Abstract

Despite the fact that acceptability/grammaticality judgements form the basic unit of data in generative syntactic theory, it is common that individuals disagree on what is or is not acceptable/grammatical. Though there are many workarounds to this problem, ranging from focusing on just a single speaker to the use of statistical measures to determine typical judgements, such variation in judgement has rarely been tackled directly. As it turns out, however, disagreements in judgements are not random in nature but are fundamentally principled; there are specific, measurable factors that influence our judgements in different ways. If we understand and control these factors, we can bring out underlying uniformity on seemingly contentious issues, allowing for syntactic hypotheses to be reliably tested and replicated at the level of the individual. In particular, various interpretative phenomena, such as bound variable anaphora and quantifier scope, can be made completely dependent on syntactic structure once the proper controls are applied. This allows the availability of such readings to be used as a reliable and definite probe for c-command relations, enabling new progress on previously intractable issues and uncovering a host of new areas for research as well. Such techniques represent a powerful and easily utilized research tool for syntacticians looking to expand their methodological inventories as generative syntax seeks new frontiers of scientific rigor in the 21^st century.”

1

Judgements Revisited:�Testing Syntactic Theories in the Face of Variation

​

Daniel Plesniak (plesniak@usc.edu)

Seoul National University Linguistics Colloquium

2022/09/23

2

Talk Goals

1. To introduce a new way of dealing with syntactic data that allows us to deal with the problem of judgement variation.

​

• To provide some examples of both why such a method is needed and how it can be applied.

​

• To sample just a few results from my own research that illustrate the successful application of the above.

3

Further note on goals

• One of the aims is for this talk to be useful to you in your own research.
• The focus is not only on what I have done but also what you might do!

​

• While I do not have time to review every detail, I hope to give you a sense of the “heart” of this program, so you can think about how it could be relevant to you.

​

• I am “around”, so if you’d like to talk more about how you can implement these techniques for yourself, feel free to let me know.

4

A quick plug

• Another avenue to learn more is through a book (co-edited and with chapters by me, Hajime Hoji, and others) coming out this November.

​

• The chapters are a bit “eclectic”, but the intent is that readers will come away from them with a detailed understanding of the relevant issues (both theoretical and practical) and the ways in which we address them.

​

• This talk will especially address topics found in Chapters 4-8 of that volume.

​

​

5

Talk Structure

• Three main parts (in order):

1. A discussion of the nature of judgements in syntactic theory

• The theoretical significance of judgement variation
• Some brief relevant examples in Korean and English
• The need to test predictions in a “controlled environment”.

​

2. A longer example involving bound variable anaphora (BVA) interpretations.

• The availability of BVA interpretations can be made to reflect syntactic structure…
• … but only if we control for non-structural factors that affect BVA availability.
• Some experiment evidence of this.

​

• 3. Some further discussion of the utility of this methodology
• New evidence for old debates (example: “possessor binding”)
• Outliers and testability (conclusory comments on data in generative linguistics)

​

6

Judgements in generative syntactic theory

• Following Chomsky (1995 and elsewhere), we postulate a Computational System that combines items into an abstract structure (a “syntax tree”), which serves as an input for both the external linear form of the sentence and the internal meaning.

​

• Syntactic theories predict, from hypothesized�properties of the abstract structure (and more)�which combinations of forms and meanings�will be generable by an individual.

​

7

Data and Predictions

• Data in syntactic theory thus consists of the actual judgements of human beings as to whether they can indeed “accept” a given sentence with a given interpretation.
• Sometimes we talk about “ungrammatical” sentences, but what we basically mean by that is that there are no possible acceptable interpretations for the given sentence.

​

• Syntactic hypotheses are thus evaluated based on whether or not they make the correct predictions.
• If a given set of hypotheses predicts that something ought to be impossible to accept (“*”), but people accept it, then that is a problem for that theory.

8

Judgement Variation

• Unfortunately, people are not always consistent in what they do or do not accept.

