=Paper=
{{Paper
|id=Vol-1607/spizzirro
|storemode=property
|title=Expanding a Theoretical Framework for English Adjective Order
|pdfUrl=https://ceur-ws.org/Vol-1607/spizzirro.pdf
|volume=Vol-1607
|authors=James Spizzirro
|dblpUrl=https://dblp.org/rec/conf/clif/Spizzirro16
}}
==Expanding a Theoretical Framework for English Adjective Order==
https://ceur-ws.org/Vol-1607/spizzirro.pdf
Expanding a theoretical framework for English adjective order
James Spizzirro
Boston College
Department of Slavic and Eastern Languages and Literatures
Lyons Hall 210
Boston College
140 Commonwealth Ave
Chestnut Hill, MA 02467
spizzirr@bc.edu
Abstract speakers, however, would claim that, according to
some intuition, the utterances in (1) and (3) are
In her 2003 study “A multifactorial corpus more natural than the utterances in (2) and (4).
analysis of adjective order in English”, Since this order cannot be explained by any known
Stefanie Wulff surveys a number of factors hierarchical relationship, we must look elsewhere
that help explain preferred adjective ordering for an explanation.
in adjective-adjective-noun constructions, A number of linguists have proposed rules that
drawing from previous work in fields ranging help predict adjective order. The vast majority of
from phonology to pragmatics. In the present
work in this field has focused on phonology (cf.
work, I pose an expansion in the criteria for
one of these factors, which should yield a Behaghel 1930, Goyvaerts 1968), syntax (cf. Biber
more effective utilization in determining 1999, Posner 1986), semantics (cf. Whorf 1945,
adjective order. In addition, I propose that one Kilgarriff 1997, Stubbs 2001, Richards 1975,
factor in Wulff 2003 provides support for the Martin 1969, Ertel 1971, Dixon 1977, Quirk et al.
expansion proposed here. Following this, I will 1985, Hetzron 1978, Richards 1977, Deese 1964)
explain areas for further research which have and pragmatics (cf. Lockhart and Martin 1969,
come to light over the course of this study, but Posner 1986, Bock 1982, Ney 1983, Lapata et al.
which have not been treated in this paper. 1999). It is worth noting that there has been little
substantial work regarding the influence of
1 Introduction morphology on adjective order, and whatever work
has been done has been largely restricted to
An interesting and largely unexplained superlative and comparative structure (cf.
phenomenon in language is the preference of a Teodorescu 2006). Stefanie Wulff presents a
particular order of adjectives over another. For comprehensive study on this topic in her 2003 work
instance, compare the grammaticality of the “A multifactorial corpus analysis of adjective order
utterance in (1) with that of the utterance in (2): in English”. In this study, Wulff surveys a number
of previously proposed factors, drawing primarily
(1) small green car from work in phonology, syntax, semantics, and
(2) *green small car pragmatics (with no section devoted to the
treatment of morphological factors). By analyzing
Or, as a more relevant example, compare the these factors at first individually, and then together
utterance in (3) with the one in (4): in a multifactorial analysis, Wulff was able to
predict adjective order with 73.5% accuracy.
(3) beautiful colourful jewels One factor presented in Wulff’s study, referred
(4) ?colourful beautiful jewels to as Nominal Character (NOMCHAR), is described
as problematic, and loses much of its predictive
Syntactically, these are all equally well-formed capability when used in a multifactorial analysis.
utterances. The vast majority of native English Wulff addresses this by saying that “the
58
operationalization of NOMCHAR applied [in her least three of the criteria in general (cf. Quirk et al.
analysis] is not inadequate, but probably 1985: 402–404).
