learning plans from english procedure...

58
Dustin Smith [email protected] MIT Media Lab Learning Plans from English Procedure Descriptions

Upload: others

Post on 26-Mar-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Dustin [email protected]

MIT Media Lab

Learning Plansfrom English

Procedure Descriptions

Page 2: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Machine Reading

- How can we get knowledge into computers?

- Machine reading: processing and understanding natural language-- accumulating relevant knowledge through inferences about text.

2

Page 3: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Approach 1: Read Children’s Stories

3

Jack was having a birthday party. Mother baked a cake.

Page 4: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Approach 1: Read Children’s Stories

- What is the relationship between Jack and Mother?

- Who will eat the cake?

- Where was Mother when she was baking the cake?

- Where was Jack?

- ... 3

Jack was having a birthday party. Mother baked a cake.

Page 5: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Eugene Charniak’s Ph.D. [1] “the chief concern motivating the model discussed here is relating a large body of knowledge to a particular story”

By preschool, children can represent and remember event sequences [2].

By the time they are reading, children have acquired a lot of world knowledge already!

4

[1] E. Charniak “Toward a Model of Children’s Story Comprehension.” 1972.[2] J. Wenner, P.J. Bauer. Bringing order to the arbitrary. 1999.

Approach 1: Read Children’s Stories

Page 6: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Big QuestionCharacterizing the problem:

How can we acquire knowledge by reading, when reading itself requires knowledge?

5

Page 7: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Big QuestionCharacterizing the problem:

How can we acquire knowledge by reading, when reading itself requires knowledge?

5

- Bootstrap solution: Gradually compose more new (more complex) representations out of existing (simpler) representations.

Page 8: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

6

Approach 2: Use Commonsense KB

Jack was having a birthday party. Mother baked a cake.

is a gameis a children’s gameis a nickname for John.is a boy’s nameis a childhood game.is a lifting device.is kind of nickname.

(40)

Use OpenMind Common Sense knowledge base [1].

[1] http://commons.media.mit.edu

Page 9: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

7

Approach 2: Use Commonsense KB

Jack was having a birthday party. Mother baked a cake.

is a gameis a children’s gameis a nickname for John.is a boy’s nameis a childhood game.is a lifting device.is kind of nickname.

(40)

bake a cake because you want toballoon used forlikely to find at toy balloonlikely to find at helium balloonbuying presents is forare fun

(17)

Page 10: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

8

Approach 2: Use Commonsense KB

Jack was having a birthday party. Mother baked a cake.

is a gameis a children’s gameis a nickname for John.is a boy’s nameis a childhood game.is a lifting device.is kind of nickname.

(40)

bake a cake because you want toballoon used forlikely to find at toy balloonlikely to find at helium balloonbuying presents is forare fun

(17)

can care for a childis a womantake care of their childrenloves her childis part of my family

(260)

Page 11: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

9

Approach 2: Use Commonsense KB

Jack was having a birthday party. Mother baked a cake.

is a gameis a children’s gameis a nickname for John.is a boy’s nameis a childhood game.is a lifting device.is kind of nickname.

(40)

bake a cake because you want toballoon used forlikely to find at toy balloonlikely to find at helium balloonbuying presents is forare fun

(17)

can care for a childis a womantake care of their childrenloves her childis part of my family

(260)

you should have an ovena birthday may make you want tobecause you want to celebrate a birthdayfirst thing add flour

(19)

Page 12: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

10

Approach 2: Use Commonsense KB

Jack was having a birthday party. Mother baked a cake.

is a gameis a children’s gameis a nickname for John.is a boy’s nameis a childhood game.is a lifting device.is kind of nickname.

(40)

bake cake because you want balloon used forlikely to find at toy balloonlikely to find at helium balloonbuying presents is forare fun

(17)

can care for a childis a womantake care of their childrenloves her childis part of my family

(260)

Problem: Retrieving only the relevant knowledge

you should have an ovena birthday may make you want tobecause you want to celebrate a birthdayfirst thing add flour

(19)

Page 13: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

11

Plans: a solution?