​

• We will see some examples shortly, but anyone who has worked on syntax has surely had the experience of not sharing judgements reported in a paper and/or disagreeing with a co-native speaker about what is acceptable in one’s own native language.

​

• Cases of such disagreements are manifested in the literature since at least the early 60’s (see Hayashishita 2004 on the Chomsky-Katz & Postal debate about inverse scope in English).
• But it has rarely been totally clear what to do about them.

9

Common Workarounds

• Rather than confront the issue of variation head on, it has been more common to try to find ways to simply choose one (provisionally) “correct” set of judgements. Methods include:
• Focusing on the judgements of just one speaker (at a specific time)
• Checking the judgements on several speakers and deciding what the rough consensus judgements are on the issue in question.
• Using averages and other statistical tools on data from many speakers to find out what can be mathematically said to be the most typical judgements/contrasts in judgements.

10

Common Workarounds

• While such approaches have plenty of uses, they are all suppressing a fundamental issue, namely that some speakers have genuine disagreements in judgement, even if one judgement is more common than the other.

​

• In this sense, it is not strictly accurate to speak of a single “correct” judgement, even if that is sometimes a useful simplifying assumption.

11

A more direct approach

• The alternative approach is to abandon the assumption that a given sentence (paired with a given interpretation) has a set acceptability.

​

• Rather acceptability is treated a function of multiple interacting factors, some of which are consistent across individuals and some of which vary.

​

• The base sentence structure is one of these factors, but its effects will only be reliably seen when other factors are not interfering.

12

Relativized Predictions

• To the extent that structure is consistent across individuals, if acceptability judgements rely on that factor without interference from others, there will indeed be predictable and shared judgements across individuals.
• To get to those predictable judgements though, we need to achieve control of the interfering factors.
• Essentially, like scientists in a lab, we need to achieve a “controlled environment” in which to gather our data.
• Predictions will thus be relativized to only apply once certain controlled conditions are met.

13

Bound Variable Anaphora

• Before discussing achieving a “controlled environment”, let me briefly illustrate the problem of variation with the case of Bound Variable Anaphora (BVA) in Korean and English.
• BVA, the so-called “quantificationally bound reading”, occurs in sentences like the following:

​

(1) 모든 사람이 자신의 개를 사랑해.� “Every person loves his dog.”

​

• The BVA reading is the one in which, like the diagram above, the individual denoted by the bindee (자신 ‘his’ in 자신의 개 ‘his dog’) ranges across the individuals “denoted” by the binder. (모든 사람 ‘every person’)

14

A Korean Example

• Which of the following sentences are acceptable to you when paired with the BVA interpretation schematized below?

​

(2) USC와 UCLA가 거기의 학생들에게 칭찬 받았다.� USC and UCLA were praised by their students.

(3) 거기의 학생들에게 USC와 UCLA가 칭찬 받았다.� By their students, USC and UCLA were praised.

(4) 거기의 학생들이 USC와 UCLA를 칭찬했다.� Their students praised USC and UCLA.

(5) USC와 UCLA를 거기의 학생들이 칭찬했다.� USC and UCLA, their students praised.

​

15

Praise

USC’s Students USC

Praise

UCLA’s Students UCLA

A Korean Example (backup)

• My understanding is that some people really do not like using 거기 ‘there’ in this way, so those individuals can use these versions as an alternative, swapping 자신 ‘itself’/‘its own’ for 거기.
• My experience is that 자신 is more consistent in judgements than 거기, so it is less useful for demonstrating variation, but judgements do still differ.

​

(2’) USC와 UCLA가 자신의 학생들에게 칭찬 받았다.� USC and UCLA were praised by their students.

(3’) 자신의 학생들에게 USC와 UCLA가 칭찬 받았다.� By their students, USC and UCLA were praised.

(4’) 자신의 학생들이 USC와 UCLA를 칭찬했다.� Their students praised USC and UCLA.

(5’) USC와 UCLA를 자신의 학생들이 칭찬했다.� USC and UCLA, their students praised.

