used to elicit such judgments (e.g., controlled experiments, self-introspection, or

targeted questioning about whether a specific sentence sounds fine in a specific language) as

well as the type of sample that is necessary for the results to have ecological validity (e.g., a

pool of participants that is randomly selected from the targeted linguistic community, a non-

random sample, or self-introspection) are a matter of debate (see Phillips, 2009; Gibson and

Fedorenko, 2010; Sprouse and Almeida, 2013; Branigan and Pickering, 2016). On the other hand,

no controversy exists over the fact that judgments about what forms part of a person’s linguistic

repertoire constitute a rich source of information in theoretical and experimental linguistics.

If one searches PubMed or any other database for the terms

“acceptability judgment tasks” and “language” on the one hand,

and “grammaticality judgment tasks” and “language” on the

other, one will quickly discover that the relevant experiments that

will show up are the same. They all report findings that are based

on the exact same elicitation technique. Perhaps the greatest

illustration of how the terms “acceptability” and “grammaticality”

are used, often without a clear distinction in place, comes from

Schütze’s (1996/2016) seminal book on linguistic judgments.

While the title of the book talks about “grammaticality judgments

judgments of the grammaticality/acceptability of sentences (and

other linguistic intuitions) have been the major source of

evidence in constructing grammars” (p. xi, emphasis added).

Just as linguists and other cognitive scientists have at times

used the terms “ungrammatical” and “unacceptable” roughly

synonymously, plurality and overlapping may characterize the

use of symbols like ?, *, or ??, that are employed to signal some

deviant property of the linguistic stimulus (Bard et al., 1996). To

define the relevant terms, the grammaticality of a sentence refers

to whether the sentence conforms to the syntactic rules of a given

with visible primitives; we cannot zoom in on a linguistic

stimulus until we find and tease apart an independent, self-

contained grammatical core. This means that grammaticality,

as one of the possible elements that determine acceptability,

“is not directly accessible to observation or measurement” (Lau

et al., 2016: 3). The question thus becomes: How do we know

anything about grammaticality aside of the information provided

by acceptability? Put differently, if grammaticality is defined as

“conforming to the rules of the grammar of language X” and if the

grammar of language X has the shape that its speakers’ judgments

grammaticality other than the one that goes through speakers’

perception of well-formedness (i.e., acceptability)?

and their scales might not coincide, hence his reference to

“unacceptable grammatical sentences”: sentences that do not

form part of grammar for reasons that have nothing to do

with grammar. The second aim of the present work is to chart

the variation space that is created when one disentangles the

two notions: unacceptable grammatical sentences, acceptable

ungrammatical sentences, their respective parsability, and

the process of assigning them meaning. Last, the scales of

grammaticality and acceptability will be discussed and it will

be shown that they do not coincide: there are n ways of

Humans are surprisingly good at providing accurate and

consistent judgments about what forms part of their linguistic

repertoire.

Although informants’ opinions about their linguistic

English about the acceptability of (1) or the unacceptability of (2).

The former is a grammatically well-formed sentence of English,

while the latter is a word-salad that would probably be read and

parsed in a rhythm that pertains more to lists of objects than to

connected speech.

Yet, even though such judgments are largely coherent with

the actual shape of speakers’ internalized grammar, there are

some stimuli that have the ability to trick the cognitive parser

an acceptability judgment that faithfully represents whether the

stimuli fall inside or outside the domain of predictions of the

underlying, internalized grammar. In other words, precisely

because of the high complexity of some stimuli, and due to the

fact that the cognitive parser works on the basis of processing

heuristics (Kahneman, 2011), some deviations may go unnoticed.

One such example is (4), which looks very similar to (3) but–

unlike (3)–violates a rule of grammar.

to novel information (“Novel Information Bias”; Leivada, 2017).

Acceptability is thus affected by a variety of processing factors

and cognitive biases, and so is grammaticality. For example,

although data that flout Identity Avoidance exist [(6); see

Leivada, 2017 for examples of syntactic violations], there are no

grammatically licit structures that feature five identical, adjacent

complementizers, and the prediction is that such structures will

never be in use, because a grammar would never consistently

deploy them. Even if grammars were able to generate a sentence

like *“John said that that that that that Mary kissed him,”

licensed arguments. In other words, although the rules of the

grammar of a language are subject to change in a way that

may legitimize the use/acceptability of a previously ill-formed

sentence and/or diminish the use/acceptability of a previously

attested one, certain changes are not expected to occur, because

they violate either a core principle of linguistic cognition or a

general cognitive bias.

