Download Analyzing Texts

Analyzing strings into Morphemes
1 Finding a common morph—the Present Tense.
Note: The symbol ‘¢’ may occuyr in play of esh [ʃ]
The first step is to find a common morph in a paradigm. The best
way is to attempt to find the stem if that is possible. We will
work with Russian. Consider the following paradigm, prearranged
in person and then number: The paradigm is of the verb nesti, ‘to
carry (by foot).’ Note: the phoneme /ö/ is always stressed:
Table 1
nesti
gloss
nisú
first person singular
Nisöʃ
second person singular
nisöt
third person singular
nisöm
first person plural
nisöti
second person plural
nisút
third person plural
Look for common form of stem and add tentative morpheme there.
After examining the paradigm, one should determine that the morph
/nis/ forms the root/stem:1
1
See the file on Root, Base, and Stem.
Table 2
nesti
Next
gloss
nis+ú
first person singular
nis+öʃ
second person singular
nis+öt
third person singular
nis+öm
first person plural
nis+öti
second person plural
nis+út
third person plural
consider
the
present
tense
conjugation
of
vezti ‘to carry (by conveyance).’
Table 3
vezti
As in
gloss
vizú
first person singular
Vizöʃ
second person singular
vizöt
third person singular
vizöm
first person plural
vizöti
second person plural
vizút
third person plural
the morph /viz/ is determined, since it is common to all
the forms:
Table 4
vezti
gloss
viz+ú
first person singular
viz+öʃ
second person singular
viz+öt
third person singular
viz+öm
first person plural
viz+öti
second person plural
viz+út
third person plural
Next, try find some more paradigms that are similar
to the one above. For example, vezti ‘to lead:’
Table 5
nesti
gloss
vidú
first person singular
Vidöʃ
second person singular
vidöt
third person singular
vidöm
first person plural
vidöti
second person plural
vidút
third person plural
Similarly, the morph
Table 6
nesti
gloss
vid+ú
first person singular
vid+öʃ
second person singular
vid+öt
third person singular
vid+öm
first person plural
vid+öti
second person plural
vid+út
third person plural
/vid/ is determined:
We have now analyzed three stems: /nis/, /viz/,
/vid/,
each
set
has
a
single
morph
which
may
now
consider an allomorph. But, there is more to come.
1 The Past Tense
Next we will look at the past tense paradigm of all three stems.
In the past tense, number is not marked, but the opposition
gender is marked. Gender is not marked in the plural:
Table 7
nesti
vezti
vesti
gloss
past masculine
nös
vös
völ
nislá
vizlá
vilá
nisló
vizló
viló
nislí
vizlí
vilí
past plural
carry
convey
lead
gloss
singular
past feminine
singular
past neuter
singular
Let’s start with the second row, the past feminine singular. We
find ‘lá’ in all three columns. We fin ‘lö’ in all three columns
of the third row, the past neuter singular, and we find ‘lí’ in
all three columns of the fourth row, the past plural. Note also
that /l/ occurs in the forms except the past tense masculine of
nesti and vezti. This wide distribution should lead us to hypothesize that /l/ is the default marker for the past tense. That
leaves us to explain the absence of /l/ in those two forms.
Assuming this, we will show the morphological analysis of
in :
Table 8
nesti
vezti
vesti
gloss
nös
vös
vö+l
past masculine singular
nis+l+á
viz+l+á
vi+l+á
past feminine singular
nis+l+ó
viz+l+ó
vi+l+ó
past neuter singular
nis+l+í
viz+l+í
vi+l+í
past plural
carry
convey
lead
gloss
Next we should observe that the hypothetical stem
/nis/ occur in three rows of the four rows. In the
first
row,
the
past
masculine
singular,
there
is
a
phonological change in the tentative stem vowel. The
past tense forms shares /n/ and /s/ with the tentative
stem, the only change is in the vowel. It is rather
reasonable to assume that there are two morphs: /nis/
and /nös/. Although there is not enough information to
attempt
to
account
for
this
alternation
now,
it
is
worth noting that it was said above that /ö/ occurs in
stressed positions only. Here, the feature [-Stress] is
in the same column as /ö/; the rule says to replace /ö/
with /i/ if it occurs with the feature [-Stress].
and
are notational variants, because the rule will predict
true results for either
2
or .2 /s/ is consistent for
For a proof of this, we need to go into deep logic.
the
final
consonant
hypothetically
propose
of
the
that
the
root/stem.
