The architecture of the english lexicon

6.6 Multiply suffixed words

This section looks briefly at how multiply suffixed words illustrate the interaction of the various subcategorization constraints presented above. Different suffixes present constraints which are differently ranked, with the result that the signature effect of various affixes fail to appear in some cases. A number of such cases were discussed above regarding the affixes subcategorizing for the morphological word, but similar effects are seen for affixes in the / al/, / ent/ and / ity/ classes.

The / al/ class suffixes subcategorize for stems, which means that other suffixes can themselves be "stem-ized" when suffixed by this type.
(6.44) “ncremŽntal co“ncidŽntal momŽntous

trrit—rial ˆctu‡rial sens‡tional

Words like momŽntous and co“ncidŽntal illustrate how / ent/ can become stem-ized when followed by / al/ class suffixes. Note that the / ent/ suffix itself still requires a stem, so that the structures involved are [{mom}{ent}ous], [coin{cid}{ent}al]. The root stems behave as expected, augmenting when necessary to satisfy stem constraints (i.e., m¯mŽntous), but remaining unstressed due to their position before the main stress.¹¹⁰ The behavior of the suffixes / ary/ and / ory/ when suffixed with / al/ indicates that the ranking of the subcategorization constraint of / al/ for the morphological word on its right is higher than that of / ory/ and / ary/ for the morphological word on its left:
(6.45) /-al/ Align( /-al/, R; MWd, R)

/-ory, -ary/ Align( /-ary/, L; MWd, R)

The structures that come out of this conflict appear to be ones like [{territ}{ori}al], [{actu}{ari}al]. Cases where word-subcategorizing suffixes like / ary/ are added to stem-categorizing suffixes show the same prosodic structures as in unaffixed instances, as expected, e.g., sŽdentˆry with a plausible structure [{sedd}ent]ary.

Words in /-ion-al/ are also of interest, as the suffix /-ion/ retains its position in the main-stressed foot, despite the addition of /-al/ (where we might expect *snsatio³nal). However, the suffix / ion/ is an / ity/ class suffix and therefore part of the stem; the structure of such words is likely to be similar to [{sens}{atÆon}al], the bisyllabic suffix / ation/ not showing any prosodic changes when suffixed by / al/, since it is already a stem. As was noted above (¤ 2.2.5), the contrast between na³tion and n‡tional is idiosyncratic and purely lexical, as most words in / ation-al/ do not show any vowel alternation.

Cases of sequences of / al/ class suffixes are difficult to identify; possibilities include ˆdject´val, sbstant´val, which would show the monomoraic suffix / ive/ being augmented as a stem in itself. The difficulty with these forms is that the simplex forms ‡djective, sœbstantive unexpectedly fail to show stress on heavy penults, and thus have an idiosyncratic structure. At odds with these forms is fŽstive, which itself is problematic and appears to be a single stem {festiv}. Words which show apparently incorrect subcategorization such as these will be handled lexically, and are discussed in ¤ 7.1.1.

Cases where / al/ and / ent/ suffixes are extended by / ity/ class suffixes are more commonplace. In all cases, the subcategorization for the / ity/ class suffixes to take a morphological word on their right and to form part of a stem with the morpheme on their left overrides all other subcategorization constraints, and the constraints governing these affixes must be the most highly ranked of the types discussed here:

(6.46) rcept’vity ˆbnorm‡lity voc‡lic
Again, the / al/ class take stems on their left; the / ity/ class suffixes force them to become part of stem that includes the suffix. The morphological word boundary, however, is at the right edge of the word, enforced by the subcategorization of the / ity/ class. Thus the structures created are [re{cept}{ivity}], [{voc}{alic}æ] rather than *[re{cept}{iv]ity}, which would produce the incorrect result. The subcategorization of / ity/ for a morphological word on the right outranks that of / al/, and other constraints evidently enforce the restriction of a single morphological word constituent to the candidates. Only in certain situations, such as in compounds (¤ 5.4.2), do constraints enforcing the presence of multiple word constituents allow such structures to surface.

