Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations - - PowerPoint PPT Presentation

▶
paraphrase to explicate
SMART_READER_LITE
LIVE PREVIEW

Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations - - PowerPoint PPT Presentation

Jun 23, 2023 454 likes •1.23k views

Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations Vered Shwartz and Ido Dagan Natural Language Processing Lab, Bar-Ilan University July 17, 2018 Noun Compounds Two or more nouns function as a unit to create a new concept hot

slide-1
SLIDE 1

Paraphrase to Explicate:

Revealing Implicit Noun-Compound Relations

Vered Shwartz and Ido Dagan

Natural Language Processing Lab, Bar-Ilan University July 17, 2018

slide-2
SLIDE 2

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package...

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-3
SLIDE 3

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-4
SLIDE 4

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Express implicit relationship between the constituent nouns:

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-5
SLIDE 5

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Express implicit relationship between the constituent nouns:

apple cake: cake made of apples

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-6
SLIDE 6

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Express implicit relationship between the constituent nouns:

apple cake: cake made of apples birthday cake: cake eaten on a birthday

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-7
SLIDE 7

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Express implicit relationship between the constituent nouns:

apple cake: cake made of apples birthday cake: cake eaten on a birthday

They are like “text compression devices” [Nakov, 2013]

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-8
SLIDE 8

Noun Compounds

Two or more nouns function as a unit to create a new concept

hot dog, hot dog bun, hot dog bun package... We focus on two-word compounds

Express implicit relationship between the constituent nouns:

apple cake: cake made of apples birthday cake: cake eaten on a birthday

They are like “text compression devices” [Nakov, 2013] We’re pretty good at decompressing them!

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 2 / 23

slide-9
SLIDE 9

We are good at Interpreting Noun-Compounds

We easily interpret noun-compounds

Even when we see them for the first time

1 Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 3 / 23

slide-10
SLIDE 10

We are good at Interpreting Noun-Compounds

We easily interpret noun-compounds

Even when we see them for the first time

What is a “parsley cake”?

1 Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 3 / 23

slide-11
SLIDE 11

We are good at Interpreting Noun-Compounds

We easily interpret noun-compounds

Even when we see them for the first time

What is a “parsley cake”?

cake eaten on a parsley? cake with parsley? cake for parsley? ...

1 Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 3 / 23

slide-12
SLIDE 12

We are good at Interpreting Noun-Compounds

We easily interpret noun-compounds

Even when we see them for the first time

What is a “parsley cake”?

cake eaten on a parsley? cake with parsley? cake for parsley? ...

1

1from http://www.bazekalim.com Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 3 / 23

slide-13
SLIDE 13

Generalizing Existing Knowledge

What can cake be made of?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 4 / 23

slide-14
SLIDE 14

Generalizing Existing Knowledge

What can cake be made of? Parsley (sort of) fits into this distribution

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 4 / 23

slide-15
SLIDE 15

Generalizing Existing Knowledge

What can cake be made of? Parsley (sort of) fits into this distribution Similar to “selectional preferences” [Pantel et al., 2007]

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 4 / 23

slide-16
SLIDE 16

We need Computers to Interpret Noun-Compounds

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 5 / 23

slide-17
SLIDE 17

Noun-Compound Interpretation Tasks

Bracketing [[pumpkin spice] latte]

slide-18
SLIDE 18

Noun-Compound Interpretation Tasks

Bracketing [[pumpkin spice] latte] Compositionality Prediction is spelling bee related to bee?

slide-19
SLIDE 19

Noun-Compound Interpretation Tasks

Bracketing [[pumpkin spice] latte] Compositionality Prediction is spelling bee related to bee? Relation Classification apple cake → ingredient birthday cake → time

slide-20
SLIDE 20

Noun-Compound Interpretation Tasks

Bracketing [[pumpkin spice] latte] Compositionality Prediction is spelling bee related to bee? Relation Classification apple cake → ingredient birthday cake → time Paraphrasing cake made of apples cake eaten on a birthday

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 6 / 23

slide-21
SLIDE 21

Noun-Compound Paraphrasing

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 7 / 23

slide-22
SLIDE 22

Motivation

Given a noun-compound w1w2, express the relation between the head w2 and the modifier w1 with multiple prepositional and verbal paraphrases [Nakov and Hearst, 2006]