We
may
underlying
form
for
nesti is {nös}.
Now let us consider the second column. We find a
similar
pattern, except we encounter a second morph
/vös/. The voicing on the stem changes. We now have two
allomorphs /viz/ and /vöz/. If we selected /z/ as the
underlying form of the root/stem, we can easily predict
that /z/ becomes voiceless at the end of a word: {vöz}
In the third column we find that /v/ is consistent
in
the
onset
position,
and
we
find
the
familiar
alternation of /i/ and /ö/ in the peak position of the
syllable: /vö/ and /vi/. In order to properly analyze
the
underlying
form
for
vesti,
we
must
include
the
present tense allomorph /vid/. Note that /vid/ occurs
when the following segment is a vowel. Note that /d/
does
not
occur
in
the
past
tense,
which
is
always
marked with /l/. And we find the expected alternation
of /i/ and /ö/. We are certain about the first two
segments (the onset and peak of the morpheme): /vöÚ/.
Is
/d/
inserted
epenthetically,
or
is
it
deleted?
Arguments epenthetic insertion include:
(1
/d/ insertion is odd and rarely if ever
found.
) a.
b.
syllable-final
consonantal
deletion
not uncommon (the coda weakens).
is
c.
Historically,
Russian
went
through
an
open syllable phase: every syllable had
to end in a vowel; this included diphthongs.
d.
assume the underlying past tense feminine
form: ##vöd+l+á##.
e.
The old and perhaps the modern rule for
syllabification is to start the syllable
immediately
after
the
vowel
within
a
word:
##$ vö $ dlá $##.
d.
There was a constraint in Indo-European
that blocked an initial syllable cluster
of a dental or alveolar stop plus /l/:
*$ {d,t}lV Ú$.
b.
this constraint forced /d/ to be deleted.
An
epenthetic
inserted,
but
vowel
the
could
Slavic
have
languages
been
did
not take this route.
c.
The morphotactic structure of a root in
nearly every Russian verb is monosyllabic
and
closed—i.e.
polysyllabic
it
has
roots
a
were
coda.
Most
borrowed.
However, stems that in ended in /er/ and
/el/.
Roots
of
epenthetic vowel.
this
sort
added
an
None of these arguments prove that there is an underlying /d/ in
the
root,
but
they
constitute
strong
evidence
for
it
plus
deleting /d/ when it immediately precedes /l/ in syllable initial
position. The three stem/roots for these three verbs are now:
(2)
{nös, vöz, vöd}/.
What is the morpheme representing the past tense in
the
masculine
singular
word
forms?
/l/
occurs
in
/vö+l/, so there is no problem here. What about the
past
masculine
forms
/nös/
and
/vös/?
There
are
a
number of linguistics who abhor ‘º’ as a morpheme. This
used to include me. These days I am getting more and
more interested in set theory. Set theory permits sets
with
small
bits
of
information
which
cannot
be
pronounced. We will write the set as ‘{f}’, where ‘f’
stands
for
a
feature.
The
feature
in
this
text
is
either [+Past] (for past tense) or [-Fem, -Pl]. For
reasons I cannot get into, ‘{f}’ is not the same thing
as ‘{º}’. Here the set has one member—’º’. When a form
is
deleted,
I
will
assume
the
result
is
a
phonologically empty set ‘{}’. Thus, the past tense has
two allomorphs: {/l/}, and {[+Past]}. The morphological
past tense masculine of the three verbs is now written
as:
(2)
nös+{[+Past]}, vöz+{[+Past]},
vö+{[+Past]}.3
3
Similarly, when a phonological segment is deleted, the result is an empty
set in syllable position of the deleted segment. Below we show that /d/ is
deleted before /l/. This would yield the past tense form of vesti: vi{}+l.