Similar to the multiply suffixed words taking the / al/ series are the stressed suffixes, such as / ese/, / ee/, / eer/, as in chöne³se, c˜mmande³er, d“vorce³e. Such suffixes will be understood as forming stems and lying within the morphological word. The simplest way to account for such structures is to regard these affixes as themselves suffixed in / æ/. Thus, the structure of such words would be as follows:

(6.47) [{chin}{es}æ] [com{mand}{er}æ] [di{vorc}{e}æ]
The suffix / æ/ subcategorizes for a stem constituent, as do the other / al/ class suffixes. The stressed suffixes would also subcategorize for stem constituents, but would be suffixed themselves, producing a situation in which they were stressed. The stem syllables immediately preceding them would be unable to take stress by NonFin(s²), and could appear either as unstressed feet or the unstressed members of bisyllabic feet, as in the examples above. Stems requiring feet would be augmented although not stressed, as in chöne³se. As with most unstressed feet that are nevertheless part of a stem, these stressless feet are not reduced, but display their long vowel. One interesting word in /-ese/ is j‡panese, whose stress pattern conflicts with that of the simplex jap‡n. The latter shows a stress pattern which would be ascribed to Kager’s third, idiosyncratic group (¤ 3.1.2). This suggests a morphological structure [{japann}æ], with both geminate and final suffixal schwa. Suffixing the stem {japann} with /-ese/ will produce the attested form, as the geminate syllable would fail to take stress due to NonFin(s²), and the usually dispreferred foot type (j‡pann) would nevertheless be the most optimal. Another interesting form is appointe³e, which must have the prosodic structure [æ(poin)(te³)æ]. The prefix /æ-/, proposed above in ¤ 5.4.3 to account for prefix syllables that are never stressed, fails to form a foot *(æ²point), even when this would be otherwise favored by the constraint hierarchy. Instead, the stem syllable {point} remains a monosyllabic foot, even though it is adjacent to the main stress.
6.7 Retraction

Kager (1989) and others have described a series of phenomena referred to as stress retraction, which was presented in ¤ 3.1.4. Many words ascribed to this category are completely regular according to the general stress pattern as analyzed here, and were only described as retracted due to theoretical assumptions within the derivational framework. For example, words of the following kind, showing so-called Strong Retraction, were thought to warrant a final stress because of the weight or length of their final syllables. The main stress in such words, which actually appears on the antepenult, was thought to be retracted via a serial derivational process:

(6.48) ‡bsolte m‡nifst hœrricˆne

t‡citrn p‡rad“se d’fficlt

Preventing the final syllable from receiving main stress, due to the extrametricality effect (enforced by NonFin(Ft’)) yields the regular pattern [(s²_m(m)s_m)(s_mm)] for most of these forms. The final feet in question, belonging to the stem constituent, may not take the main stress but are nevertheless footed, satisfying various constraints such as Edgemost, Stem-Ft-R and Lapse-s.
(6.49)

difficult /dificult/	Non-Fin(F’)	Non-Fin(s²)	Edgemost	StemFtR	Lapse-s¡	FtForm	Lapse-s
+ (d’mfim)(cumlm)t					s
(d’mfim)(cmlm)t		!*
(d’mfim)cumlmt			!s	s			sss
dim(f’mcumlm)t	!*					*	sss
(d“mfim)(cœmlm)t	!*	*

The theories cited by Kager (1989) in his analysis did not have a mechanism by which a syllable could be footed and yet not automatically receive the promotion to head foot given by End Rule: Right (apart from retraction, which was applied afterwards to undo the effects of the End Rule). While the (potentially) full vowels of the final syllables in these forms prompted earlier theorists to posit secondary stresses, here these syllables are considered to be stressless feet (¤ 5.3, ¤ 5.4.1). The characteristic sign of stressless feet is the optional reduction of short vowels in closed stem syllables (e.g., d’ffic[æ]lt); long vowels in the same position can also reduce in some dialects (e.g., hœrricane [hó²ríkæn]).

