Using Language Invention to Teach Typology and
Cross-Linguistic Universals
Autore: Matt Pearson
Data della SM: 03-17-2017
Data FL: 04-01-2017
Numero FL: Florida-000043-00
Citazione: Pearson, Matt. 2017. “Using Language Invention
to Teach Typology and Cross-Linguistic
Universals.” Florida-000043-00, Fiat Lingua,
Diritto d'autore: © 2017 Matt Pearson. Questo lavoro è concesso in licenza
sotto un'attribuzione Creative Commons-
Non commerciale, senza derivati 3.0 Licenza non trasportata.
http://creativecommons.org/licenses/by-nc-nd/3.0/
Fiat Lingua è prodotto e gestito dalla Language Creation Society (LCS). Per maggiori informazioni
riguardo alla LCS, visita http://www.conlang.org/
Using Language Invention to Teach Typology
and Cross-Linguistic Universals
Matt Pearson
[email protected]
Reed College
Abstract
I discuss a project where students learn about language typology by creating a naturalistic con-
structed language. Students review cross-linguistic variation in natural languages (in areas such
as phoneme inventory, word order, case alignment, ecc.), and then determine which properties
their invented language will have. Decisions are made at random by spinning a wheel. A-
tached to the wheel is a pie chart, where the size of each slice represents the percentage of the
world’s languages possessing a given setting for some structural parameter or set of parame-
ters. Fondamentalmente, each decision constrains subsequent decisions in accordance with known impli-
cational universals: per esempio., in determining whether the language has prepositions or postpositions,
the pie chart is adjusted based on verb-object order in the language, as decided by a previous
spin of the wheel.
1. Panoramica
In this paper I describe a linguistics class project where students collaborate to create a typolog-
ically plausible constructed language (conlang), using a semi-randomized procedure to deter-
mine structural features such as phoneme inventory, word order, and case alignment. The goal
of the project is to investigate typology in a hands-on, creative way, using language invention to
teach students about cross-linguistic variation and the nature of implicational universals and
markedness hierarchies (Greenberg 1963, Hawkins 1983, Croft 2002, et al.). By working together
to invent a conlang that conforms to known statistical tendencies, students gain exposure to the
idea that natural languages do not vary randomly in their structure, but instead follow predict-
able patterns.
The idea for this project originated with Susan Curtiss as part of a UCLA undergraduate
honors course on mental grammar. When I was a graduate student at UCLA in the 1990s, IO
worked as a teaching assistant for this course and helped to develop the project as the basis for
an independent typology-themed discussion section, following a model established by a previ-
ous TA. In describing the project here, I have refined our original procedure somewhat to reflect
the increased availability of numerical data from large language samples, most especially the
World Atlas of Language Structures (WALS) online database (Dryer and Haspelmath 2013).
In section 2 I lay out the general parameters for the exercise, and discuss some of the
sources that instructors can use to generate questions and topics related to morpho-syntactic
variation. In section 3 I describe how the structural parameters of the conlang are determined
using a semi-randomized procedure that reproduces the statistical patterns expressed by impli-
cational universals. In section 4 I briefly discuss how students work to flesh out the conlang by
adding vocabulary, cultural details, eccetera.
2
2. Focus of the project
The project may be carried out over several class periods, or even an entire semester. It can form
a unit or mini-course within a larger discussion-based class, or the discussion section activity
within a lecture course. It can also be adapted for use outside the classroom as part of an extra-
curricular activity, independent study, or typology workshop.
The project is divided into units, with the number of units depending on the topic of the
class and the time available. Each unit focuses on a domain of cross-linguistic variation for
which language universals have been proposed. At the beginning of each unit, the instructor
and students review data from natural languages to learn about the range of cross-linguistic
variation, along with any implicational generalizations which have been proposed to capture
patterns of covariation. Some types of generalizations which might be covered, with an example
of each, are listed in (1):
(1)
UN. Phonological implicational universals: If a language has voiceless nasals and approxi-
mants, it will also have their voiced counterparts (Maddieson 1984).
b. Word order correlations: If a language has OV order, it is highly likely to have postpo-
sitions; if it has VO order, it is highly likely to have prepositions (Dryer 1992).
c.
