:construction: This readme.md file is under construction.
The MWlexnorm repository provides material for generation of declensions
for the nominal forms implied by the Monier-Williams Sanskrit-English Dictionary
of 1899. It is based on the digital edition of this dictionary available at
the Cologne Sanskrit Lexicon website.
This work was begun approximately 2008 by Jim Funderburk, Malcolm Hyman, and
Peter Scharf. The current form of the work as represented in this repository
is due to Funderburk.
Statement of the problem
Consider a particular example, the word guru.
Here is how the underlying printed edition looks for the first part of the entry for guru:
Here is a link to a display of guru from the Cologne digital
edition.
For our purposes here, we are interested in only a small part of the
information about this word available in the dictionary:
गुरु mf(वी)n.
This tells us that guru is a Sanskrit word with inflected forms in each of
the three genders (masculine, feminine, and neuter).
When we look at part of the underlying xml structure of the digitization upon
which the display is based, we see the following arcane structure:
From a small familiarity with Sanskrit grammar, we can infer three
pieces of information regarding the declension of guru:
guru can be declined as a masculine noun ending in u, similar to
a model such as taru (tree, Deshpande, p. 79)
guruH, gurU, guravaH being the nominative case forms in the singular,
dual, and plural numbers, respectively
NOTE 1: These words are spelled using the SLP1 transliteration of Sanskrit.
Note 2: Madhav M. Deshpande, A Sanskrit Primer, 2003.
Note 3: We can think of guru as the stem and _mu as the model.
guru can be declined as a neuter noun ending in u, based on a model
such as maDu (honey, Deshpande, p. 113)
Note: We can think of guru as the stem and _nu as the model.
guru can be declined as a feminine noun. However, instead of being
declined like a feminine noun ending in u, such as Denu (cow, Deshpande, p. 80), it is declined like a feminine noun ending in long i (I), such as
nadI (river, Deshpande, p. 73).
Note: We can think of gurvI as the stem, and _fI as the model.
From this example of गुरु mf(वी)n, we have inferred, in a systematic way,
the stem and model according to which the declensions of this word may be
computed.
The purpose of this repository is to write computer programs that carry
forth this systematic process of deriving stem and model for all the
nominal (and indeclineable) forms in the Monier Williams dictionary.
Summary of the general method
We have divided this stem-model derivation process into three steps; each
of the steps is carried out by programs in the three subdirectories
step0, step1a, step1b.
step0: lexicalgrammar.xml
This step identifies each record in mw.xml whose markup has information in
the <lex> tag, and summarizes this information in a structure form within
the lexicalgrammar.xml file. For our example of guru, this information
looks like:
This step normalizes the information of the lexicalgrammar.xml file.
Basically, the information of the lexicalgrammar.xml file is complicated to
parse. So a simpler, more uniform representation is desireable for further
computation.
For our guru example, here is the simplification present in lexnorm.txt:
65987 guru guru/ m:f#vI:n
We present the record number (65987), the headword (guru), the extended
headword (guru/, this is the dictkey2 field of lexicagrammar.xml), and a simplified
version of the dictlex field (m:f#vI:n).
For the special categories (pronouns, cardinals, dual-plurals, present and
future participles), the simplified records are written to the lexnorm-other.txt
file.
For examples:
As mentioned in the guru example, step1b carries out the specification
of stem and model; that's what is done by step1b.
The results are in the stemmodel.txt file for most cases. For cases of
pronouns, etc., the results are in the stemmodel_other.txt file. The
stemmodel_log.txt file has a couple of instances where we aren't sure how
to carry out the process.
Here is how our guru records appears in stemmodel.txt:
MW-0065987.00 guru <f>guru m_u</f><f>gurvI f_I</f><f>guru n_u</f>
The third field contains the stem and model information present in this
record.
Note that we have computed the implied stem gurvI for the feminine form,
based on the 'f#vI' information of lexnorm.txt.
There are many other similar subtleties involved in the transition from
lexnorm.txt to stemmodel.txt. Here are a few other examples, where we
show the lexnorm.txt information and the corresponding stemmodel.txt data.
32 aMSahArin a/MSa--hArin m:f:n
MW-0000032.00 aMSahArin <f>aMSa-hArin m_in</f><f>aMSa-hAriRI f_I</f><f>aMSa-hArin n_in</f>
45 aMSin aMSin m:f:n
MW-0000045.00 aMSin <f>aMSin m_in</f><f>aMSinI f_I</f><f>aMSin n_in</f>
An n->R sandhi routine applies to the last pada (as determined by key2) to
decided wheter the feminine stem for a noun ending in 'in' should end in
'RI' or 'nI'.
27 aMSaBUta a/MSa--BUta m:f:n
MW-0000027.00 aMSaBUta <f>aMSa-BUta m_a</f><f>aMSa-BUtA f_A</f><f>aMSa-BUta n_a</f>
For the very common case of a headword ending in short 'a' and with the 'mfn'
lexical information (typically, an adjective), the feminine stem is computed
by replacing the final 'a' by 'A', with model such as *latA*.
65 aMSumat aMSu/--ma/t m:f:n
MW-0000065.00 aMSumat <f>aMSu-mat m_mat</f><f>aMSu-matI f_I</f><f>aMSu-mat n_mat</f>
We assign the model 'm_mat', in the belief that a special declension model
is applicable. We also compute the feminine stem to end in 'matI'.
103832 namat namat m:f#ntI:n
MW-0103832.00 namat <f>namat m_mat</f><f>namantI f_I</f><f>namat n_mat</f>
Here the feminine stem ends in 'antI', based on the information 'f#ntI'
provided by the MW dictionary for namat.