  • live oil

apple cake ground attack [w2] extracted from [w1] [w2] made of [w1] [w2] from [w1] boat whistle sea bass [w2] located in [w1] [w2] live in [w1] game room service door baby oil [w2] used for [w1] [w2] for [w1]

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 8 / 23

slide-23
SLIDE 23

Evaluation Setting

Available dataset: SemEval 2013 task 4 [Hendrickx et al., 2013]

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 9 / 23

slide-24
SLIDE 24

Evaluation Setting

Available dataset: SemEval 2013 task 4 [Hendrickx et al., 2013] A ranking rather than a retrieval task

Systems get a list of noun compounds

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 9 / 23

slide-25
SLIDE 25

Evaluation Setting

Available dataset: SemEval 2013 task 4 [Hendrickx et al., 2013] A ranking rather than a retrieval task

Systems get a list of noun compounds Extract paraphrases from free text

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 9 / 23

slide-26
SLIDE 26

Evaluation Setting

Available dataset: SemEval 2013 task 4 [Hendrickx et al., 2013] A ranking rather than a retrieval task

Systems get a list of noun compounds Extract paraphrases from free text Rank them

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 9 / 23

slide-27
SLIDE 27

Evaluation Setting

Available dataset: SemEval 2013 task 4 [Hendrickx et al., 2013] A ranking rather than a retrieval task

Systems get a list of noun compounds Extract paraphrases from free text Rank them

Evaluated for correlation with human judgments

Gold paraphrase score: how many annotators suggested it?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 9 / 23

slide-28
SLIDE 28

Prior Methods (1/2)

Based on constituent co-occurrences: “cake made of apple”

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 10 / 23

slide-29
SLIDE 29

Prior Methods (1/2)

Based on constituent co-occurrences: “cake made of apple” Problems:

  • 1. Many unseen compounds, no paraphrases in the corpus

rare: parsley cake or highly lexicalized: ice cream

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 10 / 23

slide-30
SLIDE 30

Prior Methods (1/2)

Based on constituent co-occurrences: “cake made of apple” Problems:

  • 1. Many unseen compounds, no paraphrases in the corpus

rare: parsley cake or highly lexicalized: ice cream

  • 2. Many compounds with just a few paraphrases

Can we infer “cake containing apple” given “cake made of apple”?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 10 / 23

slide-31
SLIDE 31

Prior Methods (1/2)

Based on constituent co-occurrences: “cake made of apple” Problems:

  • 1. Many unseen compounds, no paraphrases in the corpus

rare: parsley cake or highly lexicalized: ice cream

  • 2. Many compounds with just a few paraphrases

Can we infer “cake containing apple” given “cake made of apple”?

Prior work provides partial solutions to either (1) or (2)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 10 / 23

slide-32
SLIDE 32

Prior Methods (2/2)

  • 1. MELODI [Van de Cruys et al., 2013]:

Represent NC by applying a function to its constituent distributional vectors: vec(apple cake) = f (vec(apple), vec(cake))

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 11 / 23

slide-33
SLIDE 33

Prior Methods (2/2)

  • 1. MELODI [Van de Cruys et al., 2013]:

Represent NC by applying a function to its constituent distributional vectors: vec(apple cake) = f (vec(apple), vec(cake)) Predict paraphrase templates given NC vector

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 11 / 23

slide-34
SLIDE 34

Prior Methods (2/2)

  • 1. MELODI [Van de Cruys et al., 2013]:

Represent NC by applying a function to its constituent distributional vectors: vec(apple cake) = f (vec(apple), vec(cake)) Predict paraphrase templates given NC vector Generalizes for similar unseen NCs, e.g. pear tart

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 11 / 23

slide-35
SLIDE 35

Prior Methods (2/2)

  • 1. MELODI [Van de Cruys et al., 2013]:

Represent NC by applying a function to its constituent distributional vectors: vec(apple cake) = f (vec(apple), vec(cake)) Predict paraphrase templates given NC vector Generalizes for similar unseen NCs, e.g. pear tart

  • 2. IIITH [Surtani et al., 2013]:

Learn “is-a” relations between paraphrases: e.g. “[w2] extracted from [w1]” ⊂ “[w2] made of [w1]”

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 11 / 23

slide-36
SLIDE 36

Prior Methods (2/2)

  • 1. MELODI [Van de Cruys et al., 2013]:

