[PPT] - Linguists for Deep Learning; or: How I Learned to Stop Worrying and PowerPoint Presentation

SLIDE 1

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Linguists for Deep Learning;

r: How I Learned to Stop Worrying and

Love Neural Networks

Christopher Potts

Stanford Linguistics

*Sem 2018, June 5–6, New Orleans

1 / 42

SLIDE 2

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Linguists for Deep Learning;

r: How I Learned to Stop Worrying and

Love Neural Networks

Christopher Potts

Stanford Linguistics

*Sem 2018, June 5–6, New Orleans

1 / 42

SLIDE 3

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Linguists for Deep Learning;

r: How I Learned to Stop Worrying and

Love Neural Networks

Christopher Potts

Stanford Linguistics

*Sem 2018, June 5–6, New Orleans

1 / 42

SLIDE 4

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

2 / 42

SLIDE 5

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

Yann LeCun in 2015 [link]

The next frontier for Deep Learning is natural language understanding.

2 / 42

SLIDE 6

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

Yann LeCun in 2015 [link]

The next frontier for Deep Learning is natural language understanding.

Yann LeCun at Stanford in 2018 [link]

I would say language is number 300 in the list of 500 problems that we need to face.

2 / 42

SLIDE 7

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

Yann LeCun in 2015 [link]

The next frontier for Deep Learning is natural language understanding.

Yann LeCun at Stanford in 2018 [link]

I would say language is number 300 in the list of 500 problems that we need to face. Did deep learning swerve off the road instead?

2 / 42

SLIDE 8

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

But what does this mean?

2 / 42

SLIDE 9

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

But what does this mean?

If deep learning brings useful tools, ideas, and

insights to another field, has it thereby damaged that field? I’d say the opposite!

2 / 42

SLIDE 10

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

But what does this mean?

If deep learning brings useful tools, ideas, and

insights to another field, has it thereby damaged that field? I’d say the opposite!

So what potential does deep learning have to

improve the science of language?

2 / 42

SLIDE 11

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Signs of the apocalypse?

Neil Lawrence in 2015, quoted by Manning (2015)

NLP is kind of like a rabbit in the headlights of the Deep Learning machine, waiting to be flattened.

But what does this mean?

If deep learning brings useful tools, ideas, and

insights to another field, has it thereby damaged that field? I’d say the opposite!

So what potential does deep learning have to

improve the science of language?

My argument today

Deep learning has much to offer the study of linguistic meaning and communication.

2 / 42

SLIDE 12

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Lexical semantics

3 / 42

SLIDE 13

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Dimensions of lexical meaning

c1 c2 c3 c4 c5 · · · w1 w2 w4 w5 w6 . . .

4 / 42

SLIDE 14

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Dimensions of lexical meaning

c1 c2 c3 c4 c5 · · · w1 w2 w4 w5 w6 . . .

4 / 42

SLIDE 15

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Dimensions of lexical meaning

c1 c2 c3 c4 c5 · · · w1 w2 w4 w5 w6 . . . The stock deteriorated.

4 / 42

SLIDE 16

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The neglect of lexical meaning in semantics

5 / 42

SLIDE 17

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The neglect of lexical meaning in semantics

Thomason (1974)

The problems of a semantic theory should be distinguished from those of lexicography [. . . ]

5 / 42

SLIDE 18

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The neglect of lexical meaning in semantics

Thomason (1974)

The problems of a semantic theory should be distinguished from those of lexicography [. . . ] A central goal of (semantics) is to explain how different kinds of meanings attach to different syntactic categories; another is to explain how the meanings of phrases depend on those of their components. [. . . ]

5 / 42

SLIDE 19

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The neglect of lexical meaning in semantics

Thomason (1974)

The problems of a semantic theory should be distinguished from those of lexicography [. . . ] A central goal of (semantics) is to explain how different kinds of meanings attach to different syntactic categories; another is to explain how the meanings of phrases depend on those of their components. [. . . ] But we should not expect a semantic theory to furnish an account of how any two expressions belonging to the same syntactic category differ in meaning.

5 / 42

SLIDE 20

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The neglect of lexical meaning in semantics

Thomason (1974)

The problems of a semantic theory should be distinguished from those of lexicography [. . . ] A central goal of (semantics) is to explain how different kinds of meanings attach to different syntactic categories; another is to explain how the meanings of phrases depend on those of their components. [. . . ] But we should not expect a semantic theory to furnish an account of how any two expressions belonging to the same syntactic category differ in meaning. “Walk” and “run,” for instance, and “unicorn” and “zebra” certainly do differ in meaning, and we require a dictionary of English to tell us how. But the making of a dictionary demands considerable knowledge of the world.