​

16

Praise

USC’s Students USC

Praise

UCLA’s Students UCLA

Korean Example

• Taking as an example one particular Korean speaker, Yoona Yee, she does accept BVA sentences like (4), but her judgements are affected by a number of factors:
• The choice of binder:�(6) 거기의 학생들이 모든 대학교를 칭찬했다.� Its students praised every university.
• The choice of bindee:�(7) 자기의 학생들이 USC와 UCLA를 칭찬했다.� Their own students praised USC and UCLA.
• Other aspects of the sentence:�(8) 만약에 자기의 학생들이 USC와 UCLA를 칭찬했다면…� If their own students praised every university…
• Her judgements also change over time (and judgements on the same sentences will vary across speakers).

​

​

17

Korean Example

• Taking as an example one particular Korean speaker, Yoona Yee, she does accept BVA sentences like (4), but her judgements are affected by a number of factors:
• The choice of binder:�(6) 거기의 학생들이 모든 대학교를 칭찬했다.� Its students praised every university. Yoona: BVA impossible
• The choice of bindee:�(7) 자기의 학생들이 USC와 UCLA를 칭찬했다.� Their own students praised USC and UCLA. Yoona: BVA impossible
• Other aspects of the sentence:�(8) 만약에 자기의 학생들이 USC와 UCLA를 칭찬했다면…� If their own students praised every university… Yoona: BVA possible
• Her judgements also change over time (and judgements on the same sentences will vary across speakers).

​

​

18

English

• In the roughly 100 person experiment (Plesniak Under Review) we are going to discuss shortly, roughly 2/3 reject and 1/3 accept a BVA reading with:

​

(9) His student spoke to every teacher.

​

• In case we are tempted to dismiss the minority as somehow misunderstanding the question, I myself belong to that minority.
• I have spoken to multiple people with both kinds of judgements; they generally express shock and/or disbelief that the other kind of judgement is possible.

19

C-command

• These minority judgements are problematic for the classic account of BVA (stemming from Reinhart 1983), which holds that it requires the binder to c-command the bindee.
• Thus, while it may be acceptable with a binder as the subject and the bindee within the object (even under scrambling), we predict it should never occur with the binders the object and the bindee in the subject, as it does for some people in (9).

​

• The use of interpretations like BVA as diagnostics for c-command has been hugely influential in shaping syntactic theory since it was formulated, but precisely because there are so many apparent counterexamples, it has come under increasingly sharp attack.
• One particularly relevant example, Barker (2012), claims that BVA is simply not dependent on syntactic structure at all.

20

Multiple Sources

• While it seems true that BVA does not always require c-command, this does not mean that the two are unrelated.

​

• Rather, we should understand that BVA readings are surface level phenomena (we can “sense” them).
• There may be multiple distinct mechanisms/sources (unconscious and thus not directly sense-able) that lead to such interpretations being generated.

​

• C-command is required for one such source, but not necessarily others.
• This was in fact contended at great length in Ueyama 1998, to which we shall return.

21

A Phonology Analogy

• Consider the case of the �(surface phenomenon)�[t] in Korean.

​

• When [t] appears in a coda, �there are quite many different�underlying sources that might�have generated it.
• There is thus no simple equation that�can be made to the effect that [t] always�represents /t/ (or any other phoneme).

22

The use of tests

• Continuing with the phonology analogy, imagine we are investigating the underlying forms of words. Such neutralization poses a challenge.

​

• Fortunately, there are solutions to that challenge. For example, let us imagine we have a noun [kot], which means either ‘place’ or ‘cape’.
• We can apply a particular test, for example adding the nominative marker, which will yield [koʃi ] in the first case and [kotɕi] in the second.
• Thus, we can disambiguate, and, via empirical tests, determine that we are dealing with /kos/ and /kotɕ/ respectively.

​

• So what “phonemes” lead to [BVA], and how can we test for them?