I won’t push this analysis here, however.
In
the
past
tense
masculine, if /l/ follows an
obstruent in word final position, /l/ must be deleted,
has Russian has no other alternative: there is no syllabic /l/, and stops with a lateral release (tl, dl) are
not
permitted
in
syllable
final
position.
The
derivation of the 1st person singular and the masculine
and feminine past tense word forms is shown in the
following table:
Table 9
nös
nös+ú
vöz
vöz+ú
vöd
vöd+ú
root/stem
underlying Pres. 1st Pers.
Sing.
nis+ú
viz+ú
vid+ú
vowel reduction = final form
nös+l
vöz+l
vöd+l
underlying Past Masc. Sing.
---
---
vö+l
delete /d/ before /l/
nös+{f}
vöz+{f}
---
delete final l after
obstruent
---
vös+{f}
---
nös
vös
völ
final phonemic form: ‘+’ and
‘{f}’ erased.
nös+l+á
vöz+l+á
vöd+l+á
underlying Past Fem. Sing.
nis+l+á
viz+l+á
vid+l+á
vowel reduction
---
---
vi+l+á
delete /d/ before /l/
note that /d/ was deleted before /l/-Deletion. If the final /l/
were deleted before /d/-Deletion, the wrong from would be the
result:
(2)
vöd+l --> *vöd --> *vöt (word final
devoicing.).
This is an example an extrinsically ordered set of rules. While
intrinsically ordered rules are natural — a rule cannot apply
until its structural description is met — extrinsically ordered
rules are arbitrary as they can occur in either order. I will
introduce
Section
an
alternative
to
extrinsically
ordered
rules
in
below. Devoicing depends on the existence of a word
boundary, which is not phonetic.
This
leaves
us
with
three
tasks.
Accounting
for
the
morphophonemic changes that do occur, and how to analyze the past
masculine singular., and finally, how to analyze the present
tense endings. We will start with the obvious morphophonemic
alternations.
On the one hand, we could list the distribution of the
allomorphs in very general terms. Sometimes, that has to be done.
But, recall, we have already noted that there are phonological
conditioning rules, whether the be derived from an underlying
form,
or
whether
they
are
selected
from
a
set
based
on
phonological contexts. If a phonological context can be found, we
should take advantage of it. One of the objectives linguists try
to do is find an explanation for the things that happen. That
isn’t always possible, but if we can, we want to maintain our
goal.
First, we will give the stress reduction rule for /ö/ in
Russian as the following phonological rule of Russian (which we
give for convenience):
(2)
Stress Reduction I (Russian)
/ö/ --> /i/ / {_______, [-Stress]}.
Here the form enclosed in parentheses indicates a feature and the
position where the feature occurs. This notation is used on
computers and typewriters where vertical writing is difficult.4
Often, you will find the feature and the feature written one over
the other (vertical). This is a notational variant of the one in
. A notational variant refers to rule writing such that if one
variant is true, then the other variant must also be true, and if
one is false, the other must be false. This is a premise from
logic.5 Another notational variant is the following
(2)
Stress Reduction I (alternative, Russian)
{/ö/, [-Stress]} -->/i/††/ ______.
Here, only the relative set of features changes in the context
given. If we combine the two ways of writing
(horizontally and
vertically) and the two was of writing , we get four notating
variants of writing a phonological rule. There are other ways,
but we need not go any further into this here. Stress rules are
morphophonemic in nature: Devoicing depends on the existence of
word boundaries: only one stressed syllable per word. See the
following section.
Let us return to . There are two allomorphs for the morpheme
{NESTI}:6 /nös/ and /nis/. In all forms except the past tense
masculine, the wordforms for this paradigm are bisyllabic and the
stress occurs on the final syllable. In the underlying morpheme
hypothesis, suppose we write the underlying first person singular
as ##{nös}+(ú}##.7 The stress reduction rule is applied:
4
Chinese must be difficult to write on a typewriter as their
system is traditionally vertical
5
6
There are many notational variants used in various logical systems.