incomplete in the sense that NOMCHAR should be Triples: This survey has considered “triples”,
more adequately conceived of as a multifactorial groups containing two prenominal adjectives
construct of which the tendency towards immediately followed by a noun. Though there are
nominalization is just one aspect” (Wulff 2003). In instances in the BNC of more than two adjectives
the present work, I present one such aspect of preceding a single noun, these are rare: In the entire
Nominal Character which could aid in a more 10 million-words spoken portion of the BNC, there
effective operationalization of this factor, and are 9,647 adjective pairs. Only 426 of these pairs
which takes morphology into account, which (as are immediately followed by another adjectives, or
noted above) has largely been ignored in the 4.41%. Conversely, the vast majority of these
literature. I propose that by expanding Nominal adjective pairs (6,560, or 68%) are followed by a
Character to include adjectives which are noun (cf. Wulff 2003). Therefore this study will be
nominalized by means of removing or adding an primarily concerned with the behavior of triples
affix, rather than restricting the criteria to zero- (Adjective-Adjective-Noun constructions).
derived nouns and adjectives, Nominal Character Unbroken adjective pairs: This study treats
can be utilized more fully. Furthermore, I propose “unbroken” adjective pairs; that is, those that are
that one factor in particular presented in Wulff’s not joined by a conjunction. The behavior of
2003 study lends validity to expanding Nominal “broken” adjective strings is not well understood at
Character. the present moment. It seems that broken strings
are, at least at first glance, less sensitive to
2 Method adjective ordering restrictions. For example,
compare the grammaticality of the utterance in (5)
This study will treat adjectives and adjective with that of the utterance in (6):
strings that were dealt with Wulff’s study. For a
full explanation of the constraints on this analysis, (5) *green small car
see Wulff 2003, §2 “Scope of the investigation.” (6) ?green and small car
For purposes of clarity, I will here delineate a
number of parameters within which I have This is doubtless an area for further research,
conducted this study. but one that will not be discussed further in the
Descriptive Adjectives: There is a distinction present work.
between adjectives such as “many” and “fourth”
and adjectives such as “red,” “beautiful,” and 3 Expanding Nominal Character
“big.” This first group (described by Wulff as
“limiting adjectives”) specifies and constrains the When analyzing Nominal Character, Wulff draws
head, whereas this second group (described by primarily from Posner’s (1986) so called
Wulff as “descriptive adjectives”) serves primarily “nouniness principle” – less “noun-like” adjectives
to describe, rather than to specify, the head. It is tend to precede more “noun-like” ones. There are
this second group of adjectives that this study is two methods of analyzing the Nominal Character of
concerned with. Additionally, all adjectives used an adjective: Posner’s analysis, which has some
here must fulfill the description of either a central noted problems and cannot be effectively utilized in
or peripheral adjective, according to the four a corpus analysis1; and Wulff’s analysis, which was
criteria prescribed by Quirk et al., namely (i) created specifically for operationalization in a
attributive function; (ii) predicative function after corpus analysis. For purposes of ease and clarity,
the copula seem; (iii) ability to be modified by this study is concerned with Wulff’s (slightly
very; and (iv) gradability by the use of
morphology (-er, -est) or periphrastic comparison
(more, most). Central adjectives must fulfill at
1
For a full discussion of the problems of Posner’s analysis, see
least (i) and (ii), and peripheral adjectives must
Wulff 2003, §4.2.1. Suffice it to say that there are problems in
fulfill at least (i), otherwise they must fulfill at interpreting Posner, and his analysis relies upon intuitions that
cannot be utilized on a large enough scale in a corpus analysis.
59
altered) presentation of Nominal Character, which Here, frequencyadj is the frequency of the word
in turn draws a number of key insights from as an adjective and frequencyn is the frequency of
Posner. the word as a zero-derived noun.
Wulff’s use of Nominal Character: Wulff frequencyadj
(8) NOMCHARD = 1 –
maintains a number of the parameters of Posner’s frequencyadj + frequencydn
original study. The most important of these for the
purposes of this study is that Wulff surveys Here, NOMCHARD is “Nominal Character
adjectives and zero-derived nouns (e.g., “green” as (Derived)”, frequencyadj is the frequency of an
an adjective ‘green car’ and “green” as a noun ‘I adjective, and frequencydn is the frequency of a
love this green’) to be consistent with Posner’s noun derived from this adjective (“derived noun”).
