shorthand handwriting recognition for pen-centric interfaces

PLT 2007

CSISCSIS

Shorthand Handwriting Recognition for Pen-Centric Interfaces

Charles C. Tappert1 and Jean R. Ward2

1 School of CSIS, Pace University, New York, USA2 Pen Computing Consultant, Massachusetts, USA

PLT 2007

CSISCSIS

Thesis: Pen-Centric, Chatroom-Like Shorthand Interfaces

• Will provide critical infrastructure for many pen-centric applications

• Will provide fast text input• Will have greatest impact on applications

running on small mobile devices

PLT 2007

CSISCSIS

Agenda• Handwriting

– Fundamental Property of Writing– Handwriting Recognition Difficulties

• Historical Shorthand Alphabets• Online (Pen-Centric) Handwriting Recognition

– Online more accurate than Offline Recognition– Online Info Can Complicate Recognition Process– Design Tradeoffs/Decisions

• Pen-Centric Shorthand Alphabets• Pen-Centric Word/Phrase Shorthand• Allegro/Chatroom Experimental Shorthand System

PLT 2007

CSISCSIS

Fundamental Property of Writing

• Differences between different characters are more significant than differences between different drawings of the same character

• This makes handwritten communication possible

PLT 2007

CSISCSIS

Fundamental Property of Writing

• Property holds within subalphabets of uppercase, lowercase, and digits, but not across them

• “I”, “l”, and “1” written with single vertical stroke• “O” and “0” written similarly with an oval

PLT 2007

CSISCSIS

Handwriting Recognition Difficulties

• Shape, size, and slant variation• Similarly shaped characters – U and V • Careless writing

– in the extreme, almost illegible writing• Resolving difficult ambiguities requires

sophisticated recognition algorithms, syntax/semantics

PLT 2007

CSISCSIS

Historical Shorthand Alphabets(prior to pen computing)

• Famous writings were written in shorthand – Cicero’s orations– Martin Luther’s sermons– Shakespeare’s and George Bernard Shaw’s plays

• We focus on shorthand appropriate for PDAs• Two main types of shorthand

– Non-geometric shorthand– Geometric shorthand

• Small number of basic shapes• Shapes reused in multiple orientations

PLT 2007

CSISCSIS

Tironian Alphabet, 63 B.C.

PLT 2007

CSISCSIS

Stenographie Alphabet, 1602

PLT 2007

CSISCSIS

Stenographie Alphabet, 1602

Geometric shorthand – basic shapes/orientations

PLT 2007

CSISCSIS

Moon Alphabet, 1894• Geometric shorthand – basic shapes/orientations

PLT 2007

CSISCSIS

Other Historical Shorthand Systems

• Phonetic alphabets– Pitman (1837)– Gregg (1885)

• Systems for the blind – Braille (1824)

• Cursive shorthands – Gabelsberger (1834)

PLT 2007

CSISCSIS

Online (Pen-Centric) Handwriting Recognition

• Machine recognizes the writing as the user writes • Digitizer equipment captures the dynamic

information of the writing – Stroke number, order, direction, speed – A stroke is the writing from pen down to pen up

PLT 2007

CSISCSIS

Online (Pen-Centric) more accurate than Offline (Static) Recognition

• Can use both dynamic and static information• Can often distinguish between similarly

shaped characters – E.g., 5 versus S where the 5 is usually written

with two strokes and the S with one stroke

PLT 2007

CSISCSIS

Online Information Can Complicate Recognition Process

• Segmentation ambiguities– Character-within-character problem – cl versus d

• Large number of possible variations – E can be written with one, two, three, or four strokes, and with various

stroke orders and directions – Four-stroke E has 384 variations (4! stroke orders x 24 stroke directions)

PLT 2007

CSISCSIS

Design Tradeoffs/Decisions

• No constraints on the user– Machine recognizes user's normal writing

• User severely constrained– Must write in particular style such as handprint – Must write strokes in particular order, direction,

and graphical specification• Simplest is one stroke per character, one stroke

direction, one shape

PLT 2007

CSISCSIS

Pen-Centric Shorthand Alphabets

• Some of the earliest were for CAD/CAM• Others developed for text input on PDAs • We review geometric and non-geometric

shorthands appropriate for small devices• Historical alphabets presented above could be

used for machine recognition• In addition to shape and orientation, stroke

direction can differentiate among symbols

PLT 2007

CSISCSIS

Allen Alphabet

PLT 2007

CSISCSIS

Allen AlphabetBasic Shapes and Orientations

PLT 2007

CSISCSIS

Goldberg Alphabet

PLT 2007

CSISCSIS

Goldberg AlphabetBasic Shapes and Orientations

PLT 2007

CSISCSIS

Graffiti Alphabet(non geometric )

PLT 2007

CSISCSIS

Allegro Alphabet (non geometric)

PLT 2007

CSISCSIS

Simplified Design Tradeoffs/Decisions for Graffiti and Allegro PDA Alphabets

• Small alphabet– one case rather than both upper and lowercase

• Small number of writing variations per letter– preferably only one

• One stroke per character (character = stroke)– allows machine to recognize each character upon pen lift

• Separate writing areas for letters and digits– avoids confusion of similarly shaped letters and digits

PLT 2007

CSISCSIS

Graffiti and AllegroCommercially Successful Shorthands

• High correspondence to Roman alphabet– Easier to learn– Graffiti used in Palm OS devices

• notably the Palm Pilot and Handspring models – Allegro used in Microsoft Windows devices

• Geometric alphabets not successful

PLT 2007

CSISCSIS

Pen-Centric Word/Phrase Shorthande.g., Chatroom Shorthand

• Further increase speed of text entry• Potential applications

– Where input speed important– Where word/phrase abbreviations occur

frequently – e.g., email

PLT 2007

CSISCSIS

Allegro/Chatroom Shorthand System

• Developed for M.S. dissertation– Student was hearing impaired– Developed as output component of

communication system• Handwriting to text to speech

• Two input writing areas– One for Allegro (all-purpose)– One for chatroom-like words/phrases (e.g., CUL, F2F)

PLT 2007

CSISCSIS

Allegro/Chatroom Shorthand SystemStroke acquisition GUI

allegro strokerecognition

alphabet

sentence accumulator

Sentence display and spoken output

allegro strokelibrary

user-defined stroke library

a single stroke

other strokerecognition

word/phrasecharacter

done? no

yes

meaning

is it

PLT 2007

CSISCSIS

Allegro/Chatroom Shorthand System

PLT 2007

CSISCSIS

Allegro/Chatroom Shorthand System Preliminary Experimental Results

• Allegro/Chatroom pen-centric shorthand input faster than typing text and comparable to typing text and chatroom shorthand characters

PLT 2007

CSISCSIS

Conclusions: Pen-Centric, Chatroom-Like Shorthand Interfaces

• Will provide critical infrastructure for many pen-centric applications

• Will provide fast text input• Will have greatest impact on applications

running on small mobile devices