- are also called: plans, event chains, stories, scripts, narratives, procedures, task networks, sequential decision processes, ...

- Unite procedural and declarative semantic knowledge.

- Structure what is problematic, what questions are worth asking; embedding knowledge in a goal-driven problem solving context.

"Questions arise from a point of view–from something that helps to structure what is problematical, what is worth asking, and what constitutes an answer (or progress). It is not that the view determines reality, only what we accept from reality and how we structure it...”

-- Alan Newell in Artificial Intelligence and the Concept of Mind

Page 14: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

12

Lexical-semantics of Events

- Verbs and their arguments reference lexical-semantic frames, or event structures [1].

- Properties of verbs (TimeBank)

- Tense: {none, present, past, future}

- Grammatical Aspect: {none, progressive, perfect, progressive perfect}

- Modality: {none, to, would, should, could, can, might}

- Polarity: {positive, negative}

- Event Type: { ?? }

[1] Levin B. and Hovav, Argument Realization 2005.

Page 15: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

12

Lexical-semantics of Events

- Verbs and their arguments reference lexical-semantic frames, or event structures [1].

- Properties of verbs (TimeBank)

- Tense: {none, present, past, future}

- Grammatical Aspect: {none, progressive, perfect, progressive perfect}

- Modality: {none, to, would, should, could, can, might}

- Polarity: {positive, negative}

- Event Type: { ?? } - Need a semantic theory!

[1] Levin B. and Hovav, Argument Realization 2005.

Page 16: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

13

Lexical-semantics of Events- Some compelling evidence of this relationship from

cognitive science:

- Verb aspect influences retrieval of event knowledge:

- imperfective form (was verb-ing) versus perfect form (had verb-ed) in a statement (e.g. “I was cooking/ I had cooked”) changes the retrieval of the location “kitchen” [1].

- When observing the same event, speakers of different languages attend to the event features relevant to organizing their languages specific verb lexicon, e.g. path versus manner [2].

- When reading about actions, people’s corresponding brain motor regions are activated [3].

[1] T. R. Ferretti, M. Kutas, and K. Mcrae. Verb aspect and the activation of event knowledge. 2007.[2] A. Papafragou, J. Hulbert, and J. Trueswell. Does language guide event perception? ... 2008. [3] R. A. Zwaan and L. J. Taylor. Seeing, acting, understanding: motor resonance in lang. comp. 2006.

Page 17: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

14

Acquiring Plans the Open Mind Way

- Honda’s OMICS project [1], a clone of the Media Lab effort. Collecting knowledge exclusively about indoor common sense. High quality, manually reviewed.

- Has a parallel corpus of stories about ways to accomplish 174 different goals.

[1] http://openmind.hri-us.com/

Page 18: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

15

access the internetact as a security guardanswer the doorbellanswer the phoneapply band aidassist person standing upassist someone in walkingboil the milkbuy from vending machinecall 911calm an infantchange a baby diaperchange a bulbchange bed sheetscharge a cell phonecheck for intruderscheck for weathercheck if a store is openchop vegetablesclean a spillclean the dishesclean the floorclean the showerclean the tableclean upclean up toysclose the blindsclose the curtainscook fishcook noodlecook pastacook ricedance with the childrendo laundrydraw the curtains

dry clothesdust an objectempty the kitchen sinkempty the trashentertain childrenerase the whiteboardfeed a childfeed a pet catfeed a pet dogfeed infantfeed the fishfetch a cold drinkfetch a ladderfetch an objectfill water in containerfind a personfind an objectfind out more informationfind the timefold clothesfollow someone aroundgather all scattered toysget food from refrigeratorget mailget the newspapergive a medicinegive a messagegive a message on phonego outsidegreet a visitorguard the househandle toxic materialshang clothesheat food in microwaveheat food on kitchen gas

