Analogy in Suffix Rivalry: The Case of –ity and –ness in English

35. Jahrestagung der DGfS
Potsdam, March 12‐15, 2013
AG 7: Usage‐Based Approaches to Morphology
Analogy in Suffix Rivalry: The Case of –ity and –ness in English
Sabine Arndt‐Lappe
Heinrich‐Heine‐Universität Düsseldorf
sabine.arndt‐lappe@uni‐duesseldorf.de
Structure
• –ity / ‐ness rivalry: the phenomenon & some wellknown problems • An analogical model of suffix rivalry
• Modelling 20th century neologisms in AM
– data
– overall performance
• Analysis: Virtues and flaws in an analogical model
• Summary & conclusion
2
‐ity vs. –ness
• two main possibilities to nominalise (predominantly) adjectival bases: ‐ity and –ness (other possibilities: ‐th, ‐cy, ?‐ism)
• both possibilities are productive synchronically
examples:
20th century neologisms as (from the Oxford English Dictionary, OED)
connectibility
(< connectable)
excitingness
(< exciting)
nordicity
(< nordic)
genericness
(< generic)
metaphoricity
(< metaphoric)
blokishness
(< blokish)
prescriptivity
(< prescriptive)
commutativeness
(< commutative)
picayunity
(< picayune)
norseness
(< norse)
3
Factors affecting the variation
• form
‐ity causes stress shift and stem alternation, ‐ness does not
•
etymology: Latinate or non‐Latinate
‐ity is Latinate, ‐ness is Germanic
‐ity is attached predominantly to Latinate bases
•
selectional restrictions
adjectival suffix of the base determines choice of –ity or ‐ness
•
productivity / restrictiveness
‐ness is less restrictive than ‐ity
•
semantics
hypotheses about subtle differences in terms of • referential scope and
• transparency
(cf. esp. Marchand 1969, Aronoff 1976, Anshen & Aronoff 1981, Riddle 1985, Romaine 1983,
Giegerich 1999, Plag 2003, Baeskow 2011, Bauer et al. 2013)
4
Selectional restrictions: semi‐systematicity
• categorical vs. variable
some bases categorically select –ness, for others there is variation
e.g. Germanic –ed (informed‐ness, *informedity) vs. Latinate –ive: prescriptivity vs. prescriptiveness (OED, 20th cent.)
• degrees of preferences
e.g. –able bases (achievability, repairableness) (OED, 20th cent.)
stronger vs. weaker preferences
often: preferences debated
• phonology or morphology?
The morphological status of the generalisation is often not straightforward
e.g. massiv(ity), optimal(ity), randomic(ity) (OED, 20th cent.)
• development
Selectional restrictions are static – can they change?
5
Semi‐productivity
(Romaine 1983)
• ‐ness can, in principle, be attached to (almost?) any base
‐ness seems in effect to serve as a sort of default way of forming abstract nouns from non‐verbal categories in contemporary English. (Bauer et al. 2013: chpt. 12.2.1)
•
BUT: For some bases, ‐ity is strongly preferred over –ness
•
e.g. bases in –able/‐ible (cf. e.g. Anshen & Aronoff 1981)
6
Proposal
• Show that –ity / ‐ness competition can be accounted for in an analogical model of suffix rivalry (computational implementation: AM, Skousen & Stanford 2007)
• Provide empirical support for semi‐systematicity of selectional restrictions and semi‐productivity of ‐ity
• Data: 20th century neologisms from the Oxford English Dictionary (OED)
• Analysis: How does the analogical approach differ from rule‐
based approaches?
– selectional restrictions are emergent and form‐based: no exception marking necessary
– degrees of restrictiveness are emergent: no lexical strata necessary
7
the hypothesis: ‐ity and –ness are attached by analogy
lexemes in the Mental Lexicon
base-analogue pairs
normal
formal
local
normál-ity formál-ity locál-ity
base of new word
=
focal
?focálity
new word
a new word is formed on the basis of existing base‐derivative pairs
(cf., e.g., Becker 1990 et seq, Skousen 1989, 1992, Daelemans et al. 2005; cf. Bauer 2001, Arndt‐Lappe to appear for discussion
8
computational analogical models
• can operationalise the notions of 'similarity' and 'sets of analogues' (alternative term: 'analogical sets')
• can model variation, categorial behavior and leakage
• analogy becomes predictive
• the theory becomes testable
• algorithm:
– AM(L) (Skousen & Stanford 2007, cf. Skousen 1989, 1992, Skousen et al. 2002 et seq.)