Represent NC by applying a function to its constituent distributional vectors: vec(apple cake) = f (vec(apple), vec(cake)) Predict paraphrase templates given NC vector Generalizes for similar unseen NCs, e.g. pear tart

  • 2. IIITH [Surtani et al., 2013]:

Learn “is-a” relations between paraphrases: e.g. “[w2] extracted from [w1]” ⊂ “[w2] made of [w1]”

Our solution: multi-task learning to address both problems

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 11 / 23

slide-37
SLIDE 37

Model

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 12 / 23

slide-38
SLIDE 38

Multi-task Reformulation

Training example {w1 = apple, w2 = cake, p = “[w2] made of [w1]”}

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 13 / 23

slide-39
SLIDE 39

Multi-task Reformulation

Training example {w1 = apple, w2 = cake, p = “[w2] made of [w1]”}

  • 1. Predict a paraphrase p for a given NC w1w2:

What is the relation between apple and cake?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 13 / 23

slide-40
SLIDE 40

Multi-task Reformulation

Training example {w1 = apple, w2 = cake, p = “[w2] made of [w1]”}

  • 1. Predict a paraphrase p for a given NC w1w2:

What is the relation between apple and cake?

  • 2. Predict w1 given a paraphrase p and w2:

What can cake be made of?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 13 / 23

slide-41
SLIDE 41

Multi-task Reformulation

Training example {w1 = apple, w2 = cake, p = “[w2] made of [w1]”}

  • 1. Predict a paraphrase p for a given NC w1w2:

What is the relation between apple and cake?

  • 2. Predict w1 given a paraphrase p and w2:

What can cake be made of?

  • 3. Predict w2 given a paraphrase p and w1:

What can be made of apple?

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 13 / 23

slide-42
SLIDE 42

Main Task (1): Predicting Paraphrases

What is the relation between apple and cake?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p] (78) [w2] containing [w1] ... (131) [w2] made of [w1] ... [p] cake apple MLPp ˆ pi = 78

Encode placeholder [p] in “cake [p] apple” using biLSTM

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 14 / 23

slide-43
SLIDE 43

Main Task (1): Predicting Paraphrases

What is the relation between apple and cake?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p] (78) [w2] containing [w1] ... (131) [w2] made of [w1] ... [p] cake apple MLPp ˆ pi = 78

Encode placeholder [p] in “cake [p] apple” using biLSTM Predict an index in the paraphrase vocabulary

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 14 / 23

slide-44
SLIDE 44

Main Task (1): Predicting Paraphrases

What is the relation between apple and cake?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p] (78) [w2] containing [w1] ... (131) [w2] made of [w1] ... [p] cake apple MLPp ˆ pi = 78

Encode placeholder [p] in “cake [p] apple” using biLSTM Predict an index in the paraphrase vocabulary Fixed word embeddings, learned placeholder embeddings

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 14 / 23

slide-45
SLIDE 45

Main Task (1): Predicting Paraphrases

What is the relation between apple and cake?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p] (78) [w2] containing [w1] ... (131) [w2] made of [w1] ... [p] cake apple MLPp ˆ pi = 78

Encode placeholder [p] in “cake [p] apple” using biLSTM Predict an index in the paraphrase vocabulary Fixed word embeddings, learned placeholder embeddings (1) Generalizes NCs: pear tart expected to yield similar results

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 14 / 23

slide-46
SLIDE 46

Helper Task (2): Predicting Missing Constituents

What can cake be made of?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p]

  • f

cake made [w1] MLPw ˆ w1i = 28

Encode placeholder in “cake made of [w1]” using biLSTM

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 15 / 23

slide-47
SLIDE 47

Helper Task (2): Predicting Missing Constituents

What can cake be made of?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p]

  • f

cake made [w1] MLPw ˆ w1i = 28

Encode placeholder in “cake made of [w1]” using biLSTM Predict an index in the word vocabulary

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 15 / 23

slide-48
SLIDE 48

Helper Task (2): Predicting Missing Constituents

What can cake be made of?