So-called Weak Retraction similarly follows the usual group I stress pattern:
(6.50) e(lŽc)tr˜de sta(l‡g)m“te py(r‡mi)d˜id
Again, the final heavy syllables seem to form a monosyllabic foot. Words belonging to this group show an set of infrequent suffixes such as / oid,  ite,  ide/. A few word pairs, like sphe³roid Ü sphŽrical, suggest that these words have a similar structure to those in /-ent/, enforcing the bimoraic stem in such cases via stem augmentation (i.e., [({sfe³)(r}oid)] ). However, there are too few examples to determine the further behavior of these suffixes, and while / oid/ is quite clearly a suffix, other endings like /-ite/ or /-ode/ which can be grouped into this category are morphologically less clear. Monomorphemic words ending in long vowels show similar stress patterns.

The words described as undergoing Sonorant Retraction are numerous, and this retraction mode appears to contradict the regular stress pattern, as an apparently heavy, closed syllable is not footed but is rather skipped over:

(6.51) hŽlminth˜id c‡valcˆde ‡rgent“ne h—ttent˜t

H‡ckensˆck p‡limpsst ‡lgern˜n

These can be contrasted with words whose penults are closed by non-sonorants, as in (3.11). However, the fact that sonorants figure in the relationship recalls the solution to a different problem. There are a number of words like ‡labˆster, c‡taclysm, rŽgister which appear to show unusual stress, disregarding a stressable penult and rather displaying main stress on the antepenult. All of these forms end in surface syllabic sonorants, and the best explanation of this type involves regarding this final "syllable" as a final margin constituent, structurally similar to an onset cluster (¤ 4.1.3):
(6.52) [(‡la)(bastr)] [(k‡ta)(klizm)] [(rŽgistr)]
A similar explanation has been used to account for the apparent monosyllabicity (in the context of derivational theory, cf. ¤ 6.3) of the suffixes / ary,  ory/ by Chomsky & Halle (1968), Liberman & Prince (1977), Kager (1989), among others, who suggest an underlying structure like / orÆ/, / arÆ/ for these affixes. In phonological terms, syllabic resonants of the kind proposed in (6.52) are more plausible than such word-final syllabic glides. And if resonants in final position can be regarded as syllabic, there is no reason not to regard the resonants seen word internally in the same way:
(6.53) [(hŽlmn)(Joid)] [(k‡vl)(kd)] [(‡rïn)(tön)] [(h—tn)tot]

[(h‡kn)sak] [(p‡lmp)(sest)] [(‡lïr)(non)]

Almost all of these words have no internal morphological structure, and the internal "vowel" proposed for the penult in the traditional analysis of these forms has no distinctive vowel quality, but closely resembles the syllabic resonant in most cases. The reasons for proposing a syllable in this position in previous analyses were evidently due to orthography and the assumption that clusters of the form -CRC- could not appear word internally, although -CR# could be proposed word-finally. The postulation of such lexical forms would help explain the contrast seen between words like h—ttentot or c‡valcade and rec—ndite or appŽndix, which show the expected stress patterns for words with closed penultimate syllables. The latter two would thus have full vowels in their penults (e.g., /rekondöt/, /apendiks/), while the former are really bisyllabics (e.g., /kavlkd/, /hotntot/) with internal -CRC- clusters which surface as word-internal syllabic nasals.¹¹¹ There is no a priori reason to restrict lexical entries containing such internal consonant sequences, and this proposal provides an explanation of how such lexical entries could surface while at the same time accounting for the phenomenon of "sonorant retraction". The idea that lexical entries must consist of nicely alternating and easily syllabifiable CV sequences, which mirror their surface forms, is just another assumption of derivational theory.¹¹²

Kager’s examples of so-called "long retraction" included words of the following type, cited above in (3.9):

(6.54) pŽregrinˆte c‡tamarˆn man’pulat˜ry

amŽliorˆte hœllaball˜o hallœcinat˜ry

detŽriorˆte r’gamar˜le art’culat˜ry
The types in / ate/ and / atory/ have been discussed above in (6.7) and (6.21). The / ate/ forms are multiply suffixed, and the stem-subcategorizing suffixes (here / or/ and / ine/¹¹³) are unable to take the main stress due to their position at the end of the morphological word. In such forms, the stress remains on the stem foot, which is thus one syllable further to the left than had there been no internal suffixation. The morphological word boundary is inside the prosodic word, and so the sequence of stressless syllables does not incur any fatal violations. The situation is similar in the / atory/ words.