23

Sources of BVA

• Following the model initiated by Ueyama (1998) and further expanded by Hoji (2022a), there are (roughly) four sources of BVA:

​

• Formal Dependency (FD), which requires, among other things, X to c-command Y in the syntactic structure.
• Indexical Dependency (ID), which requires, among other things, X to precede Y in the linear form of the sentence.
• "Quirky Binding”, which Hoji splits into:

3a. “Binder Quirky”, with certain idiosyncratic requirements on the binder.

3b. “Bindee Quirky”, with certain idiosyncratic requirements on the bindee.

24

Control of Sources

• While all four sources of BVA can induce a degree of variation, our main focus today will be on controlling the variation induced by Binder and Bindee Quirky.
• Their requirements are complicated, but involve things like whether the binder can be construed as “prominent enough” to be the “topic” of the sentence and whether the bindee can be construed as “non-individual denoting”.
• What is crucial to recognize is that these depend on:
• a. what the binder/bindee in question is.
• b. the individual in question’s current “cognitive state”.

25

Binder and Bindee

• There does not seem to be any simple way to know in advance which binders and bindees can or cannot be “quirky” (see discussion in Ueyama 1998, Hayashishita 2004, and Hoji 2022a).
• Even the same person at different times can have differences in which lexical items are “quirky”; in Hoji 2022a (Chapter 5 of the forthcoming book), the author tracks his judgements on the matter across three different “stages”.

​

• Thus, our only hope is, for a given person at a given time, to diagnose whether a given choice of binder and bindee induces any “quirky effects” or not.
• If we find a pair that does not induce quirky effects, and we control for other relevant things like precedence, then BVA should only depend on structural factors, and we can test our desired hypotheses.

26

Plesniak (Under Review)

• There are various ways to achieve such control, but let me present a rather simple one.
• This is in fact the first full experiment of this type that I performed (circa early 2019).
• The experiment touches on more than just control of “quirky” factors, so we will only address certain parts of it (though I am happy to answer further questions later).
• Because of the preliminary nature of the investigation, rather than trying to find a non-quirky binder and bindee for each individual, we focused on just one choice, with ‘every’ as the quantifier and ‘his’ as the bindee.
• Other experiments have used multiple choices; I will touch on some of them briefly later and can also talk about them more during questions.

27

A bit of context

• What the experiment was attempting to demonstrate was that, under controlled conditions, BVA depends on c-command, and not something else (like precedence).
• As such, the crucial contrast was between sentences like (9) and (11), (rather than (9) and (10), because in both (9) and (11), binder does not precede bindee, but under “reconstruction”, ‘every teacher’ can c-command ‘his’ in (11), but no such option is available in (9).
• (9) His student spoke to every teacher.
• (10) Every teacher spoke to his student.
• (11) To his student, every teacher spoke.

28

BVA response patterns

• Based on participants’ responses as to the availability of BVA in (9) and (11) (repeated below), their judgements could be categorized in one of three ways.

​

• (9) His student spoke to every teacher.
• (11) To his student, every teacher spoke.

​

• Accepting BVA in (11) but not in (9): A c-command pattern
• Accepting BVA in (9) (regardless of (11)): A non-c-command pattern
• Anything else, e.g. not accepting BVA in either (9) or (11): A neutral pattern

29

Experiment Goals

​

• While overall, we expect a range of all three response patterns across individuals, when we control for interfering factors like quirky binding, we expect that no one should have the “non-c-command” pattern of judgement.

​

• We can talk about why either “c-command” or “neutral” patterns are possible under those conditions, but notice that neither of them allow for BVA when binder does not c-command bindee.

30

Experiment Goals

• The possibility of a “quirky binder” or a “quirky bindee” was to be checked by examining other types of interpretations.

​

• The possibility of a given element being a quirky binder/bindee (for a given individual at a given time) is known to be conserved across different interpretations.

​

• Indeed, another goal of this experiment was establishing precisely that conservation property.

31

Diagnosing Non-Quirky Binders

• Choices of binder that are quirky for BVA are also quirky for “distributive readings” (DR). Consider (12) below:

​

(12) Every teacher spoke to two students.

​

• The DR reading here would be that each teacher spoke to two different students, so if there are four teachers, eight students total were spoken to.