5 / 42

SLIDE 21

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Jerrold Katz (1972) on meaning

The arbitrariness of the distinction between form and matter reveals itself [. . . ]

6 / 42

SLIDE 22

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Jerrold Katz (1972) on meaning

The arbitrariness of the distinction between form and matter reveals itself [. . . ] The question “What is meaning?” broken down:

What is synonymy?
What is antonymy?
What is superordination?
What is semantic ambiguity?
What is semantic truth?
What is a possible answer to a question?
. . .

6 / 42

SLIDE 23

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Children are situated word learners

Children learn word meanings

1. with incredible speed
2. despite relatively few inputs
3. by using cues from

◮ contrast inherent in the forms they hear ◮ social cues ◮ assumptions about the speaker’s goals ◮ regularities in the physical environment.

(Frank et al. 2012; Frank & Goodman 2014)

7 / 42

SLIDE 24

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

8 / 42

SLIDE 25

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

High-dimensional

8 / 42

SLIDE 26

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

High-dimensional
Meaning from dense linguistic inter-relationships

8 / 42

SLIDE 27

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

High-dimensional
Meaning from dense linguistic inter-relationships
Meaning solely from (nth-order) co-occurrence

8 / 42

SLIDE 28

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

High-dimensional
Meaning from dense linguistic inter-relationships
Meaning solely from (nth-order) co-occurrence
No grounding in physical or social contexts

8 / 42

SLIDE 29

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Purely distributional meaning

High-dimensional
Meaning from dense linguistic inter-relationships
Meaning solely from (nth-order) co-occurrence
No grounding in physical or social contexts
Not symbolic

8 / 42

SLIDE 30

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Faruqui et al. (2015): Retrofitting to graphs

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βij

qi − qj
2

Balances fidelity to the

riginal vector ˆ

qi against looking more like

ne’s graph neighbors.

Forces are balanced with α = 1 and β =

1 Degree(i)

9 / 42

SLIDE 31

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Faruqui et al. (2015): Retrofitting to graphs

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βij

qi − qj
2

Balances fidelity to the

riginal vector ˆ

qi against looking more like

ne’s graph neighbors.

Forces are balanced with α = 1 and β =

1 Degree(i)

9 / 42

SLIDE 32

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

What retrofitting to WordNet might do

10 / 42

SLIDE 33

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

What retrofitting to WordNet might do

Cluster mammal with dog and puppy even though

mammal has a different, unusual distribution.

10 / 42

SLIDE 34

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

What retrofitting to WordNet might do

Cluster mammal with dog and puppy even though

mammal has a different, unusual distribution.

Avoid polarity mistakes like modeling superb and

awful as similar (though beware antonym edges!).

10 / 42

SLIDE 35

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

What retrofitting to WordNet might do

Cluster mammal with dog and puppy even though

mammal has a different, unusual distribution.

Avoid polarity mistakes like modeling superb and

awful as similar (though beware antonym edges!).

Holistic consistency:

Figure 3: Two-dimensional PCA projections of 100-dimensional SG vector pairs holding the “adjective to adverb” relation, before (left) and after (right) retrofitting.

10 / 42

SLIDE 36

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Concerns about identity retrofitting

No attention to edge

semantics; edges mean ‘similar to’.

Presupposes a uniform

initial embedding space.

No modeling of missing

edges.

Athelas Kingsfoil Black Breath Nazgûl Aragorn

Is Treats Treats Uses Causes

11 / 42

SLIDE 37

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Hand-build functions from Mrkˇ si´ c et al. (2016)

AntonymRepel:
(i,j)∈A

ReLU

1.0 − d(qi, qj)
SynonymAttract:
(i,j)∈S

ReLU

d(qi, qj) − 0
VectorSpacePreservation:
i
j∈N(i)

ReLU

d(qi, qj) − d( ˆ

qi, ˆ qj)

12 / 42

SLIDE 38

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

13 / 42

SLIDE 39

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

Faruqui et al.

fr(qi, qj) =

qi − qj
2

with βijr = 0

13 / 42

SLIDE 40

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

Linear

fr(qi, qj) =

Arqj + br − qi
2
ρ(fr) = Ar2
We initialize Ar = 1 and br = 0
Initialization can be different for different relations,

e.g., Aantonym = −1

13 / 42

SLIDE 41

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

Simplest neural (akin to Sutskever et al. 2009)

fr(qi, qj) = tanh(q⊤

i Arqj)

