The architecture of the english lexicon

Fill-m, to use a formalism from Prince & Smolensky (1993). Moras required by the lexical entry (via Moraic Prominence), while violating this constraint, will violate it equally for every candidate and thus do not crucially violate it. Only additional moras without lexical segmental correspondents will incur further violations. The concept behind Fill-m conflates *m and Moraic Prominence in this way.

64The exceptional words are almost all in /-ic/, such as hŽretic, and suggest stems like /hŽretik/, without the final /-æ/. Some of these forms have irregular morphology (cf. hŽresy) and could be interpreted as unitary stems.

65‘G’ standing for geminates. As geminates would have no effect on the surface form of monomorphemic bisyllables in this position, it is not possible to say how many items in the set of these bisyllables are lexically geminates, without checking morphologically related forms where the gemination might make itself known. For this table, that proportion is in any case irrelevant.

66At least not under the "strong" variant of OT pursued here, which rejects "co-phonologies" as a device which weakens the theory and simply imports unwanted strategies from derivational theory (see ¤ 1.4.3).

67See ¤ 7.1.1 for ways of treating such exceptions to otherwise general subcategorization patterns in the context of OT.

68Burzio evidently posits for these an empty final syllable, characterized as /â/; thus the (LL) foot is no longer word-final.

69Note that the current hierarchy cannot account for words like br’gand, which will be dealt with below, ¤ 5.3.

70The formalism of the No-Intervening constraint family, which can express references to both constituent edges, allows for the two Alignment constraints given here to be collapsed into one.

71In example (5.7) the difference between the optimal candidate and the ternary foot candidate is of course purely theoretical, as both would surface identically. Differences can be seen in longer words, e.g., (5.8).

72This final syllable can receive a secondary stress in some pronunciations; however it is normally unstressed. For the issue of footed yet stressless syllables, see ¤ 5.4.1, ¤ 6.5.

73While I would prefer a formulation of this constraint which would not include a complex argument like s_mm, I was unable to arrive at any other correct formulation. In any case this complex structure has been used as a constraint argument in other studies such as Sherer (1994), Rosenthall (1994).

74This constraint addresses the same issues as Prince’s (1990: 361) concept of Harmony within the foot, but differs in the more limited range of evaluation results offered here.

75Note that a candidate *(t—_mn_m)æ would violate the highly ranked Final Margins constraint (4.17); this is one factor preventing the mora from linking to the root-final consonant.

76Due to the slightly different nature of the Margin, which can encompass more than one segment, the formulation of Margin Coherence would differ from Moraic Coherence. A plausible formulation would be NI-RL(Seg, M, m).

77Note that in this position, the *m constraint will not be able to eliminate desirable augmented forms such as so³nant (5.24).

78This is of course a theory-internal argument about abstract structure, as the surface forms for both types would be identical.

79While this constraint remains in the hierarchy, it is understood as ranked so low as to have no significant effect on the choice of optimal candidate. Thus it need not be included in the subsequent discussion, as is the case with most constraints.

80Similarly, we have words like (vˆlor)(öz)(a³t)ion, where the lengthened item is a monomoraic suffix (in this case /-iz-/) and the lengthening is also optional. This will be discussed in ¤ 6.2.

81The failure of "level II" suffixes to foot, proposed in ¤ 4.2.1, can be explained by limiting Lapse-s to the prosodic word. Thus, syllables outside the prosodic word do not incur violations of this constraint and do not need to be footed. Requiring feet to align with the right edge of the prosodic word, for example with some kind of Edgemost constraint, would actually prevent these syllables from being footed, as footing them would incur violations.

82It should be clear that members of any constituent hierarchy can act as constraint domains; that is, both morphological and prosodic constituents can be constraint domains.

83Note however that neither Stem-Coh nor m-Coh will prevent true compensatory lengthening within the stem resulting from either segment deletion or shift (see ¤ 5.2.2).

84Another example might be ˆdject´val, with /-iv/ forming a foot of its own due to subcategorization constraints imposed by / al/. This form, along with the similar sbstant´val, are however problematic due to the irregular stress seen in their unsuffixed counterparts (see ¤ 6.6).

