Semantic Roles are descriptions of the semantic relation between predicate and its arguments with application in relation extraction , question answering
6. SRL Web Service
Web Server
JSON Web Services
••Multiple Parameters
support
••Modular Design. New
functionalities can be
added with URL
rerouting
SRL Visualization
••Scala+ AJAX + HTML +
Java Scripts
SRL System Jar (Java)
All information can be
made accessible by
web service
Docker
Container
User
Web Service or SRL
Visual page
You can see the the links for webpage and web
service here :
(More details in later
slides.)
http://wiki.emotibot.com:8090/display/NLP/SRL
7. SRL Process from Web Request to end result
Web Server
JSON Web Services
••Multiple Parameters
••Backward Compatible across
SRL Versions
••Modular Design. New
functionalities can be added
with URL rerouting
SRL Visualization Web
Page
••Scala+ AJAX + HTML + Java
Scripts enabled Web page
SRL System
1) Word
segmentation
(NLP)
2) POS Tagging
3) Dependency
Parsing
4) SRL
Prediction
5) SRL Format
Parser + JSON
Docker
User/module
Use Web Service
or SRL Web Page
1
2
3.1 3.2 3.3 3.4 3.5
Number
Indicate steps in
the processing
8. SRL Data Tagging Process
SRL Data Ready for Model training
1) SRL Model 2) POS Model* 3) Dependency Parser Model*
4) Run Data Validation Program (SRL System)
Run Validation
Fix Mistakes and repeat running program until all data
is validated
3) Human Tagging
1) Add/Modify SRL Tags 2) Add/Modify Predicates 3) Fix POS *
4) Fix Dependency
Parse*
2) SRL System
1) First Create CONLL Format Sentence
2) Create HIT format Sentence (easier
for tagging)
3) Output is TSV format data that can be
opened in excel
1) Input : Sentences
The whole
process is
performed
again for
new data
POS Model is not trained anymore as we use it from NLP module .
Dependency Parser Model is trained mainly to fix bugs.
11. SRL Model Training Process
Model Evaluation
4. Save a single model
3. Training Pipeline of Classifiers
1) Predicate Disambiguation 2) Argument Identification Module 3) Argument Classification
2. Data Format Conversion
1) Data Format Converter
••Converts from HIT to CONLL Format
2) Data Validator
••Tells us if any data mistakes
1. Input : HIT Data Format