apparent intentions. For use in her analysis, Wulff Alternatively, NOMCHARD can analyze words
determined the number of times each word was with an opposite derivation pattern, i.e. an
tagged as an adjective or as a noun in the BNC, adjective that is derived (morphologically) from a
and used these numbers to calculate a word’s noun and the noun from which it was derived. In
Nominal Character value. According to Wulff’s this case, the formula will take the form in (9):
formula for Nominal Character, the higher a
word’s Nominal Character value, the more likely frequencyda
it would be to appear as the second adjective in a (9) NOMCHARD = 1 –
frequencyda + frequencyn
string (and conversely, the lower its value, the
more likely it would be to appear as the first The only difference is that here, frequencyn is
adjective in a string). See §3.1 below for a fuller the frequency of a noun, and frequencyda is the
discussion of Wulff’s formula. frequency of an adjective derived from this noun
(“derived adjective”).
3.1 Beyond zero-derivation The output of this formula provides the same
predictions as Wulff’s formula: The higher a
In order to approach a fuller operationalization of word’s Nominal Character (Derived) value, the
Nominal Character, we must stray a bit from more likely it is to be the second adjective in a
Posner’s original criteria: Zero-derivation should string, and vice-versa.
not be a strict criterion in determining Nominal Due to a lack of resources and technical
Character. If we expand our analysis to include expertise, this analysis was not performed
overtly derived adjectives and nouns, we can automatically, and as a result of this, the test pool is
highlight relationships currently considered necessarily smaller than the one presented in Wulff
outside of the scope of Nominal Character. There – 528,714 words. On the one hand, this manual
is a significant problem with expanding this analysis corrects any instances of incorrect tagging
analysis past zero-derivation, however: In Wulff’s in the BNC. On the other hand, however, this
operationalization of nominal character, the manual analysis increases the possibility for human
BNC’s tags were sufficient to provide the error – I have checked and double-checked all
variables needed to solve the equation for Nominal results, but any errors are my own.
Character, whereas a simple tag search for a single
word will not provide us with the information 3.2 Interpreting the Results
necessary for morphologically distinct, rather than
zero-derived, forms. In order to analyze Since this analysis is manual, interpretation of the
morphologically distinct forms, we must slightly output is manual as well – by utilizing native
alter Wulff’s formula for Nominal Character. The speaker intuitions, and drawing generalizations and
original formula is represented by (7), and the conclusions when comparing them with a word’s
revised formula is presented in (8): NOMCHARD value. A noted difference between
Nominal Character and our expanded Nominal
frequencyadj
(7) NOMCHAR = 1 – Character (Derived) is that, whereas the vast
frequencyadj + frequencyn majority of adjectives (89.1% in Wulff’s analysis)
have a Nominal Character value between 0 and 0.1,
60
Nominal Character (Derived) values seem to vary Outputs between .5 and 1: When the input of
more widely. As a result of this, conclusions frequencya (or frequencyda) is less than the input of
regarding the influence of Nominal Character frequencydn (or frequencyn), the output of Nominal
(Derived) on adjective order can be made utilizing Character (Derived) will fall somewhere between .5
a larger portion of the Nominal Character and 1, non-inclusive ({.5 freqdn/
7). When we pair these to modify some noun, say
1 134 90 “traits,” we find that the ordering is as their values
2 17 354 predict. Compare the grammaticality of the utterance
3 77 48 in (10) with that of the utterance in (11):
4 115 83
5 157 0 (10) certain (.040) wonderful (.384) traits
6 80 3 (11) ?wonderful (.384) certain (.040) traits
7 359 15
8 94 0 Native speakers generally accept the
grammaticality of the utterance in (10). On the other
Figure 4.2: Adjective-noun frequencies. hand, the grammaticality of the utterance in (11) is
context-dependent at best: In the absence of any
Pair # NOMCHARD special emphasis or changes in prosody, this seems a
1 .402 less-grammatical utterance.