13 / 42

SLIDE 42

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

Neural T ensor Network (akin to Socher et al. 2013)

fr(qi, qj) = ur⊤ tanh(q⊤

i Arqj)

where Ar ∈ Rd×d×k and ρ(fr) = Ar2 + ur2

13 / 42

SLIDE 43

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional relations (Lengerich et al. 2018)

Framework

i∈V

αi

qi − ˆ

qi

2 +
(i,j,r)∈E

βijrfr(qi, qj) −

(i,j,r)∈E−

βijrfr(qi, qj) + λ

r

ρ(fr)

Your favorite graph embedding method

Bordes et al. 2013; Wang et al. 2014; Lin et al. 2015; for

verviews, see Nickel et al. 2011; Hamilton et al. 2017.

13 / 42

SLIDE 44

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

FrameNet evaluation

Model ‘Inheritance’ ‘Using’ ‘Reframing’ ‘Subframe’ ‘Perspective On’ (2132/992) (1552/668) (544/312) (356/168) (336/148) None 87.58 88.59 85.60 91.24 89.59 Faruqui et al. 90.79 87.87 87.02 94.50 94.24 FR-Linear 92.92 92.04 89.37 94.65 94.73 FR-Neural 92.46 92.54 89.57 95.65 94.04 Model ‘Precedes’ ‘See Also’ ‘Causative Of’ ‘Inchoative Of’ (220/136) (268/76) (204/36) (60/16) None 87.30 85.11 86.11 82.50 Faruqui et al. 85.26 83.81 84.49 78.33 FR-Linear 87.00 91.93 92.09 82.50 FR-Neural 89.16 93.25 94.33 85.00

14 / 42

SLIDE 45

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A drug–disease knowledge graph

Faruqui et al. FR-Linear Model ‘Treats’ (9152/2490) None 72.02 ± 0.50 Faruqui et al. 72.93 ± 0.82 FR-Linear 84.22 ± 0.82 FR-Neural 73.52 ± 0.89

15 / 42

SLIDE 46

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Functional complexity in the lexicon

Category Semantic type noun properties intransitive verbs properties transitive verbs entities to properties adjectives properties to properties prepositions entities to (properties to properties) determiner properties to sets of properties Vector space models tend to be monotyped, but see Clark et al. 2011.

16 / 42

SLIDE 47

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some lexical generalizations

17 / 42

SLIDE 48

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some lexical generalizations

1. Some transitive verbs entail the existence of their

direct object (see) and some do not (seek).

17 / 42

SLIDE 49

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some lexical generalizations

1. Some transitive verbs entail the existence of their

direct object (see) and some do not (seek).

2. Across languages, verbs lexicalize manner or result,

but not both (Rappaport Hovav & Levin 2010):

◮ Manner: nibble, scribble, sweep, flutter ◮ Result: clean, cover, empty, fill 17 / 42

SLIDE 50

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some lexical generalizations

1. Some transitive verbs entail the existence of their

direct object (see) and some do not (seek).

2. Across languages, verbs lexicalize manner or result,

but not both (Rappaport Hovav & Levin 2010):

◮ Manner: nibble, scribble, sweep, flutter ◮ Result: clean, cover, empty, fill

3. Some adjectives predicate distributively across

their arguments, others do not (Glass 2018):

◮ Box A and Box B are new.

(entails both are new)

◮ Box A and Box B are heavy.

(does not entail both are heavy)

17 / 42

SLIDE 51

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some lexical generalizations

1. Some transitive verbs entail the existence of their

direct object (see) and some do not (seek).

2. Across languages, verbs lexicalize manner or result,

but not both (Rappaport Hovav & Levin 2010):

◮ Manner: nibble, scribble, sweep, flutter ◮ Result: clean, cover, empty, fill

3. Some adjectives predicate distributively across

their arguments, others do not (Glass 2018):

◮ Box A and Box B are new.

(entails both are new)

◮ Box A and Box B are heavy.

(does not entail both are heavy) Can we develop deep learning systems that derive such generalizations? No training against them; that’s just restating them!