We use caps to write the morpheme for the verb and use the infinitival form
to represent the particular verb.
‘##’ represents a word boundary. This form may be and is often shortened to
##nös+u##.
7
##{nös}+{ú}
(2
) c.
d.
##/nis/+/ú/##.
Each form is enclosed in curly brackets: ‘{’ and ‘}’. The curly
brackets represent an underlying morphophonemic form. Applying
the stress reduction rule here creates a phonemic string, with
the curly brackets replaced by the slants.
In the following section we will show that there
are two allomorphs of the past tense morpheme: {/l/,
{}†}.
2 Morphophonemic Rules and Phonetic Rules
There was a successful attempt to merge morphophonemic rules from
phonetic rules. There former deals with rules that are affected
by
morphological
considerations.
The
latter
are
phonetic
by
phonetic rules with no reference to morphological considerations.
In Russian consonants with exception of palatal sibilants are
palatalized before front or palatal vowels:
(2)
C -> Cj / ____ jV.
We can rewrite fronts as
jV:
ji
= /i/,
je
= /e/,
jo
= /ö/,
ju.
=
/ü/, ja = /ä/.
(2)
/njisú/ -> [njjisú]
This transcription system is certainly not standard. It based on
a system used by Dr. Ross Saunders, now retired from S.F.U. [nj]
and [ji] are each distinct segments. This system clearly shows
the assimilative nature of the Palatalization Rule (). This rule
applies unconditionally in reference to morphological properties.
A morphophonemic rule deals with an alternation of forms
that
is
partly
or
completely
dependent
on
morphological
information. For example, the ‘D-Deletion’ rule can occur only if
/l/ marks verbal inflection:
(2)
/d/ --> {} / ____ + [/l/, verbal inflection]
The L-Deletion rule also depends on morphological information:
/l/ must be the verbal inflection marker:
(2)
[/l/, verbal inflection] --> {} / [+Obstruent] +
____##
Phonetic rules do not depend on morphological information.
The Assimilation Rule
is one such example. Assimilation must
apply. The only apparent exception is not really an exception.
The
palatal
consonants
in
Russian
/ß/
and
/dz/
cannot
be
palatalized. The vowel becomes the unfronted counterpart of a
fronted vowel when there is one. I will now propose a filter:
(2)
*/CjV/.
Referring back to example
and the following paragraph the ‘j’
either marks frontedness or the palatal feature of front vowels
or it marks palatalized consonants. It can be segmented with the
vowel or with the consonant:
(2)
The filter
/C//jV/ or /Cj//V/.
refers ti either segmented version in . Russian chose
to avoid the filter by palatalizing a preceding consonant as in .
In the case of the palatal sibilants, the palatal sibilant cannot
be assimilated. In order to avoid violating Filter:
(2)
/ʃje/ --> /ʃe/.
For example, /ßjitj/ ‘to sew’ becomes [ʃit] (or [ʃyt] in the
other transcription). [y] is normally written as the Russian
letter jery ( ы).
2 An alternative to Extrinsically Ordered Rules
Here I will propose an alternative to extrinsically ordered rules
(for Russian, at least). Suppose we start the derivation at the
stem and work outwards in both directions. Here I will discuss
the deletion of dental stops before /l/ and the deletion of /l/
after obstruents and before the empty set {}. I will start with
the past of the lexeme VÖD:
(2)
VÖD+[+Past]+[Masc. Sing. Agr]: ##vöd+ l + {}##.
Start with the stem and apply. Check to see if there is a rule
that deletes /d/ before /l/ + {}. There is. Delete /d/:
(2)
##vö{}+l+{}##
Note that the first empty phonological set {} has no function. It
can be erased. The second one marks masculine singular agreement.
This one should not be erased since it has a function:
(2)
##vö+l+{}##
There is a rule that deletes the phonological content of the past
tense marker, but it must follow an obstruent. It does not.