Implicational universals related to grammatical categories: If a language has a grammati-
cal gender system, it will also make a number distinction (Greenberg 1963).
d. Markedness asymmetries: If a language has a paucal number, it will also have a dual
numero; if a language has a dual number, it will also make a singular/plural distinc-
tion. In languages that make grammatical number distinctions morphologically, plu-
ral marking involves at least as much morphology as singular marking, and du-
al/paucal marking involves at least as much morphology as plural marking. Questi
generalizations implicate a markedness hierarchy for number: PAUCAL > DUAL >
PLURAL > SINGULAR. (Croft 2002, Corbett 2000)
After learning about the patterns of cross-linguistic variation and markedness found in natural
lingue, students must then decide which grammatical features their conlang will have. Es-
sentially this involves generating a series of ‘design questions’ which must be answered—or, In
Generative Grammar terms, a series of parameters which must be set for the conlang. In (2) Essere-
low I list some possible topics for units, along with sample design questions for each unit. Questo
list is by no means exhaustive; it is merely meant to suggest the range of decisions that must be
made in developing a grammatically fleshed-out conlang.
(2)
UN. Phoneme inventory (Maddieson 1984):
• Which vowels will the language have and how will they be arranged in the vow-
el space? (E.g., will the language have a 3-vowel system, a 5-vowel system, ecc.?)
• Which places and manners of articulation will be distinguished among the con-
sonants?
• Will [VOICE] be a contrastive feature of consonants? Will [NASAL] be a contrastive
feature for vowels?
• Will length be a distinctive feature of consonants and/or vowels?
• Will the language have tones?
3
b. Syllable structure and phonotactic constraints (Blevins 1995):
• Will the language permit codas? (I.e., will the language include both open and
closed syllables, or open syllables only?)
• Will the language permit complex onsets? Se è così, what combinations of consonants
•
are allowed?
If coda consonants are permitted, will there be restrictions on which classes of
consonants can occur in coda position? Will the language allow complex codas?
c. Morphological types (Comrie 1989, Baker 1996, Nichols 1986):
• Will the language be analytic or synthetic? If it is synthetic, will it include fea-
•
•
tures of polysynthesis (such as noun incorporation)?
If the language is (poly)synthetic, will it have strictly agglutinating morphology,
or will it also include fusional and non-concatenative morphology (per esempio., templatic
morphology)?
If the language is (poly)synthetic, will it be primarily suffixing, primarily prefix-
ing, or a combination of both?
• Will the language have reduplication? If so, what functions will be associated
with reduplication (plurality, imperfective aspect, ecc.)?
• To express argument-structure dependencies (case roles), will the language make
use of head marking (indexation/agreement), dependent marking (morphologi-
cal case), both, o nessuno dei due?
d. Lexical and functional categories (Schachter and Shopen 2007, Harley and Ritter 2002):
• Will the language include a category of adjectives, as distinct from nouns and
verbs?
• Will the language have a category of adpositions, or will it express relational
concepts by other means?
• What sort of pronominal/phi-feature system will the language have? Which per-
son, numero, and gender/animacy distinctions will be made by the pronouns
and agreement markers?
• Will the language have articles?
• Will the language have a tense category? If so, what kinds of tense distinctions
will it make?
e. Basic constituent order (Greenberg 1963, Hawkins 1983, Dryer 1992, 2007):
• Will the language have OV or VO as its unmarked order for verb and object, o
•
•
•
will both orders be allowed (non-configurationality)?
If the language is VO, will the subject precede or follow the verb (SVO, VSO,
VOS)?
If the language has an adposition category, will it be prepositional or postposi-
tional?
If the language has an adjective category, will attributive adjectives precede the
modified noun (Adj-N) or follow the modified noun (N-Adj) within the noun
phrase?
• Will relative clauses precede the modified noun (Rel-N) or follow the modified
•
sostantivo (N-Rel)?