​

• (12) is an example of “surface scope DR”, which is generally acceptable.
• The key question is whether “inverse scope DR” is possible.

32

Diagnosing Non-Quirky Binders

​

(13) Two students spoke to every teacher.

​

• If an individual accepts the “two per each” reading in (13), then this may be the result of “every teacher” being a quirky binder for them.
• There are other reasons that might happen too, so we’d need to run more tests, but recall, our purpose is to find people for whom “every teacher” is not a quirky binder.
• Thus, we want to focus on individuals who reject DR in (13), in particular those for whom that rejection is accompanied by acceptance of DR in surface scope contexts, like (12), or even better, (14):

(14) To two students, every teacher spoke.

33

Diagnosing Non-Quirky Bindees

• A very similar procedure is used for the bindees, only now, the interpretation considered is simply coreference. Consider the below:
• (15) John spoke to his student.
• (16) To his student, John spoke.
• (17) His student spoke to John.
• John can be understood to c-command ‘his’ in (15) and (16), but not in (17).
• Accepting coreference in cases like (17) can be a potential indicator of ‘his’ being a quirky bindee for BVA, so we focus on those who reject coreference there but do accept coreference under c-command as in (16).
• Some of you may have questions about how constrained or not coreference is in English; I will be happy to address those further during Q&A.

34

Full Paradigm

• Putting the tests all together, we have the following:

​

(11) To his student, every teacher spoke.

(9) His student spoke to every teacher.

(14) To two students, every teacher spoke.

(13) Two students spoke to every teacher.

(16) To his student, John spoke.

(17) His student spoke to John.

​

• Those who accept DR in (14) but not (13) and accept Coref in (16) but not (17) are said to have neither quirky “every teacher” nor quirky “his”, and are thus in a “controlled environment” for judgement making.
• We thus definitively predict they will never accept BVA in (9), even though they may (and hopefully will) accept it in (11).

35

Working with Online Participants

• Since the experiment was to be deployed to participants recruited via a survey platform (Prolific), additional steps needed to be taken.
• Namely, we have to make sure that online participants understand the task, are paying attention, not guessing randomly, etc.

​

• This is not unusual in experimental linguistics (lots of experiments use things like “catch trials”).
• Those interested can see Hoji 2022b and Plesniak 2022c (Chapters 6 and 8 in the forthcoming book), and also Hoji 2015, from which the particular tests used in this experiment were adapted.

​

• Besides these, the methodology used is quite similar to how we would check our own judgements or those of our colleagues.

​

36

Results

• Let us represent each individual by a colored dot and position them (arbitrarily for now) in a diagram.
• Crucially, green and yellow individuals reject BVA without c-command, whereas red individuals accept it.
• Individuals represented:
• Total: 106
• C-command pattern: 36
• Neutral pattern: 33
• Non-c-command pattern: 37

37

CC

Neutral

Non-CC

Results

• First, based on the results of our “catch trials”, we can pick out those who seem to be paying attention/understanding the task (and whose data is therefore reliable).

​

• Those with this property:
• Total: 83
• C-command pattern: 30
• Neutral pattern: 28
• Non-c-command pattern: 25

38

CC

Neutral

Non-CC

Attentive Individuals

Results

• Now we add in consideration of whether the tests diagnosed each individual’s ‘every’ as being non-quirky.

​

• Those with both properties:
• Total: 17
• C-command pattern: 9
• Neutral pattern: 5
• Non-c-command pattern: 3

39

CC

Neutral

Non-CC

Attentive Individuals

Those for whom ‘every’ is not quirky.

Results

• Lastly, we add in consideration of whether the tests diagnosed each individual’s ‘his’ as being non-quirky.

​

• Those with all three properties:
• Total: 10
• C-command pattern: 7
• Neutral pattern: 3
• Non-c-command pattern: 0

40

CC

Neutral

Non-CC

Attentive Individuals

Those for whom ‘every’ is not quirky.

Those for whom ‘his’ is not quirky.