Therefore /l/ remains. This results in the broad phonetic transcription:
(2)
[völ].
In a somewhat narrower transcription,
(2)
may be rewritten as:
[vjowl].
The vowel is a closed mid vowel with initial fronting (this
causes
palatalization
of
the
preceding
consonant)
and
is
labialized before /l/, which in inherently labialized.
Now consider the masculine singular past tense of the stem
NÖS:
(2)
##nös+ l + {}##.
Starting with the stem there exists no rule which deletes /l/
under any condition. No change. Next move to the next morpheme
/l/. There does exist a rule which deletes /s/ after an obstruent
across a morpheme boundary. Delete the /l/:
(2)
##nös + {} + {}##.
Therefore
(2)
[njos].
The above can be revised in set theory.
2 Simplified Set Theory Approach
In
the
underlying
simplified
forms
but
a
set
set
theory
that
approach,
contains
at
there
the
is
no
moment
two
allomorphs: {/nis/ and /nös/}. In some sense, we have two sets in
a string; the first set includes the verb stem allomorphs and the
second one the first person singular (allo)morph:
(2)
Another
{/nis/ and /nös/}+{/ú/}.
fact
about
Russian.
Only
syllable
in
a
word
can
be
stressed.8
(2)
8
Basic Stress Pattern of Russian Word
Both parts of a compound word can be stressed, one stronger than the other.
Only one syllable of a word may be stressed.
Now, we have to pick on the two allomorphs in the first set. In
the second there is but one allomorph. We select that one first.
Now we return to the first set. In view of Rule , we must select
the unstressed allomorph. If we select the stressed allomorph,
then the word will have two stressed syllables in violation of
Rule . Therefore we get:
(2)
##/nis/+/ú/##.
Using set theory in this way, we don’t have to make a claim that
there
is
an
underlying
morphophonemic
form
which
is
never
pronounced.
Now let us try to determine the allomorphs of the
past tense. Note that /l/ is common to all but two of
the past tense forms; /l/ does not occur in the past
masculine singular of nesti and vezti: /nös/ and /vös/.
There are two allomorphs for vesti: {viz, vös}. /s/
occurs in this paradigm in word-final position. With a
larger corpus it can be shown that voiced obstruents
never occur in word final position.Let us propose that
voiced
obstruents
become
voiceless
in
word
final
position. The proposed rule is:
(2)
Word Final Obstruent Devoicing
[+Obstr] --> [-Voice] / _____ ##.
It turns out that this rule is found not only in Russian, but in
several Slavic languages, German, and other languages in the
world. As a matter of fact, partial obstruent devoicing occurs in
English. For many speakers the devoicing process starts during
the production phase of the final obstruent. For some speakers it
starts a little sooner. However, the rule that lengthens a vowel
before voiced segments in English remains. In this way the distinction between voiced and voiceless obstruents in word final
position is maintained even though the vocalic lengthened bears
the function now.
In terms of set theory we have two allomorphs for
vezti
mentioned
above.
Because
of
the
constraint
against voiced obstruents in word final position, the
allomorph with the voiceless obstruent is selected when
no suffixes follow:
(2)
{/vöz/, /vös/}##.
Select the voiceless one:
(2)
/vös/##.
The analysis of the past tense ending presents a
bigger problem. In
as we have noted /l/ does not occur
in the masculine singular forms of two verbs, though
/l/ occurs in the past tense of vesti. Note that there
are three allomorphs:
(2)
{/vid/, /vö/, /vi/}.
Let us start with the first segment. Note that it is consistently
/v/. We may posit /v/ in the underlying form. In the vowel
position we find /ö/ and /i/, which is unstressed. Based on the
alternations discussed above, we may posit /ö/ in the underlying
form.
The fact that we are claiming that there are two
past tense allomorphs: /l/ and [+Past] shows that the
phonological approach to morphology is less than ideal.
But we will have to let it stand for the moment.
