General §
rir:TimeVL 11 §
- Ending [[[VL11] Notes IL]]
rir:Message2Semantics §
- How to implement semantics into parsing
- Why?
- Question answering
- Chatbots
- What do we want it to do?
- Represent meaning and relationships
- Logical form as a result
- What do we need?
- Verifiability
- Unambiguity
- Canonical Form
- Inference, variables
- Expressiveness
rir:Database2How to represent semantics? §
rir:DualSim1First order predicate calculus (FOPC) §
- = Mathematical Formalism to represent meaning
- Represents: Objects, Properties and relations
Why? §
- Well understood, easy to use
- Sufficient for many applications
rir:LayoutGridElements: §
rir:AlarmWarningProblems: §
- Multiple descriptions for the same semantic meaning
- Not as expressive, complicated sentences are hard to represent
- Vague information and beliefs hard to represent
rir:CheckboxMultipleBlankHow to combine semantics and syntax? §
- Syntax-driven semantic analysis
- Grammar rules need to include syntax & semantics
- Semantic information is passed from children to parents
- Example: TR Discover
- Problems:
- Incompleteness
- Natural language is NOT mathematics
- Usual problems with symbolic approaches
rir:BarChartChompsky Hierarchy §
- = Classification of grammars
- Helps for predicting the complexity of the semantic representation
- Grammars have matching Automata
rir:TreasureMapApplication areas §
- Question answering, dialogue
- Text summarisation/classification
rir:QuestionQuestion answering §
- IBM-Watson
- Reading comprehension
rir:BubbleChartDialogues §
- Often: Slot-Filling approach
- Slots to be filled with information of a given type
- How to respond?
- Retrieval
- DB with dialogues, find most similar query
- Return answer to the matched query
- Generation
- Encoder-Decoder to generate response
- Extractive summarisation
- Extracting sentences of importance from corpora
- Abstractive summarisation
rir:PagesCommon themes §
- Rule-Based Vs. ML-Based Vs. Hybrid
- Industry Vs. Academia
- Dense vectors
rir:FileResources: §