17 / 42

SLIDE 52

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional semantics

18 / 42

SLIDE 53

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A semanticist’s ideal

Every student attended a lecture

∀z ((student z) → (∃x (lecture x) ∧ (attended x z))) λg ∀z ((student z) → (g z)) λf λg ∀z ((f z) → (g z)) student λy (∃x (lecture x) ∧ (attended x y)) attended λg (∃x (lecture x) ∧ (g x)) λf λg (∃x (f x) ∧ (g x)) lecture

19 / 42

SLIDE 54

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A semanticist’s ideal

Every student attended a lecture

∀z ((student z) → (∃x (lecture x) ∧ (attended x z))) λg ∀z ((student z) → (g z)) λf λg ∀z ((f z) → (g z)) student λy (∃x (lecture x) ∧ (attended x y)) attended λg (∃x (lecture x) ∧ (g x)) λf λg (∃x (f x) ∧ (g x)) lecture

But is this really so ideal?

19 / 42

SLIDE 55

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Complete semantic representations?

MacCartney & Manning (2009)

The difficulty is plain: truly natural language is fiendishly complex. [. . . ] Consider for a moment the difficulty of fully and accurately translating

1. Every firm polled saw costs grow more than

expected, even after adjusting for inflation. to a formal meaning representation.

20 / 42

SLIDE 56

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Sparse, fragmented feature representations

S NP The NYT VP reported S NP the deal VP fell through

the 2 source_NYT T NYT 1 embedded_implicit_neg T report 1 deal_neg 1 length 7 vocab 6 . . . . . .

21 / 42

SLIDE 57

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The answer from deep learning

S1 = f

NP1 , VP1
NP1

= f

The , NYT
The

The NYT NYT VP1 = f

reported , S2
reported

reported S2 = f

NP2 , VP2
NP2

= f

the , deal
the

the deal deal VP2 = f

fell , through
fell

fell through through 22 / 42

SLIDE 58

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The answer from deep learning

h1 = f

h0 , The
The

The h2 = f

h1 , NYT
NYT

NYT h3 = f

h2 , reported
reported

reported h4 = f

h3 , the
the

the h5 = f

h4 , deal
deal

deal h6 = f

h5 , fell
fell

fell h7 = f

h6 , through
through

through 23 / 42

SLIDE 59

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The answer from deep learning

h0 h1 h2 h3 h4 h5 h6 h7 The NYT reported the deal fell through The NYT reported the deal fell through

24 / 42

SLIDE 60

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The answer from deep learning

All our parses are wrong, but perhaps we can discover the right one(s).

h0 h1 h2 h3 h4 h5 h6 h7 The NYT reported the deal fell through The NYT reported the deal fell through

25 / 42

SLIDE 61

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A new perspective on compositionality

26 / 42

SLIDE 62

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A new perspective on compositionality

Compositionality

The meaning of a complex phrase is a function of the meaning of its consituent phrases.

26 / 42

SLIDE 63

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A new perspective on compositionality

Compositionality

The meaning of a complex phrase is a function of the meaning of its consituent phrases.

Partee (1984):

Context-dependence, Ambiguity, and Challenges to Local, Deterministic Compositionality

26 / 42

SLIDE 64

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

A new perspective on compositionality

Compositionality

The meaning of a complex phrase is a function of the meaning of its consituent phrases.

Partee (1984):

Context-dependence, Ambiguity, and Challenges to Local, Deterministic Compositionality

S1 = tanh

NP1 ; VP1 W + b
NP1

VP1 = tanh

reported ; S2 W + b
reported

S2 26 / 42

SLIDE 65

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

(Bowman 2017)

27 / 42

SLIDE 66

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

Kim smoked. ↑ Kim smoked cigars. Kim didn’t smoke. ↓ Kim didn’t smoke cigars. (Bowman 2017)

27 / 42

SLIDE 67

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

Kim smoked. ↑ Kim smoked cigars. Kim didn’t smoke. ↓ Kim didn’t smoke cigars. A student smoked. ր տ A Swedish student smoked. A student smoked cigars. (Bowman 2017)

27 / 42

SLIDE 68

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

Kim smoked. ↑ Kim smoked cigars. Kim didn’t smoke. ↓ Kim didn’t smoke cigars. A student smoked. ր տ A Swedish student smoked. A student smoked cigars. No student smoked. ւ ց No Swedish student smoked. No student smoked cigars. (Bowman 2017)

27 / 42

SLIDE 69

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

Kim smoked. ↑ Kim smoked cigars. Kim didn’t smoke. ↓ Kim didn’t smoke cigars. A student smoked. ր տ A Swedish student smoked. A student smoked cigars. No student smoked. ւ ց No Swedish student smoked. No student smoked cigars. Every student smoked. ւ տ Every Swedish student smoked. Every student smoked cigars. (Bowman 2017)

