natural language learning with formal semantics · 2. formal semantics (montague) - we construct...
TRANSCRIPT
![Page 1: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/1.jpg)
Natural Language Learning with Formal Semantics
Wenhu Chen
1
![Page 2: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/2.jpg)
Outline1. Introduction2. Formal Semantics3. Its Application in NLP4. My Work & Future Direction
2
![Page 3: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/3.jpg)
Meaning
3
Meaning
![Page 4: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/4.jpg)
Meaning
4
Meaning
Semantics PragmaticsCognitive Social Science
![Page 5: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/5.jpg)
Meaning
5
Linguists can barely reach an agreement on the definition of “meaning”
![Page 6: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/6.jpg)
Meaning
6
1. Simple2. Straightforward
Raw-String
Bag of wordsUnigram/Bigram
![Page 7: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/7.jpg)
Meaning
7
1. Simple2. Straightforward
Raw-String
Bag of wordsUnigram/Bigram
Formal Semantics
1. Compositional2. Explainable
Lambda CalculusMontague Semantics
![Page 8: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/8.jpg)
Meaning
8
1. Simple2. Straightforward
Raw-String Formal Semantics
1. Compositional2. Explainable
Real-valued Vector
1. Data-Drive2. Computation
Efficient
Word2Vec/GloveBERT
Bag of wordsUnigram/Bigram
Lambda CalculusMontague Semantics
![Page 9: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/9.jpg)
Meaning
9
What aspects do you need the meaning representation to capture?
![Page 10: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/10.jpg)
Meaning
10
What aspects do you need the meaning representation to capture?
Can it lean meaning or just similarity between meaning?
Can it lean grounded meaning or learn lexical meaning?
![Page 11: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/11.jpg)
1. Raw-String Representation- Cannot directly link to the external world knowledge.- Cannot combine pieces of knowledge together to perform inference.
11
Utterance1: John has an appleUtterance2: John has a banana
Utterance: John has an apple and a banana.
![Page 12: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/12.jpg)
1. Raw-String Representation- Assuming we want to build an NLP system to understand math questions?
a. Such system cannot leverage the mathematical knowledge.b. The sample complexity becomes extremely high.
12
MathWorld
Utterance: What is the largest prime less than 10?
![Page 13: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/13.jpg)
2. Formal Semantics (Montague)- We construct the meaning of natural language in terms of variables,
predicates, arguments.
13
![Page 14: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/14.jpg)
2. Formal Semantics (Montague)- We construct the meaning of natural language in terms of variables,
predicates, arguments.- Formal Semantics aim to use explicit and grounded meaning representation
to convey information.
14
![Page 15: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/15.jpg)
2. Formal Semantics (Montague)- We construct the meaning of natural language in terms of variables,
predicates, arguments.- Formal Semantics aim to use explicit and grounded meaning representation
to convey information.- The formal semantics can help us understand human language from a
mathematical logic perspective.
15
![Page 16: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/16.jpg)
2. Formal Semantics (Montague)
16
Utterance: What is the largest prime less than 10?
