74'406 Natural Language Processing

1
74.406 Natural Language Processing

• Semantics 0
• Flakey

2
Semantics

• How it should not be.

3
My Chat with A.L.I.C.E.

• Human hello.
• ALICE Hi there!
• Human how are you?
• ALICE I am functioning within normal
  parameters.
• Human good.
• ALICE What do you really want to talk about?
• Human about you.
• ALICE Me?
• Human yes.
• ALICE Aw.

4

• Human how are you?
• ALICE Everything is going extremely well.
• Human good.
• ALICE Thanks for the compliment.
• Human do you have a grammar?
• ALICE Doesn't everyone have a grammar. I have
  a lot of friends on the Internet.
• Human what friends?
• ALICE Are you using Netscape or Explorer?
• Human feature structures.
• ALICE Do you have any conditions I should know
  about?

5
Web References

• http//www.alicebot.org/join.html
• http//alice.pandorabots.com/
• http//www.uib.no/People/hhiso/eliza.html
• http//128.2.242.152/afs/cs/project/ai-repository/
  ai/areas/classics/eliza/0.html
• http//www.org.hha.dk/es/nexcon/Intelligence/ai-ty
  pes.html

6
Semantics

• How it could be.

7
Flakey - Example

• Flakey as Communicating Agent
• Case Frame Representation
• Concrete and Generic Actions
• Effects of Actions
• Inference / Reasoning
• Two Types of Questions

8
Flakey as Communicating Agent

• "Flakey, bring this file to
  Karen."
• listener verb determiner noun preposition
  noun
• V NP
  PP
• NounPhrase Prepos. Phrase
• head direct object
  indirect object
• agent action patiens recipient

9
Case Frames for Representing NL

• "Flakey, bring this file to
  Karen.
• verb NounPhrase PPhrase
• head direct object indirect object
  
• case frame
• action bring head verb
• patiens this file direct object
• recipient to Karen indirect object
• manner now modifier - added-

10
Mapping Case Frames to Actions

• case frame
• agent Flakey
• action bring head
• patiens this file direct object
• recip. to Karen indirect object

robot action precondition have (Flakey,
file1) action bring (Flakey, file1,
Karen) effect not (have (Flakey, file1))
and have (Karen, file1)
11
Concrete and Generic Actions I
robot concrete action (instance of
action) precondition have (Flakey,
file1) action bring (Flakey, file1,
Karen) effect not (have (Flakey, file1))
and have (Karen, file1)
robot generic action (concept in
KB) precondition have (agent,
object) action bring (agent, object,
recipient) effect not (have (agent,
object)) and (have (recipient, object))
12
Concrete and Generic Actions II
generic action bring (agent, object,
recipient) precondition have (agent,
object) effect not (have (agent, object))
and (have (recipient, object))
instance of action bring (Flakey, file1,
Karen) precondition have (Flakey,
file1) effect not (have (Flakey, file1))
and have (Karen, file1) You may add objects
as features to action-frame.
13
Effects of Actions
instance of action bring (Flakey, file1,
Karen) precondition have (Flakey,
file1) effect not (have (Flakey, file1))
and have (Karen, file1) effect of this
action delete from KB have (Flakey, file1)
add to KB have (Karen, file1) preconditions
and effects are formulae describing world states,
related to thedynamic KB.
14
Flakey - Reasoning, Inference
Integrate General Rules (Proper Axioms Theory)
Axiom ?x?y?loc (have(x,y) ? (at(x,loc) ?
at(y,loc))) Reasoning / Inference have
(Flakey, object) ? at (Flakey, here) ? at
(object, here)
15
Flakey - Question Answering I

• Flakey, where did you bring the file.
• case frame
• action bring
• patiens file1
• destination ?

Compare to stored case frames case
frame action bring patiens
file1 destination Karen Conclusion and
answer I brought the file to Karen.
16
Flakey - Question Answering II

• Q Flakey, where is the file.
• case frame
• action/status is
• subject the file identify with file1
• location ? refers to loc of file1

Access dynamic KB (world state) Stored from
effect of bring-action or pre-stored ...
at (file1, Karen), ... have (Karen, file1),
... A The file is at Karen. or "Karen has the
file."