2 The Present Tense Morpheme
Now I turn our attention to the present tense morpheme. Note that
the endings for the 2nd person singular, third person singular,
first person plural, and the second person plural all share the
phoneme /ö/. Since there is a slot for the past tense, it is not
unreasonable to look for one in the present tense. /ö/ is a
reasonable candidate for this position. The four forms may be
analyzed as:
2
nesti
vezti
vesti
nis+ö+ʃ
viz+ö+¢
vid+ö+¢
2nd pers. sing.
nis+ö+t
viz+ö+t
vid+ö+t
3rd pers. sing.
nis+ö+m
viz+ö+m
vid+ö+m
1st pers. plural
nis+ö+ti
viz+ö+ti
vid+ö+ti
2nd pers. plural
1.5 Another Approach to Eliminating Extrinsically Ordered Rules
2 The First Person Singular and the Third Person
Plural Ending
The endings /ú/ and /út/ marking the first person singular and
the third person plural, respective, are the same for all three
verbs. Here, an empty set is not the best analysis for these
endings. The approach which I favour is to assume that there is
but
one
morpheme
here
and
that
this
morpheme
contains
the
features marking present tense, first person singular or the
third person singular:
2
nesti
vezti
vesti
nis+ú
viz+ú
vid+ú
1st Person Singular
nis+út
viz+út
vid+út
3rd Person Plural
We will analyze the past tense and non-progressive participles of
English in a similar fashion.9
A morpheme contains a bundle or set of grammatical
features. Grammatical features may occur anywhere in
the word or even in a clitic. We won’t cover clitics
until later. First, let us suppose that in the first
morpheme
following
the
verb
in
the
first
person
singular and the third person plural forms contain the
features [-Past], [-2nd], +[Pers],
[-Pl]. This bundle of features is a set, and it is spelled out as
the phoneme /ú/:
(2)
{[-Past], [-2nd], [+Pers], [-Pl]} = /ú/.
A similar set (morpheme) exists for the third person plural
9
in set theory two sets overlap, some members of each set are shared by both
sets. If both sets complete overlap, then the members of each set belong to
ending /út/:
(2)
{[-Past], [-Pers], [-Pl]} = /út/.
Let us number the position for the stem as ‘1’, and the suffix
that immediately follows it as ‘2’. /ú/ and /út/ both occupy
position 2.
Now
let
us
suppose
the
feature
bundles
split
into
two
morphemes or sets in the remaining forms of the present tense. We
will
assign
tense
to
position
‘2a’
and
agreement
to
‘2b’.
Agreement includes those features which are subject to agreement
rules with the subject of the sentence: person, number, and
gender:
(2)
Splitting in the Russian Verb
[2
Tense,
Agreement]
-->
[2a
Tense]
+
[2b
Agreement].
Note that all the past tense endings split this way into two
morphemes via Rule : {nös+l+{}}. {nös+l+a}, and so forth. Let us
now do a sample derivation of the second person singular form
/nisö¢/. First we start with the underlying form:
(2)
{nös+[2 [-Past], [+2nd], [+Pers], [-Pl]]}
The splitting rule applies here creating two morphemes, the first
for tense, the second for agreement:
(2)
{nös+[2a [-Past]] + [2b [+2nd] [+Pers], [-Pl]]}
The first morpheme is spelled out as /ö/ and the second morpheme
as /¢/ or /S/.
2 Sets and Filters--Optimality Theory?
both sets simultaneously. This is in effect one set. This is what is
Let me turn our attention to sets. Just a few words on sets. A
set may contain one or more members. There is no limit. And
combinations of objects may form a set, though many of these
would be rather useless. In morphology we will limit the sets to
morphological grammatical features and to phonological features.
Phonological sets are similar to phonological rules, but without
abstract
underlying
forms.
What
I
am
doing
appears
to
be
a
variant of Optimality Theory. I return to this below.
This is one of the problems we run into when we try
to define morphemes in terms of phonemes rather than as
bundles or sets of grammatical features. Applying the
splitting rules
and then the spell-out rules as indi-
cated. We won’t go into doing the above rules here as a
class
exercise.