MathWorld
![Page 17: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/17.jpg)
3. Vector Representation- Data-driven learning process- Suitable for deep learning framework
17
![Page 18: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/18.jpg)
3. Vector Representation- Data-driven learning process- Suitable for deep learning framework
18
![Page 19: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/19.jpg)
3. Vector Representation
19
Utterance Behavior
We don’t need to worry about the structure of computation in the middle
![Page 20: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/20.jpg)
3. Vector Representation
20
Tight coupling between machine learning and representation means there’s always risk that some new semantic phenomenon arises and suddenly our model is useless
Utterance Behavior
![Page 21: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/21.jpg)
Evaluation Chart
21
Type Meaning Similarity Meaning Grounded Meaning Lexical Meaning
String
Vector
Formal
![Page 22: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/22.jpg)
Evaluation Chart
22
Type Meaning Similarity Meaning Grounded Meaning Lexical Meaning
String
Vector
Formal
![Page 23: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/23.jpg)
Evaluation Chart
23
Type Meaning Similarity Meaning Grounded Meaning Lexical Meaning
String
Vector
Formal
![Page 24: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/24.jpg)
Formal Semantics- For a long time, such an explicit meaning representation has dominated the
research of NLP community- Reasoning- Perception- Action
24
![Page 25: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/25.jpg)
Formal Semantics- For a long time, such an explicit meaning representation has dominated the
research of NLP community- Reasoning- Perception- Action
25
Utterance Parser
![Page 26: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/26.jpg)
Formal Semantics- For a long time, such an explicit meaning representation has dominated the
research of NLP community- Reasoning- Perception- Action
26
Utterance Parser
Semantic Form1
Semantic Form2
Semantic Form3
QuestionAnswering
LanguageNavigation
Textual Inference
![Page 27: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/27.jpg)
List of Semantic Forms- Introduction of formal semantics
- Lambda Calculus (Turing 1937)- Combinatory categorial grammar (CCG) (Steedman et al. 1996, 2000)- Weighted linear CCG (Clark & Currant et. al 2007)- Dependency-based compositional semantics (DCS) (Liang et al. 2011)- Lambda DCS (Liang et al. 2013)- Abstract Meaning Representation (Banarescu et al. 2013)- Domain Specific Languages (DSL)
- Functional program semantics (Liang et al. 2017, Dawn et al. 2018)- SQL, etc (Zhong et al. 2017, Yu et al. 2018)
27
![Page 28: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/28.jpg)
Semantic Forms
28
Yoav et. al, Learning Compact Lexicons for CCG Semantic Parsing, 2014Yoav et. al, Weakly Supervised Learning of Semantic Parsers for Mapping Instructions to Actions, 2013Luke et. al, Learning to Map Sentences to Logical Form: Structured Classification with Probabilistic Categorial Grammars, 2005Steedman et. al, Surface Structure and Interpretation. 1996Steedman et. al, The Syntactic Process. 2000
- Combinatory Categorial Grammar- Definition- Rules- Learning
- Dependency-based compositional semantics- Definition- Rules- Learning
- Neural Symbolic Machine (Functional-Program)- Definition- Learning
![Page 29: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/29.jpg)
Combinatory Categorial Grammar- Categorial Formalism
- Compositional- Puts information on the words
29
![Page 30: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/30.jpg)
Combinatory Categorial Grammar- Categorial Formalism
- Compositional- Puts information on the words
- Transparent interface between syntax and semantics
30
![Page 31: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/31.jpg)
Combinatory Categorial Grammar- Categorial Formalism
- Compositional- Puts information on the words
- Transparent interface between syntax and semantics- Including Predicate-argument structure, quantification and information
structure.
31
![Page 32: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/32.jpg)
Combinatory Categorial Grammar- Categorial Formalism
- Compositional- Puts information on the words
- Transparent interface between syntax and semantics- Including Predicate-argument structure, quantification and information
structure.- Same expressive power as lambda calculus.
32
![Page 33: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/33.jpg)
- Basic Building Block- Capture Syntactic and Semantic Information Jointly
CCG Categories
33
ADJ :ADJ
![Page 34: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/34.jpg)
- Basic Building Block- Capture Syntactic and Semantic Information Jointly
CCG Categories
34
Syntax ADJ Semantic:
![Page 35: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/35.jpg)
- Primitive symbols: N, S, NP, ADJ and PP- Syntactic combination operation (/,\)- Slashes specify argument order
CCG Categories
35
Syntax ADJ Semantic:
Syntax (S\NP)/ADJ Semantic:
Syntax NP SemanticS:
![Page 36: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/36.jpg)
- Pair words and phrases with meaning- Meaning captured by a CCG category
CCG Lexical Entries
36
Natural Language
fun
CCG Category
|- ADJ:
![Page 37: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/37.jpg)
- Pair words and phrases with meaning- Meaning captured by a CCG category