9
The basic architecture of an analogical model
item
exemplars in the lexicon
onset‐
pen
nuc‐
pen
coda‐
pen
onset‐
fin
nuc‐
fin
coda‐
fin
generativity
r
ə
=
t
ɪ
v
ity
norseness
=
=
=
n
ɔː
s
ness
donness
=
=
=
d
ʌ
n
ness
formality
f
ɔː
=
m
ə
l
ity
formalness
f
ɔː
=
m
ə
l
ness
local
l
əʊ
=
k
ə
l
ity
onset‐
pen
nuc‐
pen
coda‐
pen
onset‐
fin
nuc‐
fin
coda‐
fin
formality
f
ɔː
=
m
ə
l
ity
formalness
f
ɔː
=
m
ə
l
ness
local
l
əʊ
=
k
ə
l
ity
item
analogical set
new word: focal + ‐x? item
onset‐
pen
nuc‐
pen
coda‐
pen
onset‐
fin
nuc‐
fin
codafin
f
əʊ
=
k
ə
l
suffix
suffix
2x –
ity, 1x ‐ness
suffix: -ity
(majority10
choice)
Question: Constraints on similarity?
item
analogical set
new word: focal + ‐x? onset‐
pen
nuc‐
pen
coda‐
pen
onset‐
fin
nuc‐
fin
coda‐
fin
formality
f
ɔː
=
m
ə
l
ity
formalness
f
ɔː
=
m
ə
l
ness
localness
l
əʊ
=
k
ə
l
ity
onset‐
pen
nuc‐
pen
coda‐
pen
onset‐
fin
nuc‐
fin
codafin
f
əʊ
=
k
ə
l
item
suffix
4 features shared
5 features shared
AM has a principled, independent method at its disposal for deciding
• which, and
• how many shared features are necessary to qualify an exemplar to end up in the analogical set
basis: homogeneity of similar exemplars w.r.t. the choice of –ity or –ness (cf. esp. Skousen 2002 a, b, 2005, 2009)
11
The data
• all –ity and –ness derivatives that were first attested in the 20th century extracted from the OED
– N = 593
N‐ness = 228
– N‐ity = 365
Coded features
• the phonological makeup of the last two syllables of the base
• basis: machine‐readable OED transcription
• unit of analysis: syllabic constituents (onset, nucleus, coda) cf. e.g. Eddington 2002, 2010, Daelemans & Bosch 2005 for the methodology)
12
The experiment
• test set to be predicted: 20th century neologisms (N = 593)
• lexicon of exemplars: 20th century neologisms (N = 592)
• ‚leave‐one‐out‘ procedure
• parameter settings
– the absence of a feature is treated as a feature value (i.e., e.g., the absence of a coda counts as meaningful information)
– the target category is computed using no. of occurrences in the analogical set, not pointers
13
AM experiment: overall performance
F‐score, macro‐averaged:
% correct predictions (overall):
F‐score for –ity:
0.87
88%
0.90
% correct –ity:
F‐score for –ness:
% correct –ness:
93%
0.83
79%
=> very good overall predictive power
=> -ity is more predictable than -ness
Let's look at the details!
14
Semi‐systematicity and semi‐
productivity in the data
very strong tendency towards ‐ity
tendency towards ‐ity
variation
categorical –ness with Germanic suffixes
(N = 573, only those morph. classes with at least 10 attestations)
16
Semi‐systematicity
• ‐ity is never selected categorically
• degrees of –ity preference: ‐ar / ‐ic / ‐able / ‐al > ‐ive / ‐ous
Semi‐productivity
• no explanation for the very clear tendency of some adjectival bases to take –ity
• no explanation of the variable behaviour of simplex bases
17
Predicting semi‐systematicity
very strong tendency towards ‐ity
tendency towards ‐ity
variation
variation is overpredicted with –ness leaning bases
19
(N = 573, only those morph. classes with at least 10 attestations)
Advantages of an account in terms of analogy
• the observed tendencies are actually predicted
• variation is expected to occur
Problems for an account in terms of analogy
• accounting for categorical behaviour
• cf. predictions for –ness leaning suffixes
20
What is the basis of emergent generalisations?
or: How circular is this all?
Crucial insights
• Analogical sets are not identical to morphological classes
• Analogical sets are form‐based (of course!)