27 / 42

SLIDE 70

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Compositional generalizations: monotonicity

Kim smoked. ↑ Kim smoked cigars. Kim didn’t smoke. ↓ Kim didn’t smoke cigars. A student smoked. ր տ A Swedish student smoked. A student smoked cigars. No student smoked. ւ ց No Swedish student smoked. No student smoked cigars. Every student smoked. ւ տ Every Swedish student smoked. Every student smoked cigars. Most students smoked. − տ Most Swedish students smoked. Most students smoked cigars. (Bowman 2017)

27 / 42

SLIDE 71

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Pragmatics

28 / 42

SLIDE 72

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

29 / 42

SLIDE 73

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.

29 / 42

SLIDE 74

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

29 / 42

SLIDE 75

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

3. I am here.

[NAACL; New Orleans; planet earth; . . . ]

29 / 42

SLIDE 76

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

3. I am here.

[NAACL; New Orleans; planet earth; . . . ]

4. We are here.

[pointing at a map]

29 / 42

SLIDE 77

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

3. I am here.

[NAACL; New Orleans; planet earth; . . . ]

4. We are here.

[pointing at a map]

5. I’m not here now.

[answering machine]

29 / 42

SLIDE 78

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

3. I am here.

[NAACL; New Orleans; planet earth; . . . ]

4. We are here.

[pointing at a map]

5. I’m not here now.

[answering machine]

6. We went to a local bar after the workshop.

29 / 42

SLIDE 79

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

1. I am speaking.
2. We won.

[A team I’m on; a team I support; . . . ]

3. I am here.

[NAACL; New Orleans; planet earth; . . . ]

4. We are here.

[pointing at a map]

5. I’m not here now.

[answering machine]

6. We went to a local bar after the workshop.
7. three days ago, tomorrow, now, . . .

29 / 42

SLIDE 80

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?

29 / 42

SLIDE 81

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.

29 / 42

SLIDE 82

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.

29 / 42

SLIDE 83

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.

29 / 42

SLIDE 84

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.

29 / 42

SLIDE 85

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.

29 / 42

SLIDE 86

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,

29 / 42

SLIDE 87

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,
b. as long as we don’t slip in the premise that they

have jet packs.

29 / 42

SLIDE 88

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,
b. as long as we don’t slip in the premise that they

have jet packs.

10. I didn’t see any.

29 / 42

SLIDE 89

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,
b. as long as we don’t slip in the premise that they

have jet packs.

10. I didn’t see any.
a. Are there typos in my slides?

29 / 42

SLIDE 90

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,
b. as long as we don’t slip in the premise that they

have jet packs.

10. I didn’t see any.
a. Are there typos in my slides?
b. Are the cookies in cupboard.

29 / 42

SLIDE 91

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

8. Where are you from?
a. Connecticut.
b. Stanford.
c. The U.S.
d. Planet earth.
9. If Kangaroos had no tails, they would fall over.
a. True,
b. as long as we don’t slip in the premise that they

have jet packs.

10. I didn’t see any.
a. Are there typos in my slides?
b. Are the cookies in cupboard.
c. Are there bookstores downtown?

29 / 42

SLIDE 92

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

29 / 42

SLIDE 93

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

12. Would you mind if I stole your pen for a second?

[politeness]

29 / 42

SLIDE 94

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

12. Would you mind if I stole your pen for a second?

[politeness]

13. He’s not exactly a genius/idiot.

[irony]

29 / 42

SLIDE 95

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

12. Would you mind if I stole your pen for a second?

[politeness]

13. He’s not exactly a genius/idiot.

[irony]

14. Great idea!

[sarcasm(?)]

29 / 42

SLIDE 96

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

12. Would you mind if I stole your pen for a second?

[politeness]

13. He’s not exactly a genius/idiot.

[irony]

14. Great idea!

[sarcasm(?)]

15. Any chance we can sort this out here, officer?

[bribery(?)]

29 / 42

SLIDE 97

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Natural language is situated and social

11. I think this is the way to the library.

[politeness]

12. Would you mind if I stole your pen for a second?

[politeness]

13. He’s not exactly a genius/idiot.

[irony]

14. Great idea!

[sarcasm(?)]

15. Any chance we can sort this out here, officer?

[bribery(?)]

16. It’d be a shame if something happened to your dog.

[threat(?)]