CCG Lexical Entries
37
fun ADJ:
(S\NP)/ADJ:
CCG NP: CCG
fun
is
|-
|-
|-
![Page 38: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/38.jpg)
- Equivalent to function application
- Two direction of application
CCG Operations
38
![Page 39: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/39.jpg)
- Equivalent to function application
- Two direction of application
CCG Operations
39
Argument Function Result
![Page 40: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/40.jpg)
- Use Lexicon to match words and phrases with their CCG categories
- Combine categories using operators
CCG Parsing
40
![Page 41: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/41.jpg)
- Use Lexicon to match words and phrases with their CCG categories
- Combine categories using operators
CCG Parsing
41
![Page 42: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/42.jpg)
- Use Lexicon to match words and phrases with their CCG categories
- Combine categories using operators
CCG Parsing
42
![Page 43: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/43.jpg)
Lexicon Problem- Key component of CCG- Same words often paired with many different categories- Difficult to learn from limited data
43
![Page 44: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/44.jpg)
Lexicon Problem- Key component of CCG- Same words often paired with many different categories- Difficult to learn from limited data
44
The house dog
The dog of the house
house |- ADJ:
house |- N:
![Page 45: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/45.jpg)
Factored Lexicons
45
house |- N:
house |- ADJ:
garden |- N:
Similar
![Page 46: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/46.jpg)
Factored Lexicons
46
house |- N:
house |- ADJ:
garden |- N:
Lexemes
(garden, {garden})(house, {house})
Templates
![Page 47: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/47.jpg)
Factored Lexicons
47
Compress
![Page 48: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/48.jpg)
Factored Lexicons- Capture systematic variations in word usage- Each variation can then be applied to compact units in lexical matching- Abstracts the compositional nature of the word
48
![Page 49: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/49.jpg)
Learning (Data)- (Natural Language, Lambda Form)
49
![Page 50: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/50.jpg)
Learning (Structure Prediction)
50
Lexicon CombinatorsPredefined
Log-Linear
![Page 51: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/51.jpg)
Learning (Log-Linear Scorer)
51
![Page 52: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/52.jpg)
Learning (Log-Linear Scorer)
52
![Page 53: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/53.jpg)
GENLEX
53
Latent
![Page 54: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/54.jpg)
GENLEX
54
Lexical Generation Procedure
Given:Sentence Validation Function Log-Linear Model Lexicon
![Page 55: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/55.jpg)
GENLEX
55
Lexical Generation Procedure Lexemes
(garden, {garden})(house, {house})
Templates
Lexicon
OverlyGenerate
Given:Sentence Validation Function Log-Linear Model Lexicon
![Page 56: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/56.jpg)
GENLEX
56
Lexical Generation Procedure
Given:Sentence Validation Function Log-Linear Model Lexicon
Lexicon
Prune
![Page 57: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/57.jpg)
Unified Learning Algorithm
57
![Page 58: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/58.jpg)
Unified Learning Algorithm
58
Iterate over data
![Page 59: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/59.jpg)
Lexical generation
59
Generate a large set of potential lexical entries
![Page 60: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/60.jpg)
Lexical generation
60
Get top parses
![Page 61: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/61.jpg)
Lexical generation
61
Get lexical entries from highest scoring
valid parses
![Page 62: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/62.jpg)
Lexical generation
62
Update model’s lexicon
![Page 63: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/63.jpg)
Update parse scorer
63
Re-parse parsers and divide them into
“good” and “bad” ones
![Page 64: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/64.jpg)
Update parse scorer
64
Pick the “good” and “bad” parsers violating
the margin
![Page 65: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/65.jpg)
Update parse scorer
65
Update towards ‘good’ parses and against
‘bad’ parses
![Page 66: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/66.jpg)
Update parse scorer
66Output Lexicon &Log-Linear Model
![Page 67: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/67.jpg)
Semantic Forms
67
Liang et. al, Learning dependency-based compositional semantics, 2011Liang et. al, Lambda dependency-based compositional semantics, 2013Jonathan et. al, Semantic parsing on Freebase from question-answer pairs, 2013Jonathan et. al, Semantic parsing via paraphrasing, 2014
- Combinatory Categorial Grammar- Definition- Rules- Learning
- Dependency-based compositional semantics- Definition- Rules- Learning
- Neural Symbolic Machine (Functional-Program)- Definition- Learning
![Page 68: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/68.jpg)
Dependency-based Compositional Semantics- Weakness of CCG Parse
- Too much hand-coded rules for lexicon construction.- Learning algorithm is complicated.- The grammar rules are too strict.- Learning requires annotating logic form.