(23) made (28) apple (4145) cake ... (7891) of (1) [w1] (2) [w2] (3) [p]

  • f

cake made [w1] MLPw ˆ w1i = 28

Encode placeholder in “cake made of [w1]” using biLSTM Predict an index in the word vocabulary (2) Generalizes paraphrases:

“[w2] containing [w1]” expected to yield similar results

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 15 / 23

slide-49
SLIDE 49

Training Data

Collected from Google N-grams

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 16 / 23

slide-50
SLIDE 50

Training Data

Collected from Google N-grams Input:

Set of NCs Templates of POS tags (e.g. “[w2] verb prep [w1]”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 16 / 23

slide-51
SLIDE 51

Training Data

Collected from Google N-grams Input:

Set of NCs Templates of POS tags (e.g. “[w2] verb prep [w1]”)

Weighting by frequency and length

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 16 / 23

slide-52
SLIDE 52

Training Data

Collected from Google N-grams Input:

Set of NCs Templates of POS tags (e.g. “[w2] verb prep [w1]”)

Weighting by frequency and length 140k instances

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 16 / 23

slide-53
SLIDE 53

Evaluation

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 17 / 23

slide-54
SLIDE 54

Ranking Model

Predict top k paraphrases for each noun compound

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 18 / 23

slide-55
SLIDE 55

Ranking Model

Predict top k paraphrases for each noun compound Learn to re-rank the paraphrases

to better correlate with human judgments

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 18 / 23

slide-56
SLIDE 56

Ranking Model

Predict top k paraphrases for each noun compound Learn to re-rank the paraphrases

to better correlate with human judgments

SVM pair-wise ranking with the following features:

POS tags in the paraphrase Prepositions in the paraphrase Length Special symbols Similarity to predicted paraphrase

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 18 / 23

slide-57
SLIDE 57

Results

non-isomorphic isomorphic 20 40 60

54.8 13 40.6 13.8 17.9 23.1 23.1 25.8 28.4 28.2 MELODI [Van de Cruys et al., 2013] SemEval 2013 Baseline [Hendrickx et al., 2013] SFS [Versley, 2013] IIITH [Surtani et al., 2013] PaNiC [Shwartz and Dagan, 2018]

slide-58
SLIDE 58

Results

non-isomorphic isomorphic 20 40 60

54.8 13 40.6 13.8 17.9 23.1 23.1 25.8 28.4 28.2 MELODI [Van de Cruys et al., 2013] SemEval 2013 Baseline [Hendrickx et al., 2013] SFS [Versley, 2013] IIITH [Surtani et al., 2013] PaNiC [Shwartz and Dagan, 2018] rewards recall and precision

slide-59
SLIDE 59

Results

non-isomorphic isomorphic 20 40 60

54.8 13 40.6 13.8 17.9 23.1 23.1 25.8 28.4 28.2 MELODI [Van de Cruys et al., 2013] SemEval 2013 Baseline [Hendrickx et al., 2013] SFS [Versley, 2013] IIITH [Surtani et al., 2013] PaNiC [Shwartz and Dagan, 2018] rewards recall and precision rewards

  • nly precision
slide-60
SLIDE 60

Results

non-isomorphic isomorphic 20 40 60

54.8 13 40.6 13.8 17.9 23.1 23.1 25.8 28.4 28.2 MELODI [Van de Cruys et al., 2013] SemEval 2013 Baseline [Hendrickx et al., 2013] SFS [Versley, 2013] IIITH [Surtani et al., 2013] PaNiC [Shwartz and Dagan, 2018] rewards recall and precision rewards

  • nly precision

“conservative” models Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 19 / 23

slide-61
SLIDE 61

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-62
SLIDE 62

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-63
SLIDE 63

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

  • 3. Incorrect prepositions

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-64
SLIDE 64

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

  • 3. Incorrect prepositions

E.g., n-grams don’t respect syntactic structure: “rinse away the oil from baby ’s head” ⇒ “oil from baby”

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-65
SLIDE 65

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

  • 3. Incorrect prepositions

E.g., n-grams don’t respect syntactic structure: “rinse away the oil from baby ’s head” ⇒ “oil from baby”

  • 4. Syntactic errors

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-66
SLIDE 66

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

  • 3. Incorrect prepositions

E.g., n-grams don’t respect syntactic structure: “rinse away the oil from baby ’s head” ⇒ “oil from baby”

  • 4. Syntactic errors
  • 5. Borderline grammatical

(“force of coalition forces”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-67
SLIDE 67