What
we
will
do
is
set
up
an
equivalence relation:
(2)
Correspondences for the Russian Verb
{[-Past], [-2nd], [+Pers], [-Pl]} /ú/
{[Past], [+Pers], [-Pl]} /út/.
{2b [-Past], [+2nd, [+Pers], [-Pl]}  ö+/ʃ/
{2b [-Past], [-Pers], [-Pl]}  /ö+t/
{2b [-Past], [-2nd], [+Pers], [-Pl]}  /ö+m/
{2b [-Past], [+2nd], [+Pers], [-Pl]}  ö+/ti/
The
first
correspondence
that
the
first
person
singular
is
spelled out as /ú/, and the second as /út/.
These correspondences also work in reverse. If you find
happening here.
{/ú/}, then you know you have the first person singular form.
Things get more complicated as more text comes in, but we needn’t
worry about it now. Correspondences are simpler than a rule
format which I won’t cover here at this time.
Now I will focus on sets and how they the analysis may be a
form of Optimality Theory. Let us start with the past tense
morpheme {l} which has two allophones:
(2)
{/l/, {}} = P
P is a set containing two allomorphs, which form
the set [+Past]
(2)
{/l/, {}} ¯_ [+Past]
The grammatical feature [+Past] corresponds with
the set P.
(2)
{[+Past], [-Past]} = T(ense)
The two members of Tense, abbreviated to ‘T’,
are [+Past] and [-Past].
We
now
need
constraint:/l/
That
can
is
occur
true,
because
almost
of
the
distributional
anywhere,
but
{}
word
final
position—specifically when it follows a consonant. Let us set up
constraint or filter for the empty allomorph:
(2)
Final ‘d+l’ Filter
*/d+l/ _____ ##.
This filter is similar to the rule which deletes /d/ after ‘C+’
before a word boundary:
(2)
‘D’-Lateral Deletion
d_{} / ____ + /l/ ##
What about the deletion of the lateral? It is deleted after an
obstruent:
(2)
Lateral Deletion
/l/ _ {} / [+Obstr] ____ ##
The equivalent filter is:
(2)
Final Lateral Filter
*/l/ / [+Obstr] ____ ##
Now, let us see how the filter works. Remember, we are assume the
variant of Optimality Theory which permits morphemic sets and the
selection of an allomorph. There are three allomorphs for the
verb vezti ‘to lead’:
(2)
{vö, vid, vi}.
Before
we
can
analyze
this,
we
also
have
to
consider
phonemic sets and their filters. Of concern to us here is the
alternation of /ö/ and /i/. Recall that the former may occur only
in
stressed
positions.
We
have,
therefore,
the
phonological
alternation set:
(2)
{ö, i}.
These used to be called morphophonemes. Let us redefine the
morphophoneme:
(2)
Morphophoneme
An morphophoneme is a set of two or more
phonemes which predictably alternate in a
lexeme.
Let
us
represent
the
morphophoneme
in
CAPS,
and
select
the
allomorph which is unique. /ö/ occurs only in . /i/ occurs in
other morphophonemes: {é, i}, {ä’, Å} (ä’ is stressed). Let us
abbreviate the set
(2)
to be written as {Ó}.
Ö = {ö, i}.
In a similar fashion we can write the morphophoneme {d, {}} as
{D}, and morphophoneme {l, {}} as:
(2)
D = {d, {}}.
L = {l, {}}.
/v/ so far has no alternate phonemes. We will write as {V}.
V = {v}.
No morphophoneme is complete until all the data is in.
Another filter is required. Only one syllable may
be stressed in the Russian word.10
Now we can set up the lexeme for the Russian verb
under consideration: VöD. /v/ is the only member V, /ö/
is the unique member of {Ó}, and /d/ is the unique
member of {D}.
Table 10
10 Compounds are made out of two word.
stressed, one stronger than the other.