29 / 42

SLIDE 98

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

SHRDLU (Winograd 1972)

Fully grounded system that parses the user’s input, maps it to logical form, interprets that logical form in its world, tries to take appropriate action, and generates responses. One project did succeed. Terry Winograd’s program SHRDLU could use English intelligently, but there was a catch: the only subject you could discuss was a micro- world of simulated blocks.

30 / 42

SLIDE 99

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

SHRDLU (Winograd 1972)

1. Person: WHAT DOES THE BOX CONTAIN?
2. Computer: THE BLUE PYRAMID AND THE BLUE

BLOCK.

3. Person: WHAT IS THE PYRAMID SUPPORTED BY?
4. Computer: THE BOX.
5. Person: HOW MANY BLOCKS ARE NOT IN THE BOX?
6. Computer: FOUR OF THEM.
7. Person: IS AT LEAST ONE OF THEM NARROWER

THAN THE ONE WHICH I TOLD YOU TO PICK UP?

8. Computer: YES, THE RED CUBE.
9. Person: IS IT SUPPORTED?
10. Computer: YES, BY THE TABLE.

30 / 42

SLIDE 100

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: sarcasm

sarcastic? w h0 h1 h2 h3 Yeah great idea Yeah great idea

31 / 42

SLIDE 101

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: sarcasm

sarcastic? w h0 h1 h2 h3 user context Yeah great idea Yeah great idea SARC (Khodak et al. 2017); Hazarika et al. 2018

32 / 42

SLIDE 102

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Colors in context

Context Utterance xxxx xxxx xxxx blue

T able: Example from the Colors in Context corpus from the Stanford Computation & Cognition Lab

33 / 42

SLIDE 103

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Colors in context

Context Utterance xxxx xxxx xxxx blue xxxx xxxx xxxx The darker blue one

T able: Example from the Colors in Context corpus from the Stanford Computation & Cognition Lab

33 / 42

SLIDE 104

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Colors in context

Context Utterance xxxx xxxx xxxx blue xxxx xxxx xxxx The darker blue one xxxx xxxx xxxx dull pink not the super bright one

T able: Example from the Colors in Context corpus from the Stanford Computation & Cognition Lab

33 / 42

SLIDE 105

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Colors in context

Context Utterance xxxx xxxx xxxx blue xxxx xxxx xxxx The darker blue one xxxx xxxx xxxx dull pink not the super bright one xxxx xxxx xxxx Purple

T able: Example from the Colors in Context corpus from the Stanford Computation & Cognition Lab

33 / 42

SLIDE 106

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Colors in context

Context Utterance xxxx xxxx xxxx blue xxxx xxxx xxxx The darker blue one xxxx xxxx xxxx dull pink not the super bright one xxxx xxxx xxxx Purple xxxx xxxx xxxx blue

T able: Example from the Colors in Context corpus from the Stanford Computation & Cognition Lab

33 / 42

SLIDE 107

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Literal neural speaker S0

c1 c2 cT h h; 〈s〉 h;x1 h;x2 x1 x2 〈/s〉 LSTM Fully connected softmax

34 / 42

SLIDE 108

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Neural literal listener L0

x1 x2 x3 (μ, Σ) c1 c2 c3

c3

Embedding LSTM Softmax

35 / 42

SLIDE 109

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Neural pragmatic agents

36 / 42

SLIDE 110

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Neural pragmatic agents

Neural pragmatic speaker (Andreas & Klein 2016)

S1(msg | c, C; θ) = L0(c | msg, C; θ)

msg′∈X L0(c | msg′, C; θ)

36 / 42

SLIDE 111

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Neural pragmatic agents

Neural pragmatic speaker (Andreas & Klein 2016)

S1(msg | c, C; θ) = L0(c | msg, C; θ)

msg′∈X L0(c | msg′, C; θ)

where X is a sample from S0(msg | c, C; θ) such that msg∗ ∈ X.

36 / 42

SLIDE 112

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Neural pragmatic agents

Neural pragmatic speaker (Andreas & Klein 2016)

S1(msg | c, C; θ) = L0(c | msg, C; θ)

msg′∈X L0(c | msg′, C; θ)

where X is a sample from S0(msg | c, C; θ) such that msg∗ ∈ X.

