Writing Support in Local and WebBased Applications for Slow Typists

1
Writing Support in Local and Web-Based
Applications for Slow Typists

• Christian Bühler, Helmut Heck, Dirk Clemens, Dirk
  Sischka

2
Contents

• Introduction
• Methods of Word Prediction
• Application Areas

3
Introduction

• Means of text input
• QWERTY keyboard
• Special keyboards
• Mouth/head stick
• Switches, virtual keyboard scanning

4
Introduction

• Input Rates (strokes/min)
• Professional typist gt300
• Average 150
• Mouthstick 75 120
• Foot keyboard 70
• Switch/scanning a virtual keyboard 3-10

5
Introduction

• Scanning

6
Introduction

• Scanning

7
Introduction

• Scanning

8
Introduction

• Scanning

9
Introduction

• Scanning

10
Introduction

• Scanning

11
Introduction

• Scanning

12
Introduction

• Scanning

13
Introduction

• Scanning

14
Introduction

• Scanning

15
Introduction

• Scanning

16
Introduction

• Measures to improve input rates
• Abbreviation Expansion
• pre-defined abbreviations expand to phrases or
  whole sentences
• Word Prediction
• offering a list of words based on one or more
  letters or words already written

17
Introduction

• Enhancement of text input rate by word prediction
• Monitoring the recent textual input
• Prediction of likely text continuations
• Presentation of likely text continuations
• Easy selection and insertion

18
Introduction

• Treasury of Words (German)
• Passive vocabulary 200,000 words
• Passive vocabulary of an individual 50,000 words
• Active vocabulary 2,000 4,000 words
• (professional speaker 10,000 words,
• Goethe 80.000 words)

19
Methods
morphology
statistics
Word prediction system
collocation
semantics
heuristics
20
Methods

• A simple word prediction system
• develops predictions right out of a dictionary
• no additional measures

21
Methods
Source public domain English word dictionary
OPTED (182,702 entries)
22
Methods

• Statistical prediction

Source Brown Corpus (1,014,312 words/15 text
categories)
23
Introduction

• Basic forms of statistical models
• Elements mapped to the frequency of use
  (unigrams)
• Groups of two elements (bigrams)
• Groups of three elements (trigrams)
• Groups of n elements (n-grams)
• Elements letters, syllables, words

24
Introduction
Reports
Novels
Periodicals
Newspapers
Tracts
Test corpus
Documents
..
Statistical model (Data base)
Processed words for the ASK-IT writing support
120 million
25
Methods

• Linguistic Approach Morphology
• Relations that connect words by rules
• Part-of-speech
• Plural formation
• Inflection

26
Methods
Linguistic Approach Collocation Sequence of
words which co-occur often
accident car owner pool
27
Methods
Linguistic Approach Collocation

• arrangement
• convenient
• armchair

28
Methods

• Linguistic Approach Semantics
• Assignment of semantic categories by
• hand
• parsing method
• (probabilistic) Latent Semantic Analysis
• Latent Dirichlet Allocation

29
Methods

• Linguistic Approach Semantics

30
Methods

• Additional measures
• Heuristics
• u after q
• Blank after punctuation marks
• Blank after word completion
• Capitalisation after punctuation marks
• Personalisation
• Adaptation to the typists style of writing
• Tailoring the systems appearance to the typists
  preferences

31
Methods
Composition
text corpus
heuristics
static
statistic model
linguistic model
dynamic
text input
prediction algorithm(s)
prediction result
32
Application Areas

• Stand-alone application
• Text production aid
• Communication aid
• Learning aid

33
Application Areas

• Client-server application
• Public terminals
• Kiosks
• Web-based applications

34
Application Areas
35
Application Areas
Distributed functionality
http request
xajax_getProposal
Client (user interface)
ASK-IT Server (xajax)
Word Prediction Server
XML data
proprietary data structure
Browser client
Server side system
36
Application Areas

• Positioning within ASK-IT
• e-learning
• e-working
• related use cases
• Supported languages (so far)
• French
• English
• Greek
• German

37
Application Areas

• Thank you for your attention