• look at 3 examples
• ‐able bases (tendency to take –ity)
• ‐y bases (categorically take –ness)
• ‐ous bases (variability attested and predicted)
22
Analogical sets are not morphological classes
‐able bases
‐ous bases
‐y bases
morphological class
N
111
26
59
analogical set min. size – max. size
2 – 128
3 ‐ 20
6 ‐ 31
analogical set median size
12
7
15
analogical set mean size
15.28
8.77
16.71
 They are smaller.
 Generalisations are much more local.
23
Analogical set composition
Test item
examples of analogues
equiprobable adsorbability, clubbableness
startable orientability, optimality
treey
geekiness, peakiness, trendyness
mopey
stripiness, mintiness
numinous
conterminousness, radioluminosity
monozygous
zygosity, bogusness, analyzability
ana. set size
2
28
6
29
3
15
 locality
 more weight is given to more similar exemplars
 no privileged features – analogues do not all share the same features with the target
24
 form‐based
Predicting degrees of restrictiveness
Why can –ness attach to more different bases than –ity?
Degrees of restrictiveness ~ degrees of locality
• ‚local analogies‘: The members of the analogical set are highly similar to the target word
example
test item: supraconductive (supraconductivity)
analogical set: superconductivity, destructivity
• ‚general analogies‘: The members of the analogical set are not highly similar to the target word
example
test item: withdrawn (withdrawnness)
analogical set: 18 exemplars none of which shares more than 2 features with the target
e.g. westernness, rewardingness, awakedness
26
Is there a difference between –ity
and –ness in terms of the degree of the locality of the prediction?
Yes, there is!
(W = 67488.5, p‐value < 2.2e‐16, Wilcoxon Test)
27
Restrictiveness and locality
• Analogical sets for –ness differ from those for –ity in terms of their similarity with the target
– ‐ity preferring sets • tend to be dominated by highly similar exemplars
• classification is relatively local
– ‐ness preferring sets • tend to be dominated by more sparse similarity relations
• classification is relatively less local
(cf. e.g. Derwing and Skousen 1994, Eddington 2000, Keuleers 2008 , Keuleers & 28
Sandra to appear for a similar argument for the English past tense)
Summary and conclusion: the facts
• semi‐systematicity
– OED neologisms confirm earlier claims about the semi‐
systematicity of –ity / ‐ness choices
– evidence for strong –ity preference for bases other than –
able (‐al, ‐ar, ‐ive, maybe –ous)
• semi‐productivity
– ‐ity / ‐ness competition cannot be explained on the basis of a global appeal to productivity
29
Summary and conclusion: the model
• –ity / ‐ness selection can be predicted in an analogical model
• differences to rule‐based accounts
– locality
– relevance of features
• implications for predictions
– semi‐systematicity
– differences in restrictiveness / productivity
• ‚niche‘ productivity, cf. Lindsay 2012 for further evidence
– diachronic change expected
30
some desiderata
• problem: categorical predictions of –ness preferring bases
• how is the account compatible with other factors influencing the variation
– ?semantics
– ?stylistic / register‐related factors
31
predicting diachronic change
How did the distribution of –ity and –ness come about, diachronically?
• data: OED neologisms from 3 different centuries
– 20th century (N = 593)
– 19th century (N = 348, sample)
– 18th century (N = 305, sample)
33
‚migration‘
34
Can AM model ‚migration‘?
• ‚Migration‘ is reflected in – an increase in the degree of certainty – a decrease of predicted variability
....of AM predictions
• test: 3 AM models
– test: 20th century, exemplars: 20th century
– test 20th century, exemplars: 19th century
– test: 19th century, exemplars: 18th century
•
gradient predictions:
random selection of exemplar => probability that –ity or –ness is selected
35
(N = 332)
=> contemporary data: great certainty of predictions for –able, ‐al, and –ic bases
36
predictions of diachronic experiments:
•
decreasing variability
•
increasing certainty
37
Summary and conclusion: diachronic change
• the facts
– not only is –ity productive; – its productivity for certain bases seems to increase (cf. ‚niche‘ behaviour, Lindsay 2012)
• the model
– diachronic change is expected and predicted
– key: changes in the lexical distribution
– locality of the prediction
38
Thank you very much for your attention!*
*... and thanks to
- Ingo Plag, Gero Kunter, and Melanie Bell for their support, assistance and critical input.
- Julia Homann, Alexander Kerber, Vineeta Michallow, and students of the HS Analogy (Universität
Siegen, 2011) for helping with the coding
39
References
Anshen, Frank & Mark Aronoff. 1981. Morphological productivity and phonological transparency. Canadian Journal of Linguistics 26. 63–72.