Neural pragmatic listener

L1(c | msg, C; θ) ∝ S1(msg | c, C; θ)

36 / 42

SLIDE 113

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Pragmatic image captioning

Mao et al. (2016); Vedantam et al. (2017): Captions that are true and distinguish their images from related ones. Reasoning about all possible utterances/captions? (Cohn-Gordon et al. 2018)

37 / 42

SLIDE 114

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Pragmatic image captioning

Mao et al. (2016); Vedantam et al. (2017): Captions that are true and distinguish their images from related ones. Reasoning about all possible utterances/captions? ⇒ Sample from S0 (Cohn-Gordon et al. 2018)

37 / 42

SLIDE 115

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Example: Pragmatic image captioning

Mao et al. (2016); Vedantam et al. (2017): Captions that are true and distinguish their images from related ones. Reasoning about all possible utterances/captions? ⇒ Full pragmatic reasoning about characters (Cohn-Gordon et al. 2018)

37 / 42

SLIDE 116

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

Some pragmatic generalizations

1. Scalar implicature: general terms tend to signal

that their more specific alternatives are pragmatically marked.

2. I-implicature: if a general term has prototypical

instantiations in context, then it might be refined to pick out just those prototypes.

3. Manner implicature: unusual events are described

with unusual language; normal events with normal language.

4. Metaphor: metaphorical language is pervasive and

enables the speaker to highlight specific dimensions of meaning efficiently.

5. Contextual refinement: word and phrase meanings

are flexible and respond to the social context.

38 / 42

SLIDE 117

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The next frontier

39 / 42

SLIDE 118

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The next frontier

The Human Speechome Project (Roy et al. 2006)

39 / 42

SLIDE 119

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The next frontier

The Human Speechome Project (Roy et al. 2006) And, not to play into stereotypes of linguists, but some symbolic reasoning would be useful!

39 / 42

SLIDE 120

Signs of the apocalypse? Lexical semantics Compositional semantics Pragmatics The next frontier

The next frontier

The Human Speechome Project (Roy et al. 2006) And, not to play into stereotypes of linguists, but some symbolic reasoning would be useful!

Thanks!

39 / 42

SLIDE 121

References

References I

Andreas, Jacob & Dan Klein. 2016. Reasoning about pragmatics with neural listeners and speakers. In Proceedings of the 2016 conference on empirical methods in natural language processing, 1173–1182. Association for Computational Linguistics. http://aclweb.org/anthology/D16-1125. Bordes, Antoine, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston & Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data. In Advances in neural information processing systems, 2787–2795. Bowman, Samuel R. 2017. Modeling natural language semantics in learned representations. Stanford, CA: Stanford University dissertation. Clark, Stephen, Bob Coecke & Mehrnoosh Sadrzadeh. 2011. Mathematical foundations for a compositional distributed model of meaning. Linguistic Analysis 36(1–4). 345–384. Cohn-Gordon, Reuben, Noah D. Goodman & Christopher Potts. 2018. Pragmatically informative image captioning with character-level inference. In Human language technologies: The 16th annual conference of the north american chapter of the Association for Computational Linguistics, Stroudsburg, PA: Association for Computational Linguistics. Faruqui, Manaal, Jesse Dodge, Sujay Kumar Jauhar, Chris Dyer, Eduard Hovy & Noah A. Smith. 2015. Retrofitting word vectors to semantic lexicons. In Proceedings of the 2015 conference of the north american chapter of the association for computational linguistics: Human language technologies, 1606–1615. Stroudsburg, PA: Association for Computational Linguistics. http://www.aclweb.org/anthology/N15-1184. Frank, Michael C. & Noah D. Goodman. 2014. Inferring word meanings by assuming that speakers are

informative. Cognitive Psychology 75(1). 80–96. doi:doi:10.1016/j.cogpsych.2014.08.002.

Frank, Michael C., Joshua B. T enenbaum & Anne Fernald. 2012. Social and discourse contributions to the determination of reference in cross-situational word learning. Language, Learning, and Development . Glass, Lelia. 2018. Deriving the distributivity potential of adjectives via measurement theory. Proceedings of the Linguistic Society of America 3(49). 1–14. doi:10.3765/plsa.v3i1.4343. Hamilton, William L., Rex Ying & Jure Leskovec. 2017. Representation learning on graphs: Methods and applications. In Ieee data engineering bulletin, 52–74. IEEE Press. 40 / 42

SLIDE 122

References

References II

Hazarika, Devamanyu, Soujanya Poria, Sruthi Gorantla, Erik Cambria, Roger Zimmermann & Rada

Mihalcea. 2018. CASCADE: Contextual sarcasm detection in online discussion forums.