2 .954 To take another example of the predictive
3 .384 capability of Nominal Character (Derived), another
4 .410 random pairing: Take smooth (from pair 6) and
5 * colourful (from pair 2), along with a noun, say,
6 .036 “dress.” Compare the grammaticality of the
7 .040 utterance in (12) with that of the utterance in (13):
8 *
(12) smooth (.036) colourful (.954) dress
Figure 4.3: Nominal Character (Derived) Values (13) ?colourful (.954) smooth (.036) dress
The distinction here is, admittedly, less clear
than the distinction between (10) and (11). The
difference in two adjectives’ Nominal Character
(Derived) values should not be taken as a measure
2
It has been brought to my attention that I overlooked this of the rigidity of the ordering preference, but rather
word’s function as a determiner, and due to time constraints I
have not been able to rectify this in the present study. See §6
as a general guideline for ordering preference.
for further discussion.
62
It isworth noting that there are pairings that (Derived) under the umbrella of Nominal
will not fit this general guideline. Take, for Character, we must make adjustments that take into
example, “beautiful” (from pair 1) and “smooth” consideration the fact that, while Nominal
(from pair 6). The predicted ordering is Character (Derived) values cover a very large
represented in (14), while the preferred ordering is range, the vast majority of Nominal Character
shown in (15): values fall between 0 and .1 (Wulff 2003). Doing so
simply involves multiplying the Nominal Character
(14) ?smooth (.036) beautiful (.954) door (Derived) value by .1 (in general, however, in order
(15) beautiful (.954) smooth (.036) door to account for the minority of adjectives with
values between .1 and 1, some adjustments should
This example violates the principle of be made which, at present, have not been included).
Nominal Character (Derived). For a full discussion External support for NOMCHARD: By drawing
of exceptions to Nominal Character (Derived), see generalizations from a pragmatic factor in adjective
§§6-7 below. Given the apparent multifactorial order, we find that Nominal Character (Derived)
nature of adjective order, it is not surprising that fits into Wulff’s framework. Wulff presents a
there are some exceptions to Nominal Character factor, drawing on the work of Bock (1983) and
(Derived). Ney (1982), called General Frequency, that
Now, returning to pairs (5) and (8), which presents a correlation between the number of times
could not be properly analyzed in the corpus an adjective occurs in a corpus (its general
analysis: We can use a reverse analysis to attempt frequency) and its proximity to the noun: The more
to find the Nominal Character (Derived) value of frequently an adjective occurs, the more likely it is
these two adjectives. Consider the preferred to appear as adjective1, further from the noun. By
ordering of “heavy” and “smooth”, shown in (16), generalizing this factor to take into account an
and the non-preferred ordering, shown in (17): adjective’s relative frequency (i.e., the number of
times it occurs as an adjective rather than a noun),
(16) smooth (.036) heavy (*) door we find another way of interpreting Nominal
(17) ?heavy (*) smooth (.036) door Character (Derived). In this view, the greater the
frequency of an adjective, and the lower the
For the purposes of this study, this frequency of its corresponding noun, the more
generalization will suffice: By comparing an likely it is to appear as adjective1 in a string (and
adjective (which is either derived from a noun, or vice-versa).
from which a noun is derived) with an unknown
Nominal Character (Derived) value with an 6 Discussion
adjective with a known Nominal Character
(Derived) value, we can approximate a range of In general, Nominal Character (Derived) can be
values. For instance, in (16) we see that the used as a secondary aspect of Nominal Character in
Nominal Character (Derived) value for “heavy” cases where Nominal Character would not
likely falls somewhere above .036. With enough accurately analyze an adjectives trend toward
adjective pairs, the range can be narrowed, and we nominalization. Due to the apparent multifactorial
can estimate a more accurate value. nature of adjective ordering restrictions, however,
there are times when it seems nominal Character
5 Situating NOMCHARD (Derived) cannot properly predict adjective order.
When there is significant influence from other
Nominal Character (Derived) is, in essence, a factors that have not been considered in the present
single aspect of Nominal Character. Use of analysis, there may be a discrepancy in predicting
Nominal Character (Derived) should generally be adjective order with Nominal Character (Derived).
restricted to those adjectives that would not be In these cases, it seems that this factor’s efficacy
properly analyzed through Nominal Character (for may be dwarfed by other factors. Nonetheless, by
instance, “colourful” is considerably less likely to expanding the criteria of Nominal Character we are
appear as a zero-derived noun than it is as able to more accurately represent a phenomenon
“colour”). In order to use Nominal Character which likely accounts, in part, for adjective order.