The reason for preserving the last pada designation, as implied by MW key2,
also has to do with n->R sandhi that will occur in the process of declension.
(I am thinking of cases like 3s of rAma, which is rAmeRa rather than rAmena
due to n->R sandhi.) For instance consider trayImaya:
MW-0087487.00 trayImaya <f>trayI-maya m_a</f><f>trayI-mayI f_I</f><f>trayI-maya n_a</f>
The 3s of trayImaya would be trayImayeRa if 'trayI' and 'maya' are considered
in the same pada; but would be trayI-mayena if the 'r' of 'trayI' is in a
different pada, as MW information provides.
current limitations of the 'model'
In terms of number of headwords involved, the model designations in
stemmodel.txt are adequate. This is because most nominals have a stem
ending in a vowel and have declined forms following a simple model based
on the ending vowel (along with the gender).
However, there are significant numbers of important, frequently used
nominals that end in a consonant. For these consonant ending nominals, the
current model designations in stemmodel.txt must be viewed with scepticism
until the associated declension algorithms have been implemented.
A further caveat is in order regarding nominals whose declensions are
typically regarded as 'irregular'. Two examples that come to mind are
strI (woman) and its compounds, and DI (thought) and its compounds.
Currently, these appear in stemmodel.txt with the model 'f_I':
MW-0101266.00 DI <f>DI f_I</f>
MW-0254928.00 strI <f>strI f_I</f>
which implies that they are declined like nadI. However, this is wrong.
The identification of these irregular cases remains to be done, along with
a systematic means of specifying their models so that an ensuing
declension engine will know how to generate correct declensions.
:construction: This readme.md file is under construction.
The MWlexnorm repository provides material for generation of declensions for the nominal forms implied by the Monier-Williams Sanskrit-English Dictionary of 1899. It is based on the digital edition of this dictionary available at the Cologne Sanskrit Lexicon website.
This work was begun approximately 2008 by Jim Funderburk, Malcolm Hyman, and Peter Scharf. The current form of the work as represented in this repository is due to Funderburk.
Statement of the problem
Consider a particular example, the word guru.
Here is how the underlying printed edition looks for the first part of the entry for guru:
Here is a link to a display of guru from the Cologne digital edition.
For our purposes here, we are interested in only a small part of the information about this word available in the dictionary:
गुरु mf(वी)n.
This tells us that guru is a Sanskrit word with inflected forms in each of the three genders (masculine, feminine, and neuter).
When we look at part of the underlying xml structure of the digitization upon which the display is based, we see the following arcane structure:
From a small familiarity with Sanskrit grammar, we can infer three pieces of information regarding the declension of guru:
From this example of गुरु mf(वी)n, we have inferred, in a systematic way, the stem and model according to which the declensions of this word may be computed.
The purpose of this repository is to write computer programs that carry forth this systematic process of deriving stem and model for all the nominal (and indeclineable) forms in the Monier Williams dictionary.
Summary of the general method
We have divided this stem-model derivation process into three steps; each of the steps is carried out by programs in the three subdirectories step0, step1a, step1b.
step0: lexicalgrammar.xml
This step identifies each record in mw.xml whose markup has information in the
<lex>tag, and summarizes this information in a structure form within the lexicalgrammar.xml file. For our example of guru, this information looks like:Most of the records like like this, but for a few subcategories, we have added additional information. Here are examples.
step1a: lexnorm.txt and lexnorm-other.txt
This step normalizes the information of the lexicalgrammar.xml file. Basically, the information of the lexicalgrammar.xml file is complicated to parse. So a simpler, more uniform representation is desireable for further computation.
For our guru example, here is the simplification present in lexnorm.txt:
We present the record number (65987), the headword (guru), the extended headword (guru/, this is the dictkey2 field of lexicagrammar.xml), and a simplified version of the dictlex field (m:f#vI:n).
For the special categories (pronouns, cardinals, dual-plurals, present and future participles), the simplified records are written to the lexnorm-other.txt file. For examples:
step1b. stemmodel.txt, etc.
As mentioned in the guru example, step1b carries out the specification of stem and model; that's what is done by step1b.
The results are in the stemmodel.txt file for most cases. For cases of pronouns, etc., the results are in the stemmodel_other.txt file. The stemmodel_log.txt file has a couple of instances where we aren't sure how to carry out the process.
Here is how our guru records appears in stemmodel.txt:
The third field contains the stem and model information present in this record.
Note that we have computed the implied stem gurvI for the feminine form, based on the 'f#vI' information of lexnorm.txt.
There are many other similar subtleties involved in the transition from lexnorm.txt to stemmodel.txt. Here are a few other examples, where we show the lexnorm.txt information and the corresponding stemmodel.txt data.
current limitations of the 'model'
In terms of number of headwords involved, the model designations in stemmodel.txt are adequate. This is because most nominals have a stem ending in a vowel and have declined forms following a simple model based on the ending vowel (along with the gender).
However, there are significant numbers of important, frequently used nominals that end in a consonant. For these consonant ending nominals, the current model designations in stemmodel.txt must be viewed with scepticism until the associated declension algorithms have been implemented.
A further caveat is in order regarding nominals whose declensions are typically regarded as 'irregular'. Two examples that come to mind are strI (woman) and its compounds, and DI (thought) and its compounds. Currently, these appear in stemmodel.txt with the model 'f_I':
which implies that they are declined like nadI. However, this is wrong.
The identification of these irregular cases remains to be done, along with a systematic means of specifying their models so that an ensuing declension engine will know how to generate correct declensions.