Goals

• Develop morphological analyzer and generator for Tigrinya (and other Ethiopian Semitic languages) which
  • Return multiple analyses in cases of ambiguity
  • Avoid states which duplicate shared knowledge
  • Handle words for which the roots are unknown ("guesser" mode)

Tigrinya verb morphology 1

• Tigrinya: Semitic language spoken in Ethiopia and Eritrea, ~6,000,000 speakers, seriously under-resourced
• Stems: root-template morphology (interdigitation, intercalation)
  
  • Root: sequence of consonants, sometimes gemination ('_') and a
  • Template
    • Pattern of consonant slots, vowels, sometimes gemination
    • Represents derivational features, depends on tense-aspect-mood (maximum of 32 possibilities for a given root)

Tigrinya verb morphology 2

• Affixes
  • Prefixes: subject agreement, negation, relativization, prepositions, conjunctions
  • Suffixes: subject agreement, object pronouns (plain or oblique), negation, conjunctions
  • Long-distance prefix-suffix dependencies
• More than 100,000 wordforms possible for a single root
• No adequate root dictionary yet available

Tigrinya verb morphology: a maximal example

Representing the structure of a verb

'abzəytɨr_axəbalun ↔ `[ (text(root)=text('rkb')), (text(sbj)=[-text(p1),-text(p2),+text(plr),+text(fem),-text(prep)]), (text(obj)=[-text(p1),-text(p2),-text(plr),-text(fem),+text(prep)]), (+text(neg)), (+text(rel)), (text(prep)=text('ab'-)), (text(suf_conj)=text(-'n')) ]`

Finite state morphology

• Finite state transducers
  • Analysis: surface wordform to lexical/grammatical representation
  • Generation: lexical/grammatical representation to surface wordform
• Inversion: generator easily derived from analyzer
• Composition of FSTs: given an ordered set of rules, the effective order is preserved in the composition of the rules

The components of a complete system

• Alternation rules
  Example: the rule which converts lexical k and q to fricatives following a vowel when not geminated
  
• Morphotactics
  

The composed FST

The problem of non-concatenative morphology

• Finite state morphology is biased to view
  • morphemes as sequences of characters
  • words as concatenations of morphemes
• Problems when morphemes are discontinuous and/or when morphemes are not concatenated to form words
  • Discontinuous morphemes (e.g., circumfixes)
  • Reduplication
  • Semitic root-template morphology

Non-concatenative example 1

Imperfective stem, three voice categories, CC_C root

Non-concenative example 2

Negation, relativization dependencies

FSTs "weighted" with feature structures (Mohri, Amtrup)

• Feature structures (FSs): sets of attribute-value pairs
• Operation on FSs: unification
• FSTs weighted with FS sets
  • Traversing path yields FS set in addition to output string, the result of the repeated unification of FS sets on the transitions in the network, starting with an initial FS set
  • Path fails not only if the input character fails to match the input character on the arc, but also if the accumulated FS set fails to unify with the FS set on the arc
• Advantages
  • FSTs output FSs
  • Long-distance dependencies handled efficiently

Weighted FSTs for Tigrinya verbs: affix dependencies: analysis 1

Weighted FSTs for Tigrinya verbs: affix dependencies: analysis 2

Weighted FSTs for Tigrinya verbs: stem: analysis

Weighted FSTs for Tigrinya verbs: stem: generation

Tigrinya verb morphology FSTs: architecture

Evaluation

• Tigrinya guesser analyzer correctly analyzes 94% of verbs.
• Tigrinya generator produces correct output for 272 representative combinations of verb root types and grammatical features.