Talking about a dissociation of acceptability and

grammaticality, unacceptable grammatical sentences are

one logical possibility. One may wonder whether the other

called illusions because they trick the parser in a way that renders

high acceptability ratings in experiments, even though the stimuli

are ill-formed (Wellwood et al., 2018). In linguistic terms, (8) is

ill-formed because the main clause subject calls for a comparison

of cardinalities of sets, but in the absence of a bare plural in the

embedded clause subject, no comparison set is made available

(Phillips et al., 2011; O’Connor et al., 2012; Wellwood et al., 2018).

Linguistic illusions are the outcome of a partial-match strategy

that is operative during processing (Reder and Kusbit, 1991;

Kamas et al., 1996; Park and Reder, 2004).

words, a processing threshold is set and the stimulus is checked

up to this threshold, hence the notion of partial matching. Given

that (8) makes use of locally coherent templates (Townsend

and Bever, 2001) that provide a “good-enough fit” (Ferreira and

Patson, 2007) for the parser, its ill-formedness may go unnoticed,

and this results in high acceptability. Evidently, the way the

parser works–via the use of processing heuristics–mediates

one’s access to the internalized knowledge of grammar. Yet,

the ease with which a sentence is unambiguously parsed is

not a guarantee for either grammaticality or acceptability.

assigning a coherent meaning to the whole” (Adger, 2018:

161). Unlike (2) or even (10), (9) can be easily parsed in a

way that pertains to connected speech. Moreover, a coherent

interpretation of it can be provided, and over the years there have

been various proposals that construe meanings for it.

Perhaps

green ideas refer to environmental considerations. One could

build a metaphorical narrative where these ideas are colorless and

sleeping because at present there is not enough effort to combat

climate change, however, their sleep is furious, something that

may suggest that some promising initiatives for change are under

attested in naturalistic speech (12a), and the relevant data also

2018 on negation). Talking about ungrammatical sentences

that are acceptable and parsable, Otero (1972) reached the

conclusion that wide acceptability is not a guarantee for

grammaticality. Even sentences that have been described as

blatantly ungrammatical may actually be acceptable to some

degree, and this degree varies across speakers of the same

language that have different developmental trajectories (e.g., late

bilinguals, heritage speakers, L1 attriters). For example, (5) is

ungrammatical because extraction out of coordinated structures

is prohibited. A similar island effect has been described for

acceptability for five different types of island violations–

including relative clause islands that received a mean score of

3.6 in an 1–5 scale, where 1 stood for the sentence sounding

perfectly natural–among late bilingual and heritage speakers of

Spanish. Importantly, the two groups differed both in terms of

their judgments and in terms of their involuntary physiological

reactions that can be proxies for processing effort. In Lowry

et al. (2019) these were measured through a pupillometry

study: pupil dilation in the ungrammatical stimuli was observed

only in the group of late bilinguals, while there was no effect

ungrammatical sentences. To illustrate this, let’s consider the

comparative illusions in Tables 1, 2 [examples (8) and (11),

respectively], which are ungrammatical but trick the parser into

acceptability (Wellwood et al., 2018; Leivada et al., 2019b).

Although various experiments have shown that these sentences

are assigned a high acceptability rating, one could say that this

does not entail that these sentences are actually parsed, in the

sense that speakers actually assign them a meaning m. A clear

exposition of this point is given by Tim Hunter as a reply to

Hornstein (2013), who suggests that such sentences may sound

than just me have been to Russia, (ii) people have been to Russia

more times than I have, and (iii) many people have been to Russia

more times than I have (see also Wellwood et al., 2018). Naturally,

this is not what the sentence says, but nevertheless, a meaning

is assigned to the sentence. Also we suggest that one should not

ignore the possibility that those speakers that seem stumped upon

being asked to provide an interpretation do not do so because

they never actually established an association (s, m), but because

in their attempt to articulate the latter, they spot the illusion.

Crucially, this does not entail that at no point were they actually

The second interesting issue with Hunter’s point has to do

with the juxtaposition of two very different ways of eliciting

judgments through asking “Is s acceptable?” or “Is there

a meaning m such that (s, m) is acceptable?” These two

questions do not tap into the same thing. Previous research on

the pragmatics of cognitive illusions has proposed that when

processing such sentences, the hearer searches for meaning

within a manipulative communication, that is, within a tricky

context that features a “manipulation (that) can be best defined

(Maillat and Oswald, 2009: 361). In this context, the hearer stops

searching for meaning after finding one that sufficiently meets

her expectations of relevance in accordance with the previous

discourse. The illusion thus arises in the process of selecting

meaning within a manipulative context that takes advantage of

(i) the parser’s limitations and (ii) the parser’s way of operating

through employing certain processing heuristics such as partial

matching or shallow processing.