85It will be proposed in ¤ 7.5.2 that what are typically referred to as segments should actually be accounted for with two different constituents, morphological segments (or phonemes), which are part of the morphological hierarchy, and prosodic segments, which are part of the prosodic hierarchy. These roughly correspond to the items comprising "input" and "output" strings in Correspondence Theory (McCarthy & Prince 1995). Following this, the behavior of /i/ here could be explained via the projection of a mora by the morphological segment /i/, but its failure to associate to the corresponding prosodic segment /i/, which instead links to the syllable onset.

86Note that this account also assumes that multiply linking /i/ to both the onset and mora constituents is dispreferred; see (5.35) below.

87The interaction of such a constraint and parallel ones taking other features (such as [+high] , [+son] and [+obs]) as arguments might form the formal basis of the sonority-based system of mora association.

88For English, this means /a, e, o/, e.g., pala³tial, amne³sia, emo³tion. Words with orthographic /u/ in this position always surface with the triphthong [Æu·] (e.g., confu³sion), which shows the same quality regardless of the following environment (e.g., confu³se, refu³sal).

89For the words in question with final long syllables, e.g., det‡in, repŽal, prof‡ne, subl’me, the final heavy syllable has the structure -VVC, suggesting a lengthened vowel, i.e., de(ta³n) from /de-tan-æ/.

90Since constituent heads (i.e., strong members) can be referred to in constraints (e.g., McCarthy & Prince 1993a: 14), there is no cogent reason not to also allow reference to weak members. The binary relationship strong/weak can be considered parallel to right/left, both being oppositional features of prosodic constituents. The constraint given here is akin to Hung’s (1994) Rhythm constraint, which implicitly refers to weak members through a grid notation.

91This can be accounted for by an optional constraint which enforces faithfulness for vowel quality features in unstressed syllables. A stronger constraint requiring quality features in stressed syllables is mandatory at all times.

92Many adjectives of this type show alternative forms which take stress on the initial in phrases, e.g., ’nterˆctive tŽlev“sion. This is likely due to a constraint governing the adjacency of Ft’ constituents in the domain of the prosodic phrase (e.g., Align(PrPhrase: Ft’, R; Ft^¡, L) ), which is ranked higher than the P-Edgemost constraint. In isolation such a constraint would be inapplicable, as only one Ft’ constituent would be present in the phrase. Constraints governing supra-word constituents will not be discussed in this work apart from for reasons of clarification.

93The best explanation for this would involve head syllables demanding an association to place features, which /æ/ lacks.

94As in preceding chapters, certain conventions will be used to mark morphological and prosodic structure in candidates. Feet will be surrounded by (), stems by {}, morphological words by [], prosodic words by (). Furthermore, ‘-’ may be used to divide stem syllables from affixes, while ‘+’ indicates a division between morphological words and affixes. The latter conventions are meant to further clarify notationally complex representations, as in (6.3).

95In the handful of words which show penultimate stress, e.g., imprŽgnate, the morphological structure of such words is evidently different, and is parallel to that of the / al/ words, i.e., the suffix subcategorized for the stem rather than the word. See ¤ 7.1.1 for how such cases would be handled lexically within OT.

96The old formulation of the Edgemost constraint, which may be renamed WdEdgemost, is of course retained in the hierarchy, but must be ranked below the new Edgemost. WdEdgemost is relevant where more than one stem is present in a given morphological word; in such cases, the rightmost stem will be expected to hold the main stressed foot (see ¤ 6.5).

97A small minority, e.g., ap—state, insŽnsate, and some variant forms such as intr’cate show a different subcategorization, similar to that for /-al/ suffixes. See ¤ 7.1.1 for treatment of this kind of problem.

98The apparent exceptions cr«a³te, cr«a³tor behave as if analyzed as coming from a stem /cret/, rather than being a suffixed defective stem /*cre/.

99Exceptions which show different stress, e.g., consp’rator, prŽdator, each of which have no corresponding verbal form in /-ate/, apparently show a different lexical subcategorization, perhaps that of the variant of / at/ proposed in ¤ 6.3.