Questions

• Q: We just went from 100+ datapoints to just 10. Isn’t that bad?
• Actually, it’s good; the raw data was full of cases where the individual�in question did not meet the prerequisites�for our predictions to apply.
• When we focus on the individuals who did meet said�prerequisites, i.e., those for whom quirky�sources of BVA were not interfering, we found�universal rejection of (9) as predicted.
• Note that the way we diagnosed these individuals�did not involve checking any of their BVA judgements,�so this was not a case of “circular logic”.
• Essentially, we used preset filters to focus on the cases where the results�would have the maximum relevance and consequences for our hypotheses, and found the results universally matched our predictions.

41

Questions

• Q: But what about the other 90+ people?
• A simplistic answer would be that their �judgements were just irrelevant because�their non-structural sources of BVA were�causing interference.
• Actually, we can say more about them�than this (looking at where each red dot is�in the chart, for example, tells us which �source facilitated their c-command-less BVA �acceptance), but I will skip over that for now.
• However, the more satisfying answer is that�if we use more choices of binder and bindee,�we can find “controlled conditions” for many more,�possibly all individuals, as is done in Plesniak 2022a and Hoji 2022b,�for example.

42

Questions

• Q: Is this enough data to really be considered reliable?
• Technically, a new data can always contradict previously established results; we could use various statistical measures to assess our confidence, but in this case, the result has in fact been replicated many times (in various ways).

43

e.g., Plesniak 2022b (Chapter 7 of the book)

(More on next slide)

Other replications

• Just a small sampling of some of the other such replications.
• No need to get into the details of each graph now, but you can see the basic pattern.
• Plesniak 2022a also replicates the same pattern in Korean and Mandarin Chinese
• Hoji 2022a finds the same pattern across Japanese as well.

44

Plesniak 2022a (for English)

Hoj 2022b (for Japanese)

Korean Paradigm

• If you want to try yourself, you can check your judgements against sentences like those on the right (rough “translations” from the English).
• If you have non-Quirky binder and bindee (accepting DR and Coref in (20) and (22) and rejecting them in (21) and (23), then you are predicted to not accept BVA in (19).
• To go even further, you can try replacing the underlined terms with other ones and seeing what happens; the judgements may change, but the prediction should still hold.

45

• (18) 자기의 학생에게 모든 선생이 말했다.
• (19) 자기의 학생이 모든 선생에게 말했다. �
• (20) 두 명의 학생에게 모든 선생이 말했다.
• (21) 두 명의 학생이 모든 선생에게 말했다. �
• (22) 자기의 학생에게 존이 말했다.
• (23) 자기의 학생이 존에게 말했다.

Takeaways from this part

• There were lots of details in the previous slides, but the important things to take away are these:

​

• First, that it is possible to use diagnostic tests to find a controlled environment in which to test c-command-based hypotheses.

​

• And second, that when this is done, (assuming the hypotheses are correct), previously contentious judgements are shown to be underlyingly uniform.
• Judgement variation is principled.

46

Further Benefits

• In the previous example, we saw how controlling the sources of BVA allow us to peer through the chaos of judgement variation to something quite orderly.

​

• In the remaining time, I want to give an example of just one of the ways in which doing this can contribute to advancing generative theory.

​

• Specifically, let us look at a case where lack of clarity about the sources of BVA has led to contentious issues in the literature, which we can now resolve.

​

47

Possessor Binding (a.k.a. Spec Binding)

• Even in Reinhart (1983)’s original formulation of the c-command constraint on BVA, she acknowledged that it seems like possessors can “bind out” of the nominals that contain them, e.g.,:

​

(24) Every author’s mother loves his books.

​

• Such cases have featured heavily in debates about the nature of c-command, syntactic structure, and BVA.
• Kayne (1994), for example, was one of many who took it as evidence that the definition of c-command and other structural details needed to be revised.
• Such cases also feature heavily in Barker (2012)’s critique of BVA’s purported connection to c-command.
• The issue remains contentious to this day.

