TRAMORPH

                       ITC-IRST - November 1998

                               main.txt


	1. Introduction

The morphological  analyzer is a  tool able to decompose  each Italian
word into its  morphemes and to  give both syntactical information and
transcription for each valid decomposition.

The  morpho-lexicon includes more  than 100,000 entries; each morpheme
belongs to a class and is associated  to its meta-transcription, which
is  an intermediate representation that  can evolve in different ways,
depending on  the adjacent  morphemes.  The morphological  engine  can
recognize an input word by combining morphemes, according  to a set of
concatenation rules.

The morpho-lexicon was obtained by  properly processing the Zingarelli
dictionary, and  adding by  hand  all possible  inflections. This base
lexicon was  then enriched  with   names and  neologisms found  in the
65,000 most frequent words of the newspaper "Il Sole 24 Ore". Also the
most frequent  Italian proper names  and  surnames (from the telephone
directory), geographical names, commonly used foreign words were added
to the  lexicon.  Each morpheme    was phonetically  transcribed   and
manually checked.


	2. TRAMORPH history

	2.1. The morphological tool

TRAMORPH is  a morphological analyzer composed of a morpho-lexicon and
a morphological engine.

The morpho-lexicon includes more  than 100,000 entries; each  morpheme
belongs to a class and is associated  to its meta-transcription, which
is an  intermediate representation that  can evolve in different ways,
depending  on   the adjacent morphemes.  All  meta-transcriptions were
hand-checked.

The  morphological  engine is  capable of  recognizing or rejecting an
input word by combining morphemes, according to a set of concatenation
rules.  The  transcription of  a  recognized  word  can  be  found  by
concatenating  the meta-transcriptions of  the  morphemes found and by
applying a post-processing  to disambiguate the meta-symbols.  In case
of ambiguities, all valid decompositions for a word can be found, each
one with the corresponding transcription.

Particular care was posed  in the design of the  classes, in order  to
minimize the   risk of recognizing  (generating)  invalid  words.  For
instance, due  to  the presence  of   irregular verbs in  the  Italian
language,  more than 60 verbal classes  were identified, each one with
the corresponding set of  inflections (about 50  per class). Nouns and
adjectives   are divided  into about   80 classes,  according to their
inflection set, pos,  gender   and  number.   Overgeneration  problems
suggested not to include  alteration   ("gamba" -> "gambizzare")   and
derivation ("casa" -> "casetta") capabilities in the rule set, as well
as  prefix  handling  ("ri-telefonare").  All  these  items  must   be
explicitly included  in the morpho-lexicon to  be recognized. 

	2.2. The morpho-lexicon

The  first  morpho-lexicon was obtained  by  properly processing about
88,000 items from the Zingarelli dictionary. Each of them was assigned
to a  class, its suffix was removed  (functional words apart)  and its
meta-transcription  was automatically generated and then hand-checked.
The sets of the  valid inflections for each  class was added  by hand.
This process led to about 95,000 morphemes.

This base lexicon was then enriched with names and neologisms found in
the 65,000 most frequent words  of the newspaper "Il  Sole 24 Ore" and
not  recognized by the first  lexicon.   The new words included mainly
neologisms, person names and surnames, company and geographical names,
acronyms, words with prefixes ("mega-centrale"), commonly used foreign
words. The most frequent Italian proper  names and surnames, extracted
from the telephone directory, were also added to the lexicon, which is
actually composed of about 100,700 morphemes.


	3. Warnings

Some words can be pronounced in  different ways, depending on regional
inflections. The most common changes include:  open vs close vowels (e
<-> E, o <-> O) and voiced vs unvoiced fricatives  (x <-> X, z <-> Z).
Sometimes  both pronunciations occur  in the morpho-lexicon, sometimes
no.

Only the  units for the Italian  language were  used to transcribe the
lexicon.   Especially  for  foreign  words,  this fact  leads  to some
awkward but effective  transcription, at least for speech  recognition
purposes.

Despite the  manual check and a number  of automatic and semiautomatic
checks, some error might be still present.