Overview

• Linguistic Digital Divide
• Morphology
• Computational morphology
• Some applications of computational morphology
• HornMorpho
• Ongoing and future work

The Digital Divide

• The Digital Revolution: an unprecedented increase in
  • quantity of information available
  • ease of access to information
  • ease with which knowledge can be produced: knowledge production resources
• But access to the information and knowledge production resources is uneven: the information/knowledge gap is probably larger than before the Digital Revolution.
• Digital Divide within countries and between the Global North and Global South

The Linguistic Digital Divide

• Availability of information and knowledge production resources across languages is also uneven.
• 6000+ languages, but about 240 languages spoken by about 96% of the people
• Perhaps 15 languages account for 95% of web pages
• Information accessibility and ease of knowledge production also depend on language-specific resources: spell checkers, voice recognition and speech synthesis software, parsers, ontologies, digital lexicons, etc.; for most languages none of these are available.
• The L³ Project seeks to address the Linguistic Digital Divide for selected "disadvantaged" languages.

Morphology

• Internal structure of words
• luckier = lucky + -er
• luckier = lucky + COMP
• Surface and lexical forms
• Stem and affixes, morphemes
• Alternation rules: y → i following C, before -er, -est, -ed
• luckier = [luck + -y] + -er
• luckier = [luck + N->ADJ] + COMP
• Root and stem
• rides = ride(V) + 3PS_PRES | ride(N) + PLUR
• Ambiguity

Computational morphology

• Software that processes words
• Analysis
  • From surface to lexical form
  • dhaqxe → dhaq- + -t + -e
  • dhaqxe → dhaq- + [SBJ: 2PS, TNS: PST] | [SBJ: 3PFS, TNS: PST]
• Generation
  • From lexical to surface form
  • bit- + [SBJ: 1PS, TNS: PST, +NEG] → binne

Uses of computational morphology

• Information retrieval
  • swallow poison
    ⇒ ...swallowing...poison..., ...swallowed...poison..., ...
    
  • summii liqimsuu
    ⇒ ...summii...liqimsu..., ...summii...liqimsitu..., ...
• Machine translation
  • Kompyuutarri ... Afaan Oromootiinis ajaja fudhatee deebii kennuu ni danda’a.
    ⇒ ...fudhatee → fudhadh- + [SBJ: 3PMS, +CONT]...
    ⇒ ...receive/accept... ⇒ Computers can accept commands and respond to them in Oromo too.
  • Computers can accept commands and respond to them in Oromo too.
    ⇒ Computer(s) ... accept ... and ... respond
    ⇒ ...fudhadh- + [SBJ: 3PMS, +CONT] → fudhatee...
    ⇒ Kompyuutarri Afaan Oromootiinis ajaja fudhatee deebii kennuu ni danda’a.
• Spell checking
• Computer-assisted language learning

Simple and complex morphology

• Languages with simple morphology
  • Lexicon: ...,accept, accepts, accepted, accepting,...
  • No need for computational morphology since all forms can be stored in the lexicon
• Languages with complex morphology
  • Lexicon: ...,fudhachuu, fudhadha, fudhatta, fudhata, fudhatti, fudhanna,...!
  • Impractical to store all forms in the lexicon

HornMorpho

• Components (as in other modern computational morphology systems)
  • Lexicon of 4,112 roots and stems (extracted from dictionaries of Tamene Bitima, Gene Gragg)
    ...,bit-, bitadh-, bitam-, bixxil-,...
  • Morphotactics: rules governing the sequence of possible morphemes
  • Alternation rules: rules governing pronunciation and spelling changes that take place where morphemes come together ()
  • Processor
• Program written in Python programming language
• Currently mainly useful for computational linguists as a tool in a larger system
• Downloadable (along with extensive documentation) at http://www.cs.indiana.edu/~gasser/Research/software.html

HornMorpho examples

% python3.1
Python 3.1.2 (r312:79360M, Mar 24 2010, 01:33:18) 
>>> import l3

>>>>> This is HornMorpho, version 2.1.1 <<<<<

>>> l3.anal_word('om', 'dubbataniiru')
Word: dubbataniiru
POS: verb, root: , citation: dubbachuu
 subject: 3, plur
 TAM: perfect

>>> l3.anal_word('om', 'gargaaramna', segment=True)
Word: gargaaramna
POS: verb, segmentation: {gargaaram}-n-a
 subject: 1, plur
 TAM: present
POS: verb, segmentation: {gargaar-am}-n-a
 subject: 1, plur
 derivation: passive
 TAM: present

>>> l3.gen('om', 'taphadh', '[sb=[+p1,+pl],tm=prs,+neg]')
taphannu

Ongoing and future work

• Adding nouns and adjectives
• Computational syntax (sentence structure)
• Evaluation and refinement of software through statistical studies of actual language use
• Machine translation within a narrow domain, in collaboration with experts in Oromia

Conclusions

• Oromos suffer a disadvantage because of the lack of documents and the lack of computational resources in their language.
• Collaboration between computational linguists and native speakers can begin to address this disadvantage.
• Afaan Oromoo is a morphologically complex language, so morphological analysis and generation is a first step toward more extensive processing of the language.