If relevance and previous context can bias an acceptability

judgment through creating the necessary conditions for an

illusion to arise, the bias will be even greater if a specific m is

given to a participant point-blank in an experiment that asks “Is

there a meaning m such that (s, m) is acceptable?” As shown

in Tversky and Kahneman’s (1974) work, options in a task are

the reference point. This does not mean that such a task cannot

provide useful and informative findings, but that possibly the

obtained findings will not be tapping directly into a speaker’s

perception of her idiolect. Instead, it will be mediated by an

anchoring effect that may cause an adjustment to the speaker’s

judgment on the basis of m. To understand this effect, consider

the following example by Kahneman (2011).

(14a) Was Gandhi more or less than 144 years old when he

died?

(14b) How old was Gandhi when he died?

Kahneman (2011: 122).

Of course nobody claimed that Gandhi was 144 years old when

he died, but it has been found that when (14b) is presented after

on the basis of an intended meaning m?

(15b) Assuming a scale from 1 to 5, how do you rate s

on the basis of your idiolect?

In (15a), the possibility of s getting a meaning is explicit and a

question that introduces the stimuli s. This can bias the rating of

s on the basis of the “anchor-and-adjust” heuristic.

To sum up, illusions do not necessarily entail that parsing fails

to produce a meaning, but that the parser can be tricked into

providing both a meaning and an acceptability rating that may

not correspond to the actual status of the stimulus in terms of

what the speaker’s internalized grammar looks like. Importantly,

a number of factors contribute to this process of tricking the

parser: context, task and stimuli presentation, as well as structural

complexity are only a few.

whether there is a way of capturing grammaticality other than the

one that goes through speakers’ perception of what is well-formed

in their native linguistic repertoire (i.e., acceptability). Having

presented the dissociation between acceptability, grammaticality

and the way the parser works, the next section deals with how

grammaticality is established and where it comes from.

Asking about the origin of the rules of grammar, Adger (2019)

suggests that we learn them: They come from the way people

speak. Although this is true, the issue is more complex, because

different people speak in different ways even within linguistic

communities that feature only one language. When one says

established consensus about what is the norm in a specific variety

of English; a norm that the grammar books describe in detail.

Put differently, if some speakers of English, Spanish, or German

accept to some degree or even produce to some degree island

violations (Lowry et al., 2019 for Spanish), missing verbs in nested

hierarchies (Häussler and Bader, 2015 for English and German),

or comparative illusions

are grammatical in English, Spanish, and German? While it

certainly appears to be the case that some speakers’ grammars

the way speakers judge sentences in an experiment (where even

examples 14–15), the way they actually speak, and the way that

prescriptive grammar says they (should) speak.

The question still holds: Where does grammaticality come

from? The tentative answer we offer is that grammaticality is

often a formal, standardized snapshot of the way the official

language looks like at a given point in time. Grammaticality is

constantly redefined through ever-changing acceptability, but it

also reflects stable properties of general cognition. In this context,

(un)grammatical today may not be (un)grammatical tomorrow,

depending on whether the new speakers of X find it acceptable or

not and whether this acceptability is generalized and established

as the norm or not. For example, Ancient Greek featured a

syntactic phenomenon called Attic syntax which permitted a

number mismatch between the plural, neuter subject and the verb

(16a). This structure is not a grammatically licit option in Modern

Greek (16b), but not because there is something intrinsically

ungrammatical about it; it simply does not form part of the

grammar anymore. Phrased differently, there is no notion of self-

“The children are playing.”

(16b) *Ta koritsia gela. [Modern Greek]

Intended meaning: “The girls are laughing.”

This claim is partially concordant with Chomsky et al.’s (2019);

see also Chomsky (1993) view that in natural language there exists

no independently given notion of grammatical well-formedness.

Indeed, the grammatical well-formedness of a linguistic stimulus

does not boil down to an independently definable grammatical

the stimulus forms part of the standardized snapshot or not

and (ii) is subject to change such as the one shown in (16a-b).

Nevertheless, this view is true only for one reading of the

term “grammatical”: grammatical as actually forming part of the

grammar of a specific language.