68
![Page 69: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/69.jpg)
Dependency-Based Compositional Semantics (DCS)
69
DCS
major cities in CA
![Page 70: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/70.jpg)
Dependency-Based Compositional Semantics (DCS)
70
DCS
major cities in CA
Predicate
![Page 71: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/71.jpg)
Dependency-Based Compositional Semantics (DCS)
71
DCS
major cities in CA
1st elem of “city” equals 1st elem of “major”
![Page 72: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/72.jpg)
Dependency-Based Compositional Semantics (DCS)
72
DCS
Weak-Supervision
Few hand-coded lexical rules
Align Syntactic & Semantic
![Page 73: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/73.jpg)
Strong vs. Weak Supervision
73
![Page 74: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/74.jpg)
Strong vs. Weak Supervision
74
![Page 75: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/75.jpg)
Lexicon vs. Triggers
75
requires strict grammar rule during parsing and harsh type constraint
CCG
![Page 76: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/76.jpg)
Lexicon vs. Triggers
76
{(city, city),(city, state),(city, river), . . . }
requires strict grammar rule during parsing and harsh type constraint
CCG DCS
requires only trigger words for predicates without grammar rule
![Page 77: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/77.jpg)
Syntactic & Semantic Alignment
77
city
major
amod
CA
nmod
incase
Utterance: major cities in CA
DependencyParsing
![Page 78: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/78.jpg)
Syntactic & Semantic Alignment
78
city
major
amod
CA
nmod
incase
Utterance: major cities in CA
DependencyParsing
Dependency-basedCompositional Semantics
![Page 79: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/79.jpg)
Parsing: Lexical Triggers
79
![Page 80: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/80.jpg)
Parsing: Lexical Triggers
80
![Page 81: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/81.jpg)
Parsing: Predicates to DCS Trees (DP)
81
![Page 82: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/82.jpg)
Parsing: Predicates to DCS Trees (DP)
82
![Page 83: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/83.jpg)
Parsing: Predicates to DCS Trees (DP)
83
![Page 84: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/84.jpg)
Denotation
84
Assuming we have our world knowledge containing 1) unary predicates 2) binary predicates in a database.
![Page 85: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/85.jpg)
Denotation
85
How can we assign values to different nodes?
?
?
?
![Page 86: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/86.jpg)
Denotation
86
Constraint 1: the value in the each node should come from its corresponding predicate table.
![Page 87: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/87.jpg)
Denotation
87
Constraint 2: the assignment of adjacent nodes need to meet the requirement in the edge.
![Page 88: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/88.jpg)
Constraint Satisfaction Problem
88
A DCS tree assignment becomes a constraint satisfaction problem (CSP)
![Page 89: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/89.jpg)
Constraint Satisfaction Problem
89
A DCS tree assignment becomes a constraint satisfaction problem (CSP)Dynamic Programming => time complexity = O(#node)
![Page 90: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/90.jpg)
Learning
90
![Page 91: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/91.jpg)
Learning
91
![Page 92: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/92.jpg)
Learning
92
![Page 93: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/93.jpg)
Semantic Forms
93
Liang, Chen et al. Neural Symbolic Machines: Learning Semantic Parsers on Freebase with Weak Supervision. 2017Liang, Chen et al. Memory Augmented Policy Optimization for Program Synthesis and Semantic Parsing. 2018Agarwal et al. Learning to Generalize from Sparse and Underspecified Rewards. 2019
- Combinatory Categorial Grammar- Definition- Rules- Learning
- Dependency-based compositional semantics- Definition- Rules- Learning
- Neural Symbolic Machine (Functional-Program)- Definition- Learning
![Page 94: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/94.jpg)
Neural Symbolic Machine- Weakness of traditional Semantic Parsing
- The previous methods heavily rely on the predicate mapping, the mapping procedure is based on rules.
- The coverage is problematic when dealing with more complex utterance.- The previous frameworks are dominated by the composition rules enforced as prior,
only few parameters to learn.- The symbolic executor and the ranking model are separated without interaction.
