1 of 83

Soft implicative entailments

Aaron Steven White

University of Maryland

LSA Annual Meeting

January 10th, 2015

2 of 83

Semantics and Pragmatics

Primary data

Inferential connections between sentences

Classification

e.g. implicature

for descriptive adequacy

entailment

v. non-entailment

3 of 83

Entailment

Bo managed to get a raise.

Bo got a raise.

strong inference

Karttunen 1971, Horn 1972

4 of 83

Contradiction

1a. Bo managed to get a raise

#but he didn’t get a raise.

Karttunen 1971, Horn 1972

5 of 83

Entailment

Bo managed to get a raise.

Bo got a raise.

strong inference

entails that

Karttunen 1971, Horn 1972

6 of 83

Non-entailment

Bo was able to get a raise.

Bo got a raise.

strong inference

Karttunen 1971, Bhatt 1999

7 of 83

Contradiction

1a. Bo managed to get a raise

#but he didn’t get a raise.

Karttunen 1971, Bhatt 1999

1b. Bo was able to get a raise

but he didn’t get a raise.

8 of 83

Entailment

Bo was able to get a raise.

Bo got a raise.

infer

doesn’t entail that

Karttunen 1971, Bhatt 1999

9 of 83

Entailment

Bo didn’t want to get a raise.

Bo got a raise.

Bo didn’t get a raise.

equally likely

10 of 83

Observation

Entailments are lexically triggered

manage

be able

want

lexical

entailment

strong lexical inference

no lexical

inference

This suggests that these inference patterns are lexically triggered in interesting ways. As is standard in the literature, I'll refer to the entailments as lexical entailments of the predicate (though keep in mind that this is just shorthand for saying they're entailments of sentences containing the predicate). And more generally, I'll refer to all of these predicate sensitive inferences---entailment or not---as lexical inferences of the predicate. Manage then lexically entails its complement, while be able triggers a strong-but-not-necessary lexical inference that its complement is true and want has no lexical inference whatsoever. In this talk, I’ll focus specifically on the sort of strong-but-not-necessary lexical inferences---which I’ll just call strong lexical inferences from now on---that be able gives rise to.

11 of 83

Question

Problem

What is the nature of strong lexical inference?

The facts aren’t clear cut.

12 of 83

Goals

Sharpen data on lexical inference
Show three patterns of lexical inference

Pattern #1: remember
Pattern #2: refuse
Pattern #3: mean

At least one pattern is actuality entailment

13 of 83

This talk

Background on lexical inferences
Two murky cases
Experiment
Explanatory sketch
Conclusion

14 of 83

Question

Two types of answer

How do strong lexical inferences arise?

Via the pragmatics

(e.g. conversational implicature)

15 of 83

Explanation #1: Implicature

“Bo was able to get a raise.”
Ability is a necessary (though not sufficient) condition for doing something
Speaker would have said more if other necessary condition weren’t satisfied.
Therefore, all others must be satisfied.
Therefore, Bo got a raise.

16 of 83

Question

Two types of answer

How do strong lexical inferences arise?

Via the pragmatics

(e.g. conversational implicature)

Via the semantics

17 of 83

Explanation #2: lexical semantics

manage

NP managed to VP NP VPed

Karttunen 1971, Karttunen & Peters 1979

see also Baglini & Francez 2013, Nadathur 2014

be able

NP managed to VP ? NP VPed

18 of 83

Contradiction

1a. Bo managed to get a raise

#but he didn’t get a raise.

Karttunen 1971, Bhatt 1999

1b. Bo was able to get a raise

but he didn’t get a raise.

19 of 83

Complications

English may be hiding relevant distinctions in the grammatical element surrounding be able

Bhatt 1999

20 of 83

Aspectually conditioned entailments

2a. Borusa na sikoso afto to trapezi ala den to sikosa

“I could.IMPFV lift this table, but I didn’t lift it.”

2b. Boresa na tu miliso #ala den tu milisa.

“I [be able].PFV to talk to John #but I did not talk to him.”

Bhatt’s (319a) and (b)

Bhatt 1999

21 of 83

Actuality entailment

Bhatt 1999, Hacquard 2006, Mari & Martin 2007

Brogonovo & Cummins 2007, Homer 2009, Kratzer 2011

be able-PERF p p

be able-IMPF p ? p

22 of 83

Actuality entailment

Bhatt 1999, Hacquard 2006, Mari & Martin 2007

Brogonovo & Cummins 2007, Homer 2009, Kratzer 2011

ROOT-PERF p p

ROOT-IMPF p ? p

23 of 83

Aspectually conditioned entailments

I was able to lift the table.