The remaining few words, which are monomoraic, are best explained through the presence of underlying /æ/ within the stem, i.e., /katæmæran/, /hulæbælu/, /rigæmær¯l/ (¤ 5.4.3). These vowels surface with no quality and are plausibly not otherwise interpretable to learners. The unstressability of /æ/ will force the attested foot structures. The limited distribution of schwas in lexical entries means that there are not so many of these marginal forms. However, in OT all forms, including these, need to be accounted for, and the proposal of /æ/ by the grammar is the most effective way of dealing with these forms, as there is then no need to rework the entire complex grammar of foot-formation on the basis of a few infrequent and lexically unanalyzable forms.

6.8 The non-Latinate grammar

The primary focus of this work concerns the Latinate vocabulary of English, a set of words borrowed into English via French and Latin, and possessing many of the morphological and grammatical features of Romance languages. In many ways, the Latinate grammar can be regarded as a foreign sub-grammar grafted onto English. In this section, the "native" English grammar will be briefly discussed, as its words also need to be accounted for via the same constraint hierarchy. However, this is not the main focus of this study, and a detailed investigation into the native lexicon will be left for future investigation.

"Native" English words are quite different from Latinate ones; there is far less morphology, and most words are only one or two syllables long. There are certainly some vowel alternations to be seen, for example w´ld Ü w’lderness, or ke³ep Ü kŽpt, w´de Ü w’dth. However, most such words or affixes show morphological idiosyncrasies, and do not display regular patterns. For example, the suffix / th/ is also seen with the irregular alternation l—ng Ü lŽngth, with stem allomorphy, while e³ight Ü e³ighth fails to present the shortening seen in f´ve Ü f’fth. The extension to w´ld seen in w’lderness is likewise unique to that word. On the other hand, the alternation seen in ke³ep Ü kŽpt appears regularly in a limited set of verbs which form their past tense in / t/. Following the principles outlined above, most of these cases can be explained in the same way as the "shortening" cases from the Latinate vocabulary (¤ 4.3.1). If the suffix / t/ is understood as subcategorizing within the stem, then morphologically conditioned lengthening will not take place, since the addition of the suffix into the stem makes it bimoraic, rendering lengthening unnecessary, i.e., [{kep}æ] Ü [{kept}æ], [{wid}æ] Ü [{wid-J}æ].

However, this does not explain the mass of monosyllabic words of the form h’t, which show what in the Latinate grammar would be considered a single light syllable. According to the current constraint hierarchy, we would expect such a syllable to be lengthened, but in reality about 28% of monosyllabic words show this structure. One solution to this problem would be to label all these stems as geminate-final (that is, they have lexical moras linked to the stem-final consonant). This might appear to be problematic, since geminates appear only in about 13% of the Latinate words, not 28%. Such a solution also appears to introduce complexity into the lexical entries involved. However, there are a number of arguments in favor of the geminate analysis. One is that many of these forms are historically geminate, such as h’t (Old English hittan), l’ck (Old English liccian), and the underlying geminate structure (i.e., a lexical consonantal mora) can be understood as surviving into the synchronic grammar.

Another point is that arguments on the basis of lexical complexity fall down when regarding monosyllables with surface long vowels, such as re³ad, e³at, co³al. These account for about 29% of monosyllabic roots, and if regarded as underlyingly long, would themselves need to be marked with a mora in the lexicon. Marking the geminates rather than the surface long vowel cases (which would then be regarded as /red/, /et/, /col/) with moras involves almost exactly the same number of stems, a bit fewer in fact. This latter interpretation also accounts for alternations like ke³ep Ü kŽpt, sle³ep Ü slŽpt, without making forms like e³ighth exceptions. Such words would be regarded rather as underlyingly long (i.e., {«t}Ü{«t-J} vs. {fiv}Ü{fiv-J}), as in the Latinate examples given above.