100There are also UK variants sal´vary and advŽrsary, but that is beside the point.

101This is likely to be enforced by a Rhythm constraint of some kind, although not NonFin(s²), which only applies within the morphological word.

102A form *mögra³tory would be prohibited by the different ranking in American English of Ft’ to MWd, see ¤ 6.1.

103There are of course exceptions, e.g., conf’scat˜ry, compŽnsat˜ry, der—gat˜ry, on the one hand, indicating that the distribution is lexical.

104This word is an example of the behavior of an underlyingly long monosyllabic stem, i.e., a stem with a lexically long vowel (/cöt/) rather than a lengthened one. The long vowel always surfaces, e.g., xcöta³tion and the /-atory/ form given here.

105For words that seem to require two morphological words, like compounds or /-ation/ forms (¤ 6.2), the ranking of subcategorization for morphological words by roots or affixes like / ion/ would simply be higher than the constraint which restricts the number of MWords in the PrWord. The subcategorization for /-ory/ would be ranked below the same constraint.

106This depends largely on whether the alternate forms are dialectally different, or equally optional for a given speech community. The Celex database does not contain this information.

107Unless this stem is underlyingly long in some individual’s grammars, and short in others, as in the variants cy³clic Ü cy²clic seen between some speakers (¤ 4.3.3).

108Note that this implies that / æ/ also fails to parse at the prosodic word boundary, as was originally proposed in ¤ 4.1.5.

109This kind of effect could of course be attributed to a constraint which enforces the surfacing of full vowels in strong syllables within the domain of the stem.

110The lack of a long initial seen in parŽntal, as well as the alternation in the unsuffixed form p‡rent Ü pa³rent suggests that this may not be a true / ent/ form but a stem {prent} with a simpler lexical alternant {parent}.

111There are a number of alternative ways to understand these forms. They could be seen as containing an underlying stem-internal /æ/, i.e., /kavælkd/, /hotæntot/. This vowel is unable to take a position as the stressed syllable in the foot (¤ 5.4.3), and so will always be the weak member of a foot when footed. Another possible interpretation would involve projecting a mora from the syllabic sonorants (by Moraic Prominence, as they appear between two less sonorant segments), which acts the nucleus of a weak foot syllable, i.e., (c‡_mvl_m)(ca_mmde).

112On the other hand, any proposed lexical form which is not the best antecedent of the optimal surface candidate, but was rather expanded by default strategies like epenthesis (e.g., if /ktmrn/ was a plausible lexical entry for c‡tamaran), would be replaced in the lexicon by a more optimal underlying form which yields the same surface form (e.g., /katæmæran/). This process is known as Lexicon Optimization (¤ 7.1). The only reason unusual forms like /kavlkd/ would remain in the lexicon is that such a form best yields the surface form c‡valcade, given the English constraint hierarchy.

113This is not a common suffix, but a suffixal identification is useful to explain similar words such as d’scipline Ü d’sciplinary.

114It is true that almost no words of the form *bŽ appear in English, but this would also be the case under the explanation offered previously.

115For example, there are languages with several declensions or conjugations for each syntactic type, each showing particular morphology, stem extensions or vocalisms.

116For example, the compilers of the Celex wordlist (Baayen, Piepenbrock & van Rijn 1993), always chose one such suffixed form as "basic", listing it, suffix included, as a unitary base morpheme.

117 The derived form impr˜vis‡tion indicates the regular nominal structure expected in true / ize/ words, e.g., stˆndardiz‡tion, gneraliz‡tion.

118On the other hand, forms like —rator may be analyzed as based upon a single stem {orat}, despite the forms or‡te, ora³tion. Alternatively, this can be another case of the stem-subcategorizing form of / at/ (¤ 6.3). Either explanation demonstrates the potential idiosyncrasy of lexical selection and subcategorization.

119A rough outline of how this might work is presented in ¤ 7.3.1; a detailed examination of such a semantic hierarchy is however beyond the scope of this work.