94
![Page 95: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/95.jpg)
Neural Symbolic Machine
95
![Page 96: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/96.jpg)
Neural Symbolic Machine
96
NSM
Simplified Domain Specific Language
Almost no rules/triggers required
Interacted Learning & Search
Learning-driven framework
![Page 97: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/97.jpg)
Domain Specific Language
97
![Page 98: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/98.jpg)
Domain Specific Language
98
CCG DCS DSL
Evolution
1. CCG/DCS are both general-purpose semantic form.
2. Evolve gradually towards more domain-specific languages.
3. Sacrifice some generalization, but greatly simplify the induction procedure.
Loose Loose
![Page 99: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/99.jpg)
Function definitions- (Hop v p): entities reached from v using relation p- (ArgHop v1 v2 p): entities in v1 from which v2 can be reached using relation p- (ArgMin v n): entities in v which have the lowest number in field n- (ArgMax v n): entities in v which have the highest number in field n- …- (Count v): number of entities in v
99
![Page 100: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/100.jpg)
Compositionality
100
(USA) CityIn
(Hop v0 CityIn)
(Argmax v1 Population)
Population
x: Largest city in the US => y: NYC
![Page 101: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/101.jpg)
Memory-aided Sequence
101
(USA) CityIn (Hop v0 CityIn) (Argmax v1 Population)
x: Largest city in the US => y: NYC
v0 (USA)
v1 (Hop ...)
v0 (USA)
The graph structure is flattened as sequence
![Page 102: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/102.jpg)
Large Search Space
102
( Argmax v1 Population )
Hop
Count
v0 Size
Elevation
![Page 103: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/103.jpg)
Compiler-aided prune
103
( Argmax v1 Population )
Hop
Count
v0 Size
Elevation
Code assistance
![Page 104: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/104.jpg)
Semantic Parsing as Seq-to-Seq
104
![Page 105: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/105.jpg)
Semantic Parsing as Seq-to-Seq
105
![Page 106: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/106.jpg)
Semantic Parsing as Seq-to-Seq
106
![Page 107: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/107.jpg)
Semantic Parsing as Seq-to-Seq
107
![Page 108: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/108.jpg)
Memory Enc-Dec Model- Overall model is conditional distribution over all token sequences.
108
![Page 109: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/109.jpg)
Memory Enc-Dec Model- Overall model is conditional distribution over all token sequences.
- NSM models the probability at each time step conditioned on previous output tokens and utterance
109
![Page 110: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/110.jpg)
Memory Enc-Dec Model- Overall model is conditional distribution over all token sequences.
- NSM models the probability at each time step conditioned on previous output tokens and utterance
- NSM uses interpreter to aid the decoder by removing impossible tokens at each step to dramatically shrink search space.
110
![Page 111: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/111.jpg)
Memory Enc-Dec Model
111
![Page 112: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/112.jpg)
Syntactic Constraint
112
![Page 113: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/113.jpg)
Semantic Constraint
113
![Page 114: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/114.jpg)
Augmented REINFORCE- REINFORCE (explore)- Iterative Maximum Likelihood (exploit)
114
![Page 115: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/115.jpg)
Wrap up- Formal Semantics does have many advantages like strong compositionality,
explainability.- However, the annotation effort or rigid structure makes it hard to scale up to
large-scale domains or more realistic datasets.- The vectorized representation in Deep Learning is still the favorable by the
community.- How to neuralize the formal semantics and apply it to modern architecture
remains an open problem.
115
![Page 116: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/116.jpg)
My Work1. Natural Language Understanding on Structured Data
a. Relation Question Answering on Knowledge Graph (Chen et al. NAACL 18)b. Table-based Fact Verification on semi-structured table (Chen et al. Arxiv)c. Multi-model Question Answering on structured scene graph (Chen et al. Arxiv)
2. Natural Language Generation on Structured Dataa. Semantically conditioned dialog generation on structured semantic form (Chen et al. ACL 19)
116
![Page 117: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/117.jpg)
Open Question
- Philosophical problem
117
Meaning
Language Inference
Machine Translation
Summarization
Meaning serves for the success of downstream tasks
![Page 118: Natural Language Learning with Formal Semantics · 2. Formal Semantics (Montague) - We construct the meaning of natural language in terms of variables, predicates, arguments. - Formal](https://reader034.vdocuments.mx/reader034/viewer/2022042105/5e83a4be73ead8181303523d/html5/thumbnails/118.jpg)
Open Question
- Philosophical problem
118
Meaning
Language Inference
Machine Translation
Summarization
If we can already achieve high accuracy, why do we still need to care about meaning?