A second explanation, which seems less promising, would be to regard non-Latinate stems as subcategorizing directly for the prosodic word, avoiding the constraints which enforce final consonant extrametricality on the morphological word. In such a situation, final consonants would be syllable-closing and moraic, hence words like h’t would be bimoraic, and lengthening would only be seen in open syllables, e.g., be³, go³.¹¹⁴ It is likely that this second explanation poses a number of further difficulties, which will not be further explored here. The issue of how the non-Latinate vocabulary of English further behaves will have to be left to future research.

6.9 Conclusions about English stress patterns

In chapters 3-6, structures have been proposed to account for the various stress patterns seen in both complex and simplex words in English. The various alternations seen between related words, and the differing stress patterns seen across the vocabulary, have largely been accounted for by proposing various morphological structures for different word types, enforcing this primarily through affix subcategorization. Taking the "Latin" stress pattern showing final syllable extrametricality (Kager’s group I) as the basic stress pattern (¤ 3.3), the general verbal pattern (group II) can be accounted for by the presence of a suffix / æ/ (¤ 3.3.4), while idiosyncratic examples of apparently light monosyllabic feet (group III) can be understood through the proposal of lexical moras linked to consonants, i.e., geminates (¤ 3.3.2). Differing patterns seen among the various affixed forms have been connected to the different morphological subcategorizations demanded by these affixes (¤ 4.2), and the resultant edge effects seen in these constituents (as seen in chapters 4-6).

Vowel alternation has been accounted for by reassessing the status of long vowels in morphologically predictable positions (¤ 4.3). By analyzing the relative frequencies of syllable types throughout the lexicon, rather than concentrating upon a few unusual forms, the unexpectedly infrequent distribution of light syllables in monosyllabic roots as compared to multisyllabic roots could be identified and accounted for through the proposal of morphologically conditioned lengthening (¤ 4.3.1). Vowels which failed to undergo "shortening" can now be understood as underlyingly long, while short vowel quantity can be seen as the unmarked case for most vowels in the lexicon (¤ 4.3.3). This reassessment of the underlying forms for Latinate roots led to further clarifications, such as the otherwise inexplicable "retraction" and "shortening" seen in forms like rŽsident (¤ 5.1).

The preceding account of the general stress patterns of English has the advantage of being expressible via a single constraint hierarchy, which is necessary for a constrained Optimality Theory account. No previous treatment of English stress has been able to account for the variety of stress patterns in a uniform manner, or for contrasts in the treatment of segmentally identical forms without resorting to unstructured exception marking and extra-theoretical provisos, excluded from this approach. Here, idiosyncratic patterns have been accounted for by linking prosodic and morphological constituents, necessary for the overall analysis, to lexical items themselves. These tactics require no further complexity than traditional lexical entries and affixation rules, but allow for a coherent OT account of English stress and vowel alternation.

The above account relies on the lexicon as the storehouse of idiosyncratic information, both segmental and constituent-based. In other words, it is "underlying" morphological and prosodic structures which are responsible for the varying surface realizations afforded to segmentally similar candidates. These different arrangements of lexically specified constituents are necessary for the grammar to account for the many different stress patterns that the language learner is confronted with when acquiring the grammar of English. Each and every word must be an acceptable output, or put another way, a lexical form which can fit into the existing context of the grammar must be found for every word.

For most words, the correspondences between the structures required by the grammar and the logical structure of the word are consistent, and it is this regularity which allows for the correct grammar to be generated as it stands. However, not all correspondences between the prosodic structure and logical morphological structure of words will be regular. In some cases, an unusual word demands a structure which puts it at odds with other parts of the grammar, perhaps with related forms in the "paradigm", or with general assumptions about word structure. Such words are commonly regarded as "exceptions", and must be accounted for as well in the grammar. The next chapter treats the concept of exceptionality, and how it may be dealt with in Optimality Theory. In doing so, it also reassesses the role of the lexicon in assigning morphological structures, and the relation of the lexicon to the rest of the grammar.