PRESENTED BY:
ANIKET RANJAN

---

[attachment=13464]
NATURAL LANGUAGE UNDERSTANDING
INTRODUCTION
Here we are using NLU in cognitive geometry tutoring.
Finding semantic content of student explanation.
Cognitive means knowledge through experience and senses.
This tutor is based on syntactic processing with description logic.
INTRODUCTION
One problem in previous tutoring system is inability to determine semantic content of student input.
Present system is an intelligent tutor which is dialogue intense.
Which can perform
Syntactic structure analysis and
Semantic interpretation
of students explanation.
INTRODUCTION
Deals with determining semantic equivalence of natural language input provided.
Such tutors provide better understanding of concepts as it uses description logic
which is a formal language representation
more expressive than propositional logic.
Present system only propose problem, check solutions and some hints to get solution.
Do not ask to justify their solution with “why”, “what”.
OLD SYSTEM
OLD SYSTEM
This system asks to select from the list of templates like MS Lindquist.
Human tutors can reason and dialogue for the step chosen.
By-pass the need for NLU by entering or selecting the reason from list.
Fail to get in-depth semantic analysis of the sentences provided.
Mitigates system and student performance.
NEW SYSTEM
REASONING
In fourth step of solving the problem:
STUDENT: is an isosceles triangle, base angles same
TUTOR: gave correct statement of geometry, but it doesn’t apply here. Please focus.
STUDENT: interior angles are congruent
TUTOR: that’s a wrong rule. Please try again.
REASONING
STUDENT: two parallel lines cut by transversal, interior angles are same.
TUTOR: you don’t need to talk about interior angles here. Try again.
STUDENT: two parallel lines cut by transversal, exterior angles are same.
TUTOR: still not correct. Change the part “exterior”.
STUDENT: two parallel lines cut by the transversal, corresponding angles are congruent.
TUTOR: Correct answer. Proceed.
NEW SYSTEM
In-depth semantic analysis, engage in dialogues and understands students input in terms of natural language.
Performs considerable inference based on knowledge of domain to get exact meaning of input to impart proper teaching.
GCT implemented on ACT-R and provide corrective measure when needed and on demand.
Detailed assessment is maintained.
NEW SYSTEM
Exploits full NLU capability and enhances student’s performance.
This NLU version of GCT is GEOMETRY EXPLANATION TUTOR.
At each step gives answer and reason in NL expression form.
High precision level.
Parsing done by left corner chart parser which uses syntactic LFG(Lexical Functional Grammar) to model English.
WORKING
System creates semantic representation of meaning in DL system called LOOM->facilitates reasoning system.
Decides among the three options to choose from:
Answers
Questions √
Explanations
Has three main works:
Determine input content
Accept or decline
Take corrective action
WORKING CONTD…
Has classes that describe the semantic content of sentences to express theorems etc.
Systematically built semantic representation of NL sentences.
Accuracy obtained with DL-> which makes system portable.
Deals with following situations efficiently:
when same sequence of words express different meaning.
when words don’t give actual semantic content.
same meaning with different word.
DEALS WITH
Syntactic structure:
passive versus active
adjunct phrase attachment
clauses
Content word:
Generic concepts
Generic relation
Specific or explanatory
Ellipses
Anaphora
DEALS WITH…
Other semantic solutions:
Plurals
Syntactic ambiguity
Noun- noun compounds
Metonymy
Reference solution
PREVIOUS SYSTEMS
Based on Statistical Approach:
AUTOTUTOR
Drawback: cannot control analysis outcome
Based on Semantic Grammar:
SOPHIE and NESPOLE
Drawback: complex semantic computation
Based on Finite State Syntax Approach:
CIRCSIM
Drawback: cannot do full input analysis
PREVIOUS SYSTEMS
Based on Frame Based Semantic Approach:
ATLAS- ANDES
Drawback: absence of logic system
Based on Knowledge:
KANT and KANTOO
Drawback: independent of input content
Based on Semantic:
GEMINI- close to present system
Drawback: complex semantic computation
NEW SYSTEM DESIGN
FURTHER DEVELOPMENT
Can be used in remote areas
Can be developed for many languages
In more sophisticated teaching
Great advantage in learning programming with loads of reasoning and interaction.