ArXiv:1805.06413. Katz, Jerrold J. 1972. Semantic theory. New York: Harper & Row. Khodak, Mikhail, Nikunj Saunshi & Kiran Vodrahalli. 2017. A large self-annotated corpus for sarcasm. arXiv preprint arXiv:1704.05579 . Lengerich, Benjamin J., Andrew L. Maas & Christopher Potts. 2018. Retrofitting distributional embeddings to knowledge graphs with functional relations. In Proceedings of the 27th international conference on computational linguistics (COLING 2018), The COLING 2018 Organizing Committee. ArXiv:1708.00112. Lin, Yankai, Zhiyuan Liu, Maosong Sun, Yang Liu & Xuan Zhu. 2015. Learning entity and relation embeddings for knowledge graph completion. In Aaai, 2181–2187. MacCartney, Bill & Christopher D. Manning. 2009. An extended model of natural logic. In Proceedings

f the eighth international conference on computational semantics, 140–156. Tilburg, The

Netherlands: Association for Computational Linguistics. http://www.aclweb.org/anthology/W09-3714. Manning, Christopher D. 2015. Computational linguistics and deep learning. Computational Linguistics 41(4). doi:10.1162/COLIa00239. Mao, Junhua, Jonathan Huang, Alexander T

shev, Oana Camburu, Alan L. Yuille & Kevin Murphy. 2016.

Generation and comprehension of unambiguous object descriptions. In Proceedings of the ieee conference on computer vision and pattern recognition, 11–20. IEEE. Mrkˇ si´ c, Nikola, Diarmuid Ó Séaghdha, Blaise Thomson, Milica Gaˇ si´ c, Lina M. Rojas-Barahona, Pei-Hao Su, David Vandyke, T sung-Hsien Wen & Steve Young. 2016. Counter-fitting word vectors to linguistic constraints. In Proceedings of the 2016 conference of the north american chapter of the association for computational linguistics: Human language technologies, 142–148. Association for Computational Linguistics. doi:10.18653/v1/N16-1018. http://aclanthology.coli.uni-saarland.de/pdf/N/N16/N16-1018.pdf. 41 / 42

SLIDE 123

References

References III

Nickel, Maximilian, Volker Tresp & Hans-Peter Kriegel. 2011. A three-way model for collective learning

n multi-relational data. In Proceedings of the 28th international conference on machine learning

(icml-11), 809–816. ACM. Partee, Barbara H. 1984. Compositionality. In Fred Landman & Frank Veltman (eds.), Varieties of formal semantics, 281–311. Dordrecht: Foris. Reprinted in Barbara H. Partee (2004) Compositionality in formal semantics, Oxford: Blackwell 153–181. Page references to the reprinting. Rappaport Hovav, Malka & Beth Levin. 2010. Reflections on manner/result complementarity. In Malka Rappaport Hovav, Edit Doron & Ivy Sichel (eds.), Syntax, lexical semantics, and event structure, 21–38. Oxford University Press. doi:10.1093/acprof:oso/9780199544325.003.0002. Roy, Deb, Rupal Patel, Philip DeCamp, Rony Kubat, Michael Fleischman, Brandon Roy, Nikolaos Mavridis, Stefanie T ellex, Alexia Salata, Jethran Guinness et al. 2006. The human speechome

project. In Symbol grounding and beyond, 192–196. Springer.

Socher, Richard, Danqi Chen, Christopher D Manning & Andrew Ng. 2013. Reasoning with neural tensor networks for knowledge base completion. In Advances in neural information processing systems, 926–934. Sutskever, Ilya, Joshua B T enenbaum & Ruslan R Salakhutdinov. 2009. Modelling relational data using bayesian clustered tensor factorization. In Advances in neural information processing systems, 1821–1828. Thomason, Richmond H. 1974. Introduction. In Formal philosophy: Selected papers of Richard Montague, 1–69. New Haven, CT: Yale University Press. Vedantam, Ramakrishna, Samy Bengio, Kevin Murphy, Devi Parikh & Gal Chechik. 2017. Context-aware captions from context-agnostic supervision. arXiv:1701.02870 . Wang, Zhen, Jianwen Zhang, Jianlin Feng & Zheng Chen. 2014. Knowledge graph embedding by translating on hyperplanes. In Twenty-eighth aaai conference on artificial intelligence, . Winograd, T

erry. 1972. Understanding natural language. Cognitive Psychology 3(1). 1–191.

42 / 42