63
Returning to the issue presented in note 2 in Nominal Character (Derived) values for each
§4: It has been pointed out that the usage of adjective. For instance, if we take the adjectives
‘certain’ and ‘certainty’ here may fall outside the green and greenish, and find that they have greatly
scope of this study: Its use in (10) and (11) is more differing Nominal Character (Derived) values, it
akin to a determiner than an adjective. Therefore, may be the case that by comparing the average of
as noted by one reviewer, it is possible that rather these two values with the other adjective in the
than preferring adjective1 position, it occupies the string, we may be able to correct our prediction.
determiner position. Unfortunately, due to time The formula for this is shown in (19):
constraints, this cannot be rectified in the current
analysis. This will, hopefully, be addressed when !"#! !⋯ !"#!
(19) AVGNCD = 1 –
this analysis is more complete. n
Regarding the future of this study, it is this
author’s hope that an automated analysis can be Where NCD is Nominal Character (Derived)
performed, in order to provide a more and n is the number of Nominal Character
representative sample of adjectives in a larger test (Derived) values that are being compared.
pool. Additionally, see §7 for a discussion of Relative Morpheme Frequency: There may
further research in morphological factors in be a relationship between the derivational affix
adjective ordering. used to derive an adjective and its place in a two-
adjective string. Following the pattern of Nominal
Character (Derived) and our generalization of
7 For further research
General Frequency, it may be the case that the more
often a derivational affix occurs, the more likely it
Over the course of this study, a number of
is to force an adjective into adjective1 position.
conceivable relationships have come to light
There are two ways to assess an affix’s frequency,
specifically regarding the status of morphology in
represented by (20) and (21):
determining adjective order that have not yet been
considered in a corpus analysis. I will here frequencyM1
describe these and give some basic considerations (20) RMF =
frequencyM1 + … frequencyMn
regarding them3.
Adverbial Character: Adverbial Character,
Where M is a morpheme, M1 is the morpheme
or ADVCHAR, analyzes the relative frequency of an
in question, and n is the total number of all
adjective and its corresponding adverb (for
derivation affixes. Alternatively:
instance, wonderful and wonderfully). This factor
predicts that the higher an adjective’s Adverbial frequencyM1
Character value, the more likely it is to occur as (21) RMF =
!
adjective1 (in contrast to Nominal Character). The
formula is shown in (18): Where S is the size of the test pool.
frequencyadj
(18) ADVCHAR = 1 – 8 Conclusion
frequencyadj + frequencyadv
In order to more effectively utilize Nominal
Average Nominal Character (Derived): For Character in predicting adjective order, it may prove
either (i) an adjective from which a number, n, of helpful to expand the analysis past zero-derivation.
nouns can be derived using distinct derivational By including derived forms, certain adjectives may
affixes, or (ii) a noun from which a number, n, of be analyzed more accurately in those situations
adjectives can be derived using distinct where Nominal Character may not correctly predict
derivational affixes, it may be possible to predict an adjective’s position in a string. More generally,
adjective order by calculating the average of the morphological considerations for adjective order
may help further our understanding of the
phenomenon as a whole.
3
What follows in this section is largely speculation. No
corpus analysis has been performed to test these hyptotheses.
64
Acknowledgements Deese, J. 1964. The Associative Structure of Some
Common English Adjectives. Journal of Verbal
I am indebted to a number of people for their Learning and Verbal Behavior, 3, 347-357.
contributions over the course of this study, most Dixon, R. M. W. 1977. Where Have all the Adjectives
notably Professor Claire Foley, for her comments Gone? Studies in Language, 1, 19-80.
and guidance, the faculty of the Boston College Ertel, S. 1971. Pränominale Adjektivfolgen und
Slavic & Eastern Languages and Literatures semantische Tiefenstruktur. Studia Psychologica, 13,
department, my peers Joseph Maimone, Harry Hoy, 127-137.
and Eddie Hasell, and my parents.