I [was able]-IMPF...

I [was able]-PERF...

I lifted the table.

entails that

???

24 of 83

Actuality entailment

3a. In those days, a fireman was able to lift 200 pounds.

but none actually did.

3b. Yesterday, a fireman was able to lift 200 pounds.

?#but none actually did.

Bhatt 1999

25 of 83

Question

Two types of answer

How do strong lexical inferences arise?

Via the pragmatics

(e.g. conversational implicature)

Via the semantics

26 of 83

Relevance

Other strong lexical inferences may have similar explanations.

27 of 83

This talk

Background on lexical inferences
Two problem cases
Experiment
Explanatory sketch
Conclusion

28 of 83

Unclear Case #1: entailment or not

Bo took out the trash.

Bo to take out the trash.

managed

remembered

entails that

29 of 83

Unclear Case #1: entailment or not

Bo didn’t to take out the trash.

Bo didn’t take out the trash.

remember

manage

entails that

Karttunen 1971

30 of 83

Unclear Case #1: entailment or not

Bo didn’t remember that he took out the trash.

Bo take out the trash.

presupposes that

Karttunen 1971

31 of 83

Unclear Case #1: entailment or not

Bo took out the trash.

Bo to take out the trash.

remembered

entails that

van Leusen 2012

32 of 83

Unclear Case #1: entailment or not

Bo didn’t to take out the trash.

Bo didn’t take out the trash.

remember

entails that

strong inference

van Leusen 2012

33 of 83

Question

Are these inferences pragmatic or semantic?

34 of 83

Unclear Case #2: entailment or not

Bo to check his voicemail.

Bo didn’t check his voicemail.

refused

meant

weak inference

entails that

35 of 83

Unclear Case #2: entailment or not

A: You just spilled soup everywhere.

B: I know. I meant to do that.

36 of 83

Unclear Case #2: entailment or not

Bo didn’t to check his voicemail.

Bo checked his voicemail.

refuse

mean

inference

37 of 83

General problem

Hard to pin down inference strength

remember: is inference entailed or not?

mean, refuse: does inference exist?

38 of 83

Question

Are these inferences pragmatic or semantic?

39 of 83

This talk

Background on lexical inferences
Two problem cases
Experiment
Explanatory sketch
Conclusion

40 of 83

cf. Karttunen 2014

41 of 83

42 of 83

Contradiction Detection Task

Similar to Karttunen 2014...

...except uses response scale that asks about necessary truth

43 of 83

Experiment materials

12 implicatives

+ 2 non-implicatives

= 14 verbs

manage
opt
remember
think
know
want
hope

hasten
hesitate
mean
refuse
forget
neglect
fail

44 of 83

Experiment materials

Two factors

Positive v. negative matrix polarity (Sent. 1)

Positive v. negative followup (Sent. 2)

45 of 83

No inference

=

close to midpoint

Strong negative inference

=

far below midpoint

Strong positive inference

=

far above midpoint

I’ll present the data on a -1 to 1 scale with negative polarity contexts marked orange and positive polarity contexts marked grey. A verb with weak or non-existent inferences should show up with bars close to the midpoint. A verb with a strong negative inference with a certain polarity will show up substantially below the midpoint. And a verb with a strong positive inference will show up substantially below the midpoint. Note that what I mean by inference strength here is the proportion of times a participant said that the second sentence was necessarily true or false given the first. This is relevant. First, I’ll show the controls. As you can see, both want and hope show up close to the midpoint though their is some error in the judgment. As we expect, manage shows an extremely strong negative inference with negative polarity and an extremely strong positive inference with positive polarity. It’s inverse, fail shows up with exactly the opposite pattern. Bo failed to get a raise entails that he didn’t get one, and Bo didn’t fail to get a raise entails that he got one.

46 of 83

Summary

Classic entailment patterns corroborated

(e.g. manage, fail)

47 of 83

*

48 of 83

Summary

Classic entailment patterns corroborated

(e.g. manage, fail)

Significant distinctions found

(e.g. manage v. remember, fail v. forget)

49 of 83

50 of 83

Summary

Classic entailment patterns corroborated

(e.g. manage, fail)

Significant distinctions found

(e.g. manage v. remember, fail v. forget)

Interesting newly discovered patterns

(e.g. refuse, mean)

51 of 83

Summary

Classic entailment patterns corroborated

(e.g. manage, fail)

Significant distinctions found

(e.g. manage v. remember, fail v. forget)

Interesting newly discovered patterns

(e.g. refuse, mean)

52 of 83

Summary

Classic entailment patterns corroborated

(e.g. manage, fail)

Significant distinctions found

(e.g. manage v. remember, fail v. forget)

Interesting newly discovered patterns

(e.g. refuse, mean)

53 of 83

This talk

Background on lexical inferences
Two problem cases
Experiment
Explanatory sketch
Conclusion

54 of 83

Claim

The strong lexical inferences of remember are actuality entailments

White 2014

55 of 83

Presupposition v. entailment

Bo didn’t remember to take out the trash.