48

Korean Examples

• Consider your judgements on the availability of BVA in following sentences:

​

• “Regular”+ S-O order:� (24) 모든 작가가 자신이 쓴 책을 항상 칭찬한다.� ‘Every author always praises his own books.’
• “Regular”+ O-S order:� (25) 자신이 쓴 책을 항상 모든 작가가 칭찬한다.� ‘His own books, every author always praises.’
• Poss. Binding + S-O order:� (26) 모든 작가의 어머니가 자신이 쓴 책을 항상 칭찬한다.� ‘Every author’s mother always praises his own books.’
• Poss. Binding+ O-S Order� (27) 자신이 쓴 책을 모든 작가의 어머니가 항상 칭찬한다.� ‘His own books, every author’s mother always praises.’

​

​

​

49

Korean Examples (backup 1)

• In case no one accepts possessor binding with 자신 ‘himself’/ ‘his own’, based on Yoona’s judgements, sometimes 각자 ‘each’ is a little easier.

​

• “Regular”+ S-O order:� (24’) 모든 작가가 각자가 쓴 책을 항상 칭찬한다.� ‘Every author always praises his own books.’
• “Regular”+ O-S order:� (25’) 각자가 쓴 책을 항상 모든 작가가 칭찬한다.� ‘His own books, every author always praises.’
• Poss. Binding + S-O order:� (26’) 모든 작가의 어머니가 각자가 쓴 책을 항상 칭찬한다.� ‘Every author’s mother always praises his own books.’
• Poss. Binding+ O-S Order� (27’) 각자가 쓴 책을 모든 작가의 어머니가 항상 칭찬한다.� ‘His own books, every author’s mother always praises.’

​

​

​

50

Korean Examples (backup 2)

• In case no one gets the precedence effect with spec-binding with 자신/ 각자 , 그 ‘that’/’he’ is often good at demonstrating it.
• “Regular”+ S-O order:� (24’’) 모든 작가가 그가 쓴 책을 항상 칭찬한다.� ‘Every author always praises his own books.’
• “Regular”+ O-S order:� (25’’) 그가 쓴 책을 항상 모든 작가가 칭찬한다.� ‘His own books, every author always praises.’
• Poss. Binding + S-O order:� (26’’) 모든 작가의 어머니가 그가 쓴 책을 항상 칭찬한다.� ‘Every author’s mother always praises his own books.’
• Poss. Binding+ O-S Order� (27’’) 그가 쓴 책을 모든 작가의 어머니가 항상 칭찬한다.� ‘His own books, every author’s mother always praises.’

​

​

​

51

Points of Consideration

• First, not everyone accepts possessor binding (at least with certain choices of binder and bindee with which they accept “regular” binding).

​

• Second, among those that do accept it, it is often degraded by scrambling in a way that “regular” binding is not, i.e., for those individuals, suspiciously it cannot undergo reconstruction.
• This suggests the role of precedence, via Ueyama 1998’s “indexical dependency”.

​

• Finally, as we will see on the next slide, among those few that do accept the scrambled case, they are all diagnosed as having quirky binder/bindee.

52

Possessor Binding Results

• As we see, when we control for all interfering sources of BVA, no one accepts possessor binding.

​

• This result is replicated in Plesniak 2022a for English, Korean, and Mandarin Chinese.

​

• No reason to revise the definition of c-command, the structure of possessives, etc.

53

CC

Neutral

Non-CC

Attentive Individuals

Those for whom ‘the binder is not quirky.

Those for whom the bindee is not quirky.

Spec-Binding

Adapted from Plesniak 2022b

The Lesson

• The case of possessor binding demonstrates what can go wrong if we skirt around thorny empirical issues.

​

• For whatever reason, the relevant debates have frequently glossed over the fact that there is variation in whether possessor binding is possible, let alone tried to create a controlled environment for testing hypotheses about it.

​

• While there is still a lot we can ask about possessors, whatever probe we use to investigate them (BVA, DR, etc.), we need to be cautious in interpreting the results, taking into account the effects of individuals, lexical choices, and more.