120The necessity for semantic, morphological and prosodic constituents is of course applicable only to verbal communication. In a signed language, a gestural hierarchy might be understood to replace the prosodic hierarchy.

121No further decomposition of such lexemes, while plausible, will be entertained in this work.

122In the diagrams found in this section, the braces ({}) represent semantic constituents, rather than stem constituents as in previous chapters.

123In the case of irregular sets of words, such as the strong verbs in English, coequal constraints governing such a set will be reinforced by the frequency of some members of the set (e.g., drive Ü drove), rather than by all members, as some may have low frequencies but maintain themselves on the strength of the group (e.g., strive Ü strove).

124Golston (1995) has also proposed representing lexical entries as specialized bundles of constraints, although in a slightly different context than that presented here.

125The formal details of this and other constraints at the level of segmental phonology will not be delved into here, but will be left for the sake of brevity to further research.

126It is possible to propose a whole series of candidates with incorrect segmentism (perhaps with every possible combination of segments), and this is one way to conceptualize how the segments found in candidates are produced by Gen. All of these candidates, apart from the ones with the "correct" segmentism, will be eliminated by high-ranking lexical selection constraints.

127The details of how this would work are left to further research.

128For example, the alternations seen in run Ü ran, shine Ü shone, could be accounted for via a subcategorization constraint aligning such verbal roots to a preterite morpheme consisting of a vowel subcategorized to appear within the stem. Being ranked higher selectionally than the stem vowel, it replaces it in the input candidate. Regular preterites in / ed/ would not subcategorize for this morpheme.

129Whether this specific constraint overrides the usual subcategorization for /re / or simply presents a different root /record/ (and whether these alternatives actually differ) is best left for future research.

130One way to conceptualize this is to imagine a series of Alignment constraints, aligning a segment to the edge of a lexeme. Ranking these constraints would produce an ordered series of segments (e.g., /cat/ is formed by Align({cat}, R;/t/, R), Align({cat}, R;/a/, R), Align({cat}, R;/k/, R), ranked in that order). A general constraint mapping lexemes to morphological root constituents would then provide the reference point between the edges of abstract lexemic categories and concrete morphological and prosodic constituents. The traditional lexicon is in effect the same thing; this simply defines such a list of ordered and classified segments in terms of OT mechanisms.

131Here "discrete" does not necessarily mean concatenative, but rather an instance in which a morphological constituent /re/ presented to the constraint hierarchy could account for the prosodified surface form, regardless of whether a correspondent phonetic sequence [re] is identifiable in that surface form.

132While OT allows for any configuration of constituents in its constraints, this does not predict that all configurations will be common or that they will occur at random. There are certainly tendencies for certain configurations of constraints to be very common cross-linguistically. However, OT does allow a mechanism by which very "marked" structures which nevertheless surface can be accounted form.

133Alternatively, the constraint can be understood as having an edge argument (L or R), rather than the (L AND R) seen in constraints like Lapse-s.

134These can be understood as abstract elements for notational purposes, but they would directly correspond to measurable acoustic phenomena.

135Similar to morphemes, phonemes would be defined by a series of constraints which place the relevant morphological features within phonemes. One such constraint has been illustrated above, as Voiceless-f (7.14). Phonemes themselves are just labeled morphological constituents; it is the feature content of the phonemes which influences their segmental and prosodic behavior.

136But note that these examples for /-ive/ show an irregular stress pattern and are of uncertain structure.

137Hereon, the initial C of syllable structures like CVCC- are understood to stand for any syllable onset. The final cluster of the root, which forms the onset of the suffix syllable, is likewise not included in the representation of the root syllable.

138Forms like equ‡te, rel‡te, dil‡te , defl‡te, are of course not verbs in /-ate/ but prefixed stems /e-qua³t/, /re-la³t/, /dö-la³t/, /de-fla³t/, and they maintain final stress in American. The word cre‡te, with its unusual stem, has evidently been interpreted in American as a monomoraic stem /kr«a³t/.

139The Celex database does not specify whether these variants are dialectal, or if they are alternate pronunciations in the same dialect.

140While its form suggests a prefixed root, this verb does show a regular nominal impr˜vis‡tion.