Bo didn’t take out the trash.

strong inference

56 of 83

Presupposition v. entailment

Bo didn’t remember that he took out the trash.

Bo took out the trash.

presupposes that

57 of 83

Presupposition v. entailment

Bo didn’t remember to take out the trash.

Bo was supposed to take out the trash.

presupposes that

Karttunen 1971

58 of 83

Presupposition v. entailment

Bo didn’t remember that he had to take out the trash.

Bo had to take out the trash.

presupposes that

59 of 83

Covert modal

Bo didn’t remember to take out the trash.

Bo had to take out the trash.

presupposes that

contains root modal

White 2014

cf. Bhatt 1999, Hackl & Nissenbaum 2012, Wurmbrand 2014

60 of 83

Claim

Lexical inference of remember to arises as actuality entailment of covert modal

61 of 83

Bo didn’t remember that he had to take out the trash.

Bo took out the trash.

inference

62 of 83

Bo didn’t remember to take out the trash.

Bo took out the trash.

inference

63 of 83

4. In those days…

a. ...a boy remembered that he had to wear his tie

but I’m not sure I ever saw one actually wear it.

b. ...a boy remembered to wear his tie

but I’m not sure I ever saw one actually wear it.

64 of 83

5. Yesterday…

a. ...a boy remembered that he had to wear his tie

but I doubt he actually wore it.

b. ...a boy remembered to wear his tie

#but I doubt he actually wore it.

65 of 83

Question

Answer

Why should the nonfinite version be sensitive to a matrix adverb?

The nonfinite cases restructure.

White 2014

66 of 83

Restructuring

6. Jo told Bo to fill the bird feeder.

a. He remembered this but didn’t fill it. The next week,

she asked again.

Bo again remembered that he had to fill the feeder.

b. He forgot this, but noticed it was empty and filled it

anyway. She told him to again the next week.

#Bo again remembered that he had to fill the feeder.

67 of 83

Restructuring

6. Jo told Bo to fill the bird feeder.

a. He remembered this but didn’t fill it. The next week,

she asked again.

#Bo again remembered to fill the feeder.

b. He forgot this, but noticed it was empty and filled it

anyway. She told him to again the next week.

#Bo again remembered to fill the feeder.

68 of 83

Restructuring

Again modifies embedded event…

...in nonfinite context

...not in finite context

69 of 83

Restructuring

Embedded event closed by matrix aspect…

...in nonfinite sentence

....not in nonfinite sentence

70 of 83

5. Yesterday…

a. ...a boy remembered that he had to wear his tie

but I doubt he actually wore it.

b. ...a boy remembered to wear his tie

#but I doubt he actually wore it.

71 of 83

Conclusion

Raised puzzle about lexical inference
Lexical inference = entailment obscured by ambiguity in English for some predicates
Other murky cases of lexical inference
Sharpened judgments of murky cases
Remember derived in the same way as be able.

72 of 83

Open Questions

How are mean and refuse-type verbs explained?

73 of 83

74 of 83

Open Questions

How are mean and refuse-type verbs explained?
What patterns of lexical inference are possible?

75 of 83

76 of 83

Thank you

UMD Attitude Verbs Group

National Science Foundation

BCS (1124338), IGERT (0801465)

77 of 83

Experiment materials

Two factors

Positive v. negative matrix polarity (Sent. 1)

Positive v. negative followup (Sent. 2)

Four items per condition

50% fillers constructed from non-control frames

78 of 83

Experiment materials

Verbs

manage
opt
remember
think
know
want
hope

hasten
hesitate
mean
refuse
forget
neglect
fail

79 of 83

Experiment materials

Subject-embedded predicate combinations

The businessperson Ved to close the sales deal.
The student Ved to take the professor’s class.
The author Ved to return the publisher’s call.
The customer Ved to talk to the manager.

80 of 83

Experiment implementation

Implemented on Ibex (experiment link)

300 participants from Amazon Mechanical Turk

native English speaker qualification test

95% acceptance rate

filtered by reaction time distributions

81 of 83

*

n.s.