In possessive noun phrases, will the possessor precede the possessed noun (Poss-
N) or follow the possessed noun (N-Poss)?
• Will the language employ wh-movement or wh-in-situ in questions?
4
f. Marking syntactic dependencies (Comrie 1978, Palmer 1994, Andrews 2007):
•
•
•
•
If the language has morphological case and/or agreement for core arguments,
will it have an accusative alignment, an ergative alignment, a split-S alignment,
or a tripartite marking system?
If the language is ergative, will it be strictly ergative, split-ergative based on
tense/aspect, split-ergative based on a person/animacy hierarchy, ecc.?
If the language has case marking, what non-core cases, se presente, will it have (dative,
instrumental, locative, ecc.)?
If the language has grammatical number (and gender) categories, will attributive
adjectives and other modifiers agree in features with the noun they modify?
There are a number of useful resources for generating topics and design questions and identify-
ing related language universals. The WALS database website, wals.info (Dryer and Haspelmath
2013), includes a comprehensive list of articles on topics related to cross-linguistic variation.
More in-depth discussions of some of these topics can be found in typology textbooks and ref-
erence works such as Maddieson (1984), Shopen (1985, 2007), Comrie (1989), Payne (1997), Croft
(2002), Velupillai (2012), and Moravcsik (2013). The Lingua Descriptive Studies questionnaire
(Comrie and Smith 1977) provides a comprehensive list of topics and questions related to
grammatical structure. Although it is intended as an outline for field linguists writing a refer-
ence grammar of a natural language, this questionnaire can easily be repurposed for determin-
ing the grammatical features of a conlang. A similar questionnaire, but more condensed and
focused on morpho-syntactic topics, is included in Payne (1997). Finalmente, the University of Kon-
stanz’s Universals Archive offers an extensive database of proposed language universals, con
citations to works in which those universals are proposed or discussed. This archive can be ac-
cessed at http://typo.uni-konstanz.de/archive/ intro/index.php.
3. Deciding on the features of the language
As originally conceived by Susan Curtiss, the major structural properties of the conlang are de-
cided at random by spinning a wheel. The intention behind this choice was, in parte, to introduce
a game-like element of chance into the creation of the language. By leaving certain decisions up
to chance, students are forced to exercise their creativity within predetermined structural con-
straints, avoiding any temptation to make their conlang either overly familiar and English-like,
or overly alien and exotic (either of these would defeat the purpose of an exercise focused on
learning about natural language typology).
At UCLA we used a carnival-style wheel built for this purpose by Curtiss’s husband.
The wheel was mounted vertically on a wooden frame with a flexible pointer attached to the
top. When the wheel was spun, the pointer would brush against nails protruding from the cir-
cumference of the wheel until it came to rest at a particular point (cfr. the spinning wheels from
the game shows Wheel of Fortune and The Price is Right). For instructors who lack the resources
to build or purchase a carnival wheel, a large spinner from a board game like Twister can also be
used, or an online application such as Wheel Decide: http://wheeldecide.com. (Other random selec-
tion methods are also possible, such as rolling a set of multi-sided dice from a role-playing
game. Qui, Tuttavia, I focus on using a wheel for this purpose.)
For each grammatical parameter to be specified, a pie chart is attached to the spinner
wheel with slices representing features or settings for the parameter in question. The size of
each slice is determined by the rough proportion of the world’s languages that have that pa-
rameter setting (in combination with other parameter settings, where relevant). When the pro-
ject was originally developed, these proportions were estimated. Ora, Tuttavia, statistics from
5
the WALS database (Dryer and Haspelmath 2013) or similar sources may be used for this pur-
pose. For each decision to be made, a student volunteer comes to the front of the room and
spins the wheel. Whichever slice of the pie chart the pointer lands on determines the parameter
setting that the conlang will have.
Per esempio, suppose that the students are studying case/agreement systems and need
to determine which core alignment system their conlang will have—that is, how subjects of in-
transitive clauses (S), subjects of transitive clauses (UN), and objects of transitive clauses (P) Volere
be differentiated. Among natural languages, the four most common alignments are accusative (