help someone carry thingiron clotheskeep the dog awaykick a ballload the dishwasherlock up the houselock windowsmail a lettermake a bedmake a dinner reservationmake a flight reservationmake a listmake a presentationmake a shopping listmake a tossed saladmake baby sleepmake breakfastmake coffeemake fresh orange juicemake hot dogmake soupmake sure children fedmake teamake toasted breadmaking omelettemove furnituremow the lawnopen a web pageopen packageopen the garageopen the mailpack a mailing boxpack a suitcasepaint a wallpay bills

perform research on specphotocopy a paperpick up dishesplace ladder near wallplay a game on the compplay a movieplay a songplay pianoplug an electric appliance in plug battery into chargerpour beer into a glassprint documentpush someone in a wheelpush somethingput away groceriesput object awayput up a paintingraise the blindsread a story to a childrecharge batteriesremove and replace garbagereplace a refrigerator filterreplace a water tap filterreplace batteries in the treplace heater filterretrieve a toolsecure all exitssecure all windowssecure the perimeter of send a faxsend party invitationsserve a drinkserve a mealset a wake up alarmset the dining table

Page 19: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

“sweep floor”

Problems with the corpus

Page 20: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

“sweep floor”

P1. many ways to say the same thing!

Problems with the corpus

Page 21: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

“sweep floor”

P2. Temporal Abstraction (nested events)

Problems with the corpus

Page 22: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

“sweep floor”

P3. Global Alignment(problem of context)

Problems with the corpus

Page 23: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

“sweep floor”

P4. Causal discontinuation (OR)

Problems with the corpus

Page 24: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

1. Parse English sentences: clean, parse, extract predicate-argument structures.

2. Find global alignment of stories: Read in stories one at a time, align sequences, use structure to either:

a) detect missing nodes / context (alignment)

b) detect disjunctions (abstraction / is-a)

c) detect nested sequences (composition / part-of)

3. Infer corresponding state descriptions: Construct corresponding situation models for each step.

- Evaluation: the narrative cloze.

The Approach

Page 25: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

22

Four example parsed narratives (of 36)

Page 26: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

23

... and related background knowledge

Page 27: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

1. Parse English sentences: clean, parse, extract predicate-argument structures.

2. Find global alignment of stories: Read in stories one at a time, align sequences, use structure to either:

a) detect missing nodes / context (alignment)

b) detect disjunctions (abstraction / is-a)

c) detect nested sequences (composition / part-of)

3. Infer corresponding state descriptions: Construct corresponding situation models for each step.

- Evaluation: the narrative cloze.

The Approach

Page 28: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

25

Sequence Alignment- Sequential modeling work from NLP on language

models, and bioinformatics on modeling nucleic acids in DNA/amino acids in proteins.

- Needleman and Wunsch sequence alignment algorithm. Given two strings, A and B, populate a Score matrix and use back-pointers to retrieve the best alignment.

- Gap penalty:

- Beyond 2 strings?

- Time Complexity for n=2 seq:

!OpenCost! (len(gap)! 1)GapExtensionCost

max (|A|, |B|)n

425 ! 1014

Page 29: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

26

n=1, pre-

Page 30: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

27

n=1, post-

Page 31: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

28

n=2, pre-

Page 32: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

29

n=2, post-

Page 33: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

30

n=3discarded

Page 34: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

31

n=4, pre-

Page 35: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

32

n=4, post-

Page 36: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

33

n=5 pre

Page 37: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

34

n=5 post

Page 38: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

35

n=7 pre

Page 39: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

36

n=7 post

Page 40: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

37

Needleman-Wunsch Results

Page 41: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

38

Beyond string matching- Match(Ai, Bj) = {1,0}

- But, how well do these match?

- get(mail) get(letter)

- open(box) open(mailbox)

- close(door) shut(door)

- return(home) go(inside)

Page 42: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

39

Beyond string matching- Match(Ai, Bj) = {1,0}

- But, how well do these match?