54

Talk Summary

• Judgement variation is not something we need to suppress; it can be dealt with directly and doing so can be informative.

​

• The interpretations like BVA are useful probes for syntactic structures, provided that their sources are well controlled.
• This must be done at the level of the individual, and requires precise, lexical-item-sensitive diagnostics.

​

• When we do this, we find much clearer judgements than otherwise, which can shed new light on previously intractable problems.

55

Closing Thoughts

• One way that we provide evidence for a scientific theory is showing that its predictions hold even (and maybe especially) in extreme or atypical situations.
• What kind of data is relevant to a given theory depends on the particular hypotheses and what they predict.
• Unless you have a “theory of everything”, certain datapoints will not have crucial bearing on the hypotheses.
• In some sense, these two principles are in tension; we want to include as wide a range of data as possible, but exclude all “irrelevant” data.

56

Closing Thoughts

​

• The answer to this challenge that I offer here is that deciding which data to include should be:
• A. Systematic, and
• B. Firmly based in hypotheses.

​

• Crucially, if we believe in something called “Universal Grammar”, the fact that certain judgements are only held by a minority is not a good reason to disregard them.
• Indeed, such cases may represent the “extreme and/or atypical” situations in which we can most strongly support our hypotheses.

57

Closing Thoughts

​

• Of course, sometimes we cannot address every exceptional case at once, but regardless of what methods we use, we can keep in mind the ultimate goal of having theories that make successfully definite and universal predictions in their domain of relevance.

​

• Striving to do so is part of how generative linguistics can continue to grow as a scientific field.

58

Wrapping up

• Hopefully some of you feel curious or inspired to try to apply the sort of methods I have discussed today to your own work.

​

• Please feel free to reach out to me by email (plesniak@usc.edu) and/or to have a look at the forthcoming The Theory and Practice of Language Faculty Science.
• If you want to have a look before it is out/not have to pay for it, I may be able to figure something out.

​

• Of course, there is no substitute to simply thinking about where judgement variation might occur in your own area of interest and figuring out how you might account for that.
• You may consider looking at the sentence paradigms presented here and how they might be adapted to what you are working on.

59

Thank you!

​

​

60

Works Cited

• Barker Chris. 2012. Quantificational binding does not require c-command. Linguistic Inquiry 43(4). 614-633.
• Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
• Hayashishita, J.-R. 2004. Syntactic and Non-Syntactic Scope. PhD dissertation, University of Southern California.
• Hoji, Hajime. 2015. Language Faculty Science. Cambridge, UK: Cambridge University Press.
• Hoji, Hajime. 2022a. Detection of c-command effects. In Hoji, Hajime, Daniel Plesniak, and Yukinori Takubo (eds), The Theory and Practice of Language Faculty Science. De Gruyter Mouton.
• Hoji, Hajime. 2022b, Replication: Predicted correlations of judgement in Japanese. In Hoji, Hajime, Daniel Plesniak, and Yukinori Takubo (eds), The Theory and Practice of Language Faculty Science. Currently under contract with De Gruyter Mouton.
• Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.

61

Works Cited

• Plesniak, Daniel. 2022a. Towards a Correlational Law of Language: Three Factors Constraining Judgement Variation. Los Angeles: University of Southern California PhD dissertation.
• Plesniak, Daniel. 2022b. Predicted correlations of judgements in English. In Hajime Hoji, Daniel Plesniak, and Yukinori Takubo (eds.) The Theory and Practice of Language Faculty Science. Berlin: Mouton de Gruyter.
• Plesniak, Daniel. 2022c. Implementing experiments on the language faculty. In Hoji, Hajime, Daniel Plesniak, and Yukinori Takubo (eds), The Theory and Practice of Language Faculty Science. Currently under contract with De Gruyter Mouton.
• Plesniak, Daniel. Under Review. Possibility-seeking experiments: Testing syntactic hypotheses on the level of the individual.
• Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. Chicago: University of Chicago Press.
• Ueyama, Ayumi. 1998. Two Types of Dependency. PhD dissertation, University of Southern California.

62