Goyvaerts, D. L. 1968. An Introductory Study on the
Ordering of a String of Adjectives in Present-day
Appendix (Abbreviations) English. Philologica Pragensia, 11, 12-28.
adj: adjective Hetzron, R. 1978. On the Relative Order of Adjectives.
In H. Seiler (Ed.), Language Universals (pp. 165-
AO: adjective order (or adjective ordering
184). Tübingen: Narr.
restrictions)
adv: adverb Kilgarriff, A. 1997. I don’t believe in word senses.
ADVCHAR: Adverbial Character Computers and the Humanities, 31, 91-113.
AVGNCD: Average Nominal Character (Derived) Lapata, M., S. McDonald, & F. Keller 1999.
BNC: (Second) British National Corpus Determinants of Adjective-Noun Plausibility.
da: derived adjective Proceedings of the 9th Conference of the European
dn: derived noun Chapter of the Association for Computational
M: morpheme Linguistics, 30-36.
NOMCHAR: Nominal Character Lockhart, R. S., & J. E. Martin 1969. Adjective Order
NOMCHARD: Nominal Character (Derived) and the Recall of Adjective-Noun Triples. Journal of
n: noun (though it may be a variable in certain Verbal Learning and Verbal Behavior, 8, 272-275.
equations) Martin, J. E. 1969. Semantic Determinants of Preferred
RMF: Relative Morpheme Frequency Adjective Order. Journal of Verbal Learning and
Verbal Behavior, 8, 697-704.
References
Ney, J. W. 1983. Optionality and Choice in the Selection
Andrews, Avery. 1983. A note on the constituent of Order of Adjectives in English. General
structure of modifiers. Linguistic Inquiry, 14, 695- Linguistics, 23, 94-128.
697.
Panayidou, Fryni. 2013. (In)flexibility in Adjective
Behaghel, O. 1930. Von deutscher Wortstellung. Ordering. Doctoral Dissertation, Queen Mary,
Zeitschrift für Deutschkunde, 44, 81-89. University of London.
Bloomfield, L. 1933. Language. Chicago, London: Posner, R. 1986. Iconicity in Syntax. The Natural Order
Chicago University Press. of Attributes. In P. Bouissac, M. Herzfeld, & R.
Posner (Eds.), Iconicity. Essays on the Nature of
Biber, D., S. Johansson, G. Leech, S. Conrad, & E.
Culture (pp. 305-337). Tübingen: Stauffenburg
Finegan (Eds.) 1999. Longman Grammar of Spoken
Verlag.
and Written English. London: Longman.
Quirk, R., S. Greenbaum, G. Leech, & J. Svartvik (Eds.)
Bock, J. K. 1982. Toward a Cognitive Psychology of
1985. A Comprehensive Grammar of the English
Syntax: Information Processing Contributions to
Language. London: Longmans.
Sentence Formation. Psychological Review, 89, 1-47.
Richards, M. M. 1975. The Pragmatic Communication
Bolinger, Dwight. 1967. Adjectives in English:
Rule of Adjective Ordering: A Critique. American
attribution and predication. Lingua, 18, 1-34.
Journal of Psychology, 88, 201-215.
Cinque, Guglielmo. 2010. The syntax of adjectives: A
Richards, M. M. 1977. Ordering Preferences for
comparative study. Linguistic Inquiry Monographs
Congruent and Incongruent English Adjectives in
57. Cambridge, MA: MIT Press.
Attributive and Predicative Contexts. Journal of
Verbal Learning and Verbal Behavior, 16, 489-503.
65
Stubbs, M. 2001. Words and Phrases: Corpus Studies of
Lexical Semantics. Cornwall: Blackwell.
Teodorescu, A. 2006. Adjective Ordering Restrictions
Revisited. Proceedings of the 25th West Coast
Conference on Formal Linguistics, 399-407.
Somerville, MA: Cascadilla Proceedings Project.
Truswell, R. 2009. Attributive adjectives and nominal
templates. Linguistic Inquiry, 40, 525-533.
Whorf, B. L. 1945. Grammatical Categories. Language,
21, 1-11.
Wulff, S. 2003. A multifactorial corpus analysis of
adjective order in English. International Journal of
Corpus Linguistics, 8:2, 245-282.
66