- get(mail) get(letter)

- open(box) open(mailbox)

- close(door) shut(door)

- return(home) go(inside)

[1] Jiang, J and Conrath, D: Semantic Similarity Based on Corpus Statistics and Lexical Taxonomy. 1997

- Jiang & Conrath generalization similarity metric [1]

Page 43: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

40

A more general description

L. De Raedt and J. Ramon. Deriving distance metrics from generality relations. 2008.

Page 44: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

41

n=4, pre-

Page 45: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

42

n=4, post-

JC similaritytext similarity

Page 46: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

43

n=5 pre

Page 47: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

44

n=5 post

JC similaritytext similarity

Page 48: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

45

Needleman-Wunsch (Relational) Results

Page 49: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

46

What about inferring hidden state?

go to the mailbox

open the mailbox

extract letter

in(mailbox, mail) at(home, Dustin) sealed(letter) closed(mailbox)

in(mailbox, mail) at(mailbox, Dustin) sealed(letter) open(mailbox)

in(hand,mail) at(mailbox, Dustin) sealed(letter) open(mailbox)

Page 50: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

- Learning a Left-Right Hidden Markov Model or Context Free Grammar (searching a model space, inducing a model w/ some MDL type constraint)

- Good for dealing with sequential, incomplete data. Not good for relational rich data. Hard to augment background knowledge.

- Relational Markov Models? Logical Hidden Markov Models (LoHMMs)?

47

Possible Solutions

Page 51: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Relational Markov Models

C. Anderson, P. Domingos, and D.S. Weld. Relational Markov Models and their Application to Adaptive Web Naviation. 2002.

A hidden Markov model

and a semantic abstraction

hierarchy (a taxonomy).

Page 52: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Relational Markov Models

C. Anderson, P. Domingos, and D.S. Weld. Relational Markov Models and their Application to Adaptive Web Naviation. 2002.

A graph rank-based heuristic

on the state transitions is used

to generalize arguments,

in favor of a compact representation.

Page 53: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Problems Representational Solutions

P1. Lexical Ambiguity

P2. Temporal Abstraction

P3. Global Alignment

P4. Causal discontinuation

1. Taxonomic (Is-A) hierarchy

2. Compositional (Part-of) hierarchy

3. Sequential

An ideal representation?

Page 54: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Representational Abstraction

51

h3 = [shape(circle) ∧ taste(sweet) ∧ color(green)]

h0

h2h1

h5h4h3

generalization lattice

<- more general m

ore specific ->

lattice = a set, a partial ordering, and greatest upper/lower bound operations.

h0 = [shape() ∧ taste() ∧ color()]

A hypothesis space of abstractions (removing details).

Page 55: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Representational Composition- Bootstrap solution: Gradually compose more

new (more complex) representations out of existing (simpler) representations.

52

composition

appletaste(sweet)

color(light-green)

shape(circle) composition

Page 56: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Structural versus Assertional

Various encodings of the statement “John loves tacos:”

LOVES: SUBJECT: John OBJECT: tacos

Loves(John,tacos)

tacos

loves

John

Different structures, same assertions!

Structural knowledge is the internal organization of the knowledge while assertional knowledge is the set of claims it makes about the world.

Page 57: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Relational Sequence Alignment

[1] Much related work by Kristian Kersting and Luc De Raedt.

- Relational (in first order logic), instead of propositional, descriptions of states.

Predicates/arity: vi/2, cd/1, ls/0, pdfview/2

Ground atoms -- predicates with non-variable terms:

vi(ch2,tex)

Ground clauses:

Generalized Clauses (with variables):

cd(X), vi(Y, tex), latex(Y, tex)

Page 58: Learning Plans from English Procedure Descriptionsalumni.media.mit.edu/~dustin/0109-PlansFromProcesses.pdf · Learning Plans from English Procedure Descriptions. Machine Reading-How

Sequential

Relational Sequence Alignment

[1] Much related work by Kristian Kersting and Luc De Raedt.