However, we suggest there is also another reading of the term

“grammatical.” To understand this other reading, one needs

to factor in that change is not without limits. Not all changes

are possible and not all linguistic stimuli are candidates for

forming part of grammar. For example, as mentioned in the

grammatical. Similarly, a sentence such as (4), which violates

the θ-criterion, is unlikely to ever form part of grammar.

discussed in the next section, certain changes are not expected

to occur, because they violate either a core principle of language

(e.g., the θ-criterion in 4) or a general cognitive bias (e.g., the

Novel Information Bias in 6). In this sense, Chomsky et al. (2019)

are right in arguing that there exists no independently given

notion of grammatical well-formedness, but we would like to add

to their claim that there do exist independently given constraints

to the set of entities that this notion can encompass. This is the

It is an uncontroversial claim that acceptability judgments are not

categorical, but form a continuous spectrum (Sprouse, 2007 and

references therein). The usual meaning of the word “continuous”

is unbroken or undivided, hence it is the nature of a continuum

to be undivided, or better, to permit repeated division without

limit (Bell, 2017). If one subscribes to the view that acceptability

should be viewed as a continuum, one also subscribes to the view

that the acceptability continuum is infinitely divisible. Although

acceptability judgment tasks that involve Likert scales feature

ungrammatical in the relative or in the absolute sense. The relative

sense pertains to the first reading of the term “grammatical”

that was mentioned above: forming part of grammar. We call

it “relative” because it is defined in the context of a given

language. For example, (16b) is ungrammatical in relation to

Modern Greek, but it is not ungrammatical per se. It is a potential

candidate for forming part of grammar, it was grammatical in

the past (16a), and it may be again in the future. Similarly, (17)

is ungrammatical in relation to Standard English, but this is an

accident, as it could potentially be grammatical (and in fact it is

and (iii) concerns violations of some core principle of language

and/or cognition, that is, structures that grammar would never

consistently deploy. Therefore, absolute ungrammaticality has to

do with structures that cannot form part of grammar.

Comparing the scales of the two notions, acceptability

and grammaticality, it is meaningful to talk about partial

acceptability (Sprouse, 2007), but not about partial or strong-

weak ungrammaticality. A rule of grammar (or more than one

rule of grammar) can be either violated or not, but it cannot

be violated just a bit. Ungrammaticality cannot be a matter of

ungrammaticality,” we would argue that this refers either to

partial unacceptability or to variation in a linguistic community.

Consider, for instance, the discussion of partial ungrammaticality

in Attinasi (1974): “A hidden assumption of homogeneity, that

the language competence of every speaker consists of the same

structures, falters when the question of partial ungrammaticality

is raised. How can some speakers totally reject, others partially

accept and still others totally accept certain sentences as

grammatical if each presumably speaks ‘English,’ or any other

language?” (p. 280). In our view, this question has to do with

grammaticality evokes an established norm that is part of a

formal snapshot. Speakers may deviate from this norm, either

because language change has occurred and the norm does not

reflect this yet, or because their idiolect simply differs from the

norm. But this should be referred to as “interspeaker variation,”

not “partial grammaticality.”

The present work has discussed the complex relation between

grammaticality, acceptability, and parsability. A number

of unacceptable grammatical sentences and acceptable

ungrammatical sentences have been presented, including

grammatical illusions, violations of Identity Avoidance, and

sentences that involve a high level of processing complexity

introduced: (Un)grammaticality in the relative sense refers to

the whether the stimulus falls within the domain of predictions

of a given grammar or not. (Un)grammaticality in the absolute

sense refers to whether the stimulus has the potential to

be a part of grammar or not. Relative (un)grammaticality is

an ever-changing property of the stimulus, whereas absolute

(un)grammaticality is stable. In both readings of the term,

grammaticality is defined by something that is external to the

stimulus (be it the grammar of a specific language or principles

of general/linguistic cognition), and it is not an inextricable

“acceptable” and “grammatical,” the overarching aim of this

work has been to aid the field in reaching a more adequate

level of terminological clarity for notions that pertain to the

evidential base of linguistics. Many details of the distinction

between relative and absolute (un)grammaticality are left to

be worked out, and this will likely be the topic of future

work. To give just one example, when we deal with island

effects of the sort discussed above, do we deal with absolute

ungrammaticality that is universal and derives from processing

or other principles of language/cognition, or with relative

which x: John likes x and oranges)? The answer is currently

unclear to us, and it probably needs novel experimental work

to be properly discussed. Recognizing this uncertainty does not

mean undermining the proposed distinction between absolute

and relative ungrammaticality. It rather suggests that progress is

underway, or as (Feynman, 1998: 27) puts it, “[b]ecause we have

the doubt, we then propose looking in new directions for new

ideas. The rate of the development of science is not the rate at

which you make observations alone but, much more important,

the rate at which you create new things to test.”

equally to the final editing of this work.

This work was supported by the European Union’s Horizon 2020

Research and Innovation Program under the Marie Skłodowska-

Curie Grant Agreement No. 746652. The publication charges for

this manuscript have been funded by a grant from the publication

fund of UiT The Arctic University of Norway. The funders had