Arndt‐Lappe, Sabine. in press. 47. Word‐formation and analogy. In Peter O. Müller, Ingeborg Ohnheiser, Susan Olsen & Franz Rainer (eds.), Word‐Formation ‐ An International Handbook of the Languages of Europe. Berlin: de Gruyter Mouton.
Aronoff, Mark. 1976. Word Formation in Generative Grammar. Cambridge, MA: MIT Press.
Baeskow, Heike. 2012. ‐Ness and ‐ity: Phonological exponents of n or meaningful nominalizers of different adjectival domains? Journal of English Linguistics 40(1). 6–40.
Bauer, Laurie. 2001. Morphological Productivity. Cambridge: CUP.
Bauer, Laurie, Rochelle Lieber & Ingo Plag. 2013. English Morphology: A Reference Guide to Contemporary English Word‐Formation and Inflection. to appear.. Oxford: OUP.
Becker, Thomas. 1990. Analogie und morphologische Theorie (Studien zur theoretischen Linguistik). München: Fink.
Daelemans, Walter & Antal d. van Bosch. 2005. Memory‐Based Language Processing. Cambridge: CUP.
Derwing, Bruce I. & Royal Skousen. 1994. Productivity and the English past tense: Testing Skousen's analogical model. In Susan D. Lima, Roberta Corrigan & Gregory K. Iverson (eds.), The Reality of Linguistic Rules, 193–218. Amsterdam / Philadelphia: John Benjamins.
Eddington, David. 2000. Analogy and the dual‐route model of morphology. Lingua 110. 281–298.
Eddington, David. 2002. A comparison of two analogical models: Tilburg Memory‐Based Learner versus Analogical Modeling. In Royal Skousen, Deryle Lonsdale & Dilworth B. Parkinson (eds.), Analogical Modeling, 141–156. Amsterdam / Philadelphia: John Benjamins.
Giegerich, Heinz. 1999. Lexical Strata in English: Morphological Causes, Phonological Effects. 40
Cambridge: CUP.
Keuleers, Emmanuel. 2008. Memory‐Based Learning of Inflectional Morphology. PhD. dissertation, Antwerpen: University of Antwerp.
Keuleers, Emmanuel & Dominiek Sandra. under revision. Similarity and productivity in the English past tense.
Lindsay, Mark. 2012. Rival suffixes: synonymy, competition, and the emergence of productivity. In Angela Ralli, Geert Booij, Sergio Scalise & Athanasios Karasimos (eds.), Morphology and the Architecture of Grammar. Proceedings of the 8th International Morphology Meeting, 192–203. Patras: University of Patras; URL: http://morbo.lingue.unibo.it/mmm.
Marchand, Hans. 1969. Categories and Types of Present‐Day English Word‐Formation. München: C.H. Beck.
Plag, Ingo. 2003. Word‐Formation in English. Cambridge: CUP.
Riddle, Elizabeth M. 1985. A historical perspective on the productivity of the suffixes ‐ness and ‐
ity. In Jacek Fisiak (ed.), Historical Semantics ‐ Historical Word‐Formation (Trends in linguistics. Studies and monographs 29), 435–461. Berlin et al.: Mouton de Gruyter.
Romaine, Suzan. 1983. On the productivity of word formation rules and limits of variability in the lexicon. Australian Journal of Linguistics 3(2). 177–200.
Skousen, Royal. 1989. Analogical Modeling of Language. Dordrecht: Kluwer.
Skousen, Royal. 1992. Analogy and Structure. Dordrecht: Kluwer.
Skousen, Royal. 2002a. An overview of Analogical Modeling. In Royal Skousen, Deryle Lonsdale & Dilworth B. Parkinson (eds.), Analogical Modeling, 11–26. Amsterdam / Philadelphia: John Benjamins.
Skousen, Royal. 2002b. Issues in Analogical Modeling. In Royal Skousen, Deryle Lonsdale & Dilworth B. Parkinson (eds.), Analogical Modeling, 27–48. Amsterdam / Philadelphia: John Benjamins.
41
Skousen, Royal. 2009. Expanding Analogical Modeling into a general theory of language prediction. In James P. Blevins & Juliette Blevins (eds.), Analogy in Grammar, 164–184. Oxford: OUP.
Skousen, Royal & Thereon Stanford. 2007. AM: Parallel : available from http://humanities.byu.edu/am/.
Skousen, Royal, Deryle Lonsdale & Dilworth B. Parkinson (eds.) (2002). Analogical Modeling. Amsterdam / Philadelphia: John Benjamins.
42