Error Analysis

False Positive

(1) 44% (2) 15% (3) 14% (4) 8% (5) 5% (6) 14%

  • 1. Valid, missing from gold-standard

(“discussion by group”)

  • 2. Too specific

(“life of women in community”)

  • 3. Incorrect prepositions

E.g., n-grams don’t respect syntactic structure: “rinse away the oil from baby ’s head” ⇒ “oil from baby”

  • 4. Syntactic errors
  • 5. Borderline grammatical

(“force of coalition forces”)

  • 6. Other errors

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 20 / 23

slide-68
SLIDE 68

Error Analysis

False Negative

(1) 30% (2) 25% (3) 10% (4) 35%

  • 1. Long paraphrase (n > 5)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 21 / 23

slide-69
SLIDE 69

Error Analysis

False Negative

(1) 30% (2) 25% (3) 10% (4) 35%

  • 1. Long paraphrase (n > 5)
  • 2. Determiners

(“mutation of a gene”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 21 / 23

slide-70
SLIDE 70

Error Analysis

False Negative

(1) 30% (2) 25% (3) 10% (4) 35%

  • 1. Long paraphrase (n > 5)
  • 2. Determiners

(“mutation of a gene”)

  • 3. Inflected constituents

(“holding of shares”)

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 21 / 23

slide-71
SLIDE 71

Error Analysis

False Negative

(1) 30% (2) 25% (3) 10% (4) 35%

  • 1. Long paraphrase (n > 5)
  • 2. Determiners

(“mutation of a gene”)

  • 3. Inflected constituents

(“holding of shares”)

  • 4. Other errors

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 21 / 23

slide-72
SLIDE 72

Recap

A model for generating paraphrases for given noun-compounds

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 22 / 23

slide-73
SLIDE 73

Recap

A model for generating paraphrases for given noun-compounds Better generalization abilities:

Generalize for unseen noun-compounds Embed semantically-similar paraphrases in proximity

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 22 / 23

slide-74
SLIDE 74

Recap

A model for generating paraphrases for given noun-compounds Better generalization abilities:

Generalize for unseen noun-compounds Embed semantically-similar paraphrases in proximity

Improved performance in challenging evaluation settings

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 22 / 23

slide-75
SLIDE 75

Recap

A model for generating paraphrases for given noun-compounds Better generalization abilities:

Generalize for unseen noun-compounds Embed semantically-similar paraphrases in proximity

Improved performance in challenging evaluation settings

Thank you!

Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 22 / 23

slide-76
SLIDE 76

References

[Hendrickx et al., 2013] Hendrickx, I., Kozareva, Z., Nakov, P., Ó Séaghdha, D., Szpakowicz, S., and Veale, T. (2013). Semeval-2013 task 4: Free paraphrases of noun compounds. In SemEval, pages 138–143. [Nakov, 2013] Nakov, P. (2013). On the interpretation of noun compounds: Syntax, semantics, and entailment. Natural Language Engineering, 19(03):291–330. [Nakov and Hearst, 2006] Nakov, P. and Hearst, M. (2006). Using verbs to characterize noun-noun relations. In International Conference on Artificial Intelligence: Methodology, Systems, and Applications, pages 233–244. Springer. [Pantel et al., 2007] Pantel, P., Bhagat, R., Coppola, B., Chklovski, T., and Hovy, E. (2007). ISP: Learning inferential selectional

  • preferences. In ACL, pages 564–571.

[Shwartz and Dagan, 2018] Shwartz, V. and Dagan, I. (2018). Paraphrase to explicate: Revealing implicit noun-compound

  • relations. In ACL, Melbourne, Australia.

[Surtani et al., 2013] Surtani, N., Batra, A., Ghosh, U., and Paul, S. (2013). Iiit-h: A corpus-driven co-occurrence based probabilistic model for noun compound paraphrasing. In SemEval, pages 153–157. [Van de Cruys et al., 2013] Van de Cruys, T., Afantenos, S., and Muller, P. (2013). Melodi: A supervised distributional approach for free paraphrasing of noun compounds. In SemEval, pages 144–147. [Versley, 2013] Versley, Y. (2013). Sfs-tue: Compound paraphrasing with a language model and discriminative reranking. In SemEval, pages 148–152. Vered Shwartz and Ido Dagan · Paraphrase to Explicate: Revealing Implicit Noun-Compound Relations · ACL 2018 23 / 23