Each
part
of
a
compound
may
be
{VÓD + L+
The morphophonemes of the lexeme + tense
‘A}
+ agreement
vÓD + L +
intermediate stage, /v/ is selected from
‘A}
V (first cycle)
viD+ L +
intermediate stage, /i/ is selected from
‘A}
Ó (first cycle)
vi + L+ ‘A
vi + l + á
vilá
intermediate stage, /{}/ is selected from
D (first cycle)
*vöD+L+’A
*vid+L+’A
intermediate stage, /l/ and /á/ are
selected from L and A
erase ‘+’.
all the above rules are morphophonemic
rules.
vjilá
assimilatory palatalization, phonetic
rule.
The above represents an over simplification that is not so
simple as it appears. As it stands, the selection of the correct
phoneme has to be selected in a certain order. For example, in
the past singular masculine, ordering is crucial. If /l/ is
selected from {L}. first before Final Lateral Filter applies,
then /l/ is deleted because it follows /d/ at the end of the
word: /vid/. This is the incorrect outcome. The elimination of
ordered selection is undesirable. Furthermore, the rule is not
truly phonologically conditioned because the syllable initial
sequence in Russian is now permitted as in dlja ‘for.’
Recall that there is morphological conditioning. In English
the feature [+Past] conditions certain alternations of irregular
verbs. Let us see if the same might hold in Russian. The only
place where we find the morphophoneme D is in the past tense
alternations. Suppose we replace the /l/ in D-Lateral Deletion
with the feature [+Past]:
(2)D-Selection (replaces D-Lateral Deletion)
*{D} +[+Past]
First, note that the context for the rule simplified. We need
specify the grammatical feature [+Past] only for the context of
the alternation. Note that the grammatical feature is always
present when generated. The set which includes [+Past] includes
phonemic and morphological information:
Table 11: The Grammaticon of Russian Entry [+Past]
properties
types
[+Past]
conceptual
+Suffix,
morphological
position 2
{L = /l/, {}}
morphophoneme
The conceptual box directly relates to the semantic field (the
eventuality
takes
place
before
the
speech
event).
The
morphological field marks its form: a suffix which occurs in
position
2
inflectional
(position
affixes).
1
is
The
for
the
lexeme,
morphophonemic
position
field
2
for
contains
the
appropriate phonological information of the form.
The for selecting the appropriate phoneme of [+Past] is
revised:
(2)
{L} Selection (replaces Lateral Deletion)
*{[+Obstr]}+{/l/}##.
If the morphophoneme D is present, /l/ cannot be selected. The
rule
is
based
on
the
morphophoneme
D,
not
on
its
phonemic
members. This rule will not block the existence of /dl/ in dlja,
since there is no morpheme boundary between /d/ and /l/. In a
cyclical application, start with phonemes in the stem. The rules
can look outward, but not inward. {D/º}-Selection applies in the
first cycle since {D} occur in the first cycle, the stem, and it
can look outward to {/l//, [+Past]}.
It is also possible to write the rule such that {L} must
occur in a suffix, but not {D}. This is tantamount to the above
rule, since the morpheme boundary ‘+’ ensures that the two morphophonemes are not in the same morpheme. The way it is written
above in
is a simpler notational variant.
Now we can claim that D-Past Selection is now unordered with
respect to L-Selection. D-Past Selection is dependent on the
feature [+Past], which is always there, while the choice of /l/
in L-Selection is dependent on a stem final obstruent and a word
boundary.
This
does
represent
another
alternative
besides
cyclical application. Which method is correct remains to be seen.
Let us perform some tests to see whether these filters
really work. Suppose we generate */vöd/ or */vöt/ (applying the
Final Obstruent Voicing Filter. If this form is interpreted as a
past tense masculine singular, it is rendered ungrammatical by
the D-Past Selection filter. Should there happen to be a noun
#/vöt/11, it would not violate the D-Past filter, since nouns
cannot be marked with the grammatical feature [+Past].
11 The ‘#’ is used to here to indicate that such a noun does not exist in the
standard literary language.
Definitions
Analysis and Rules 1
Analysis and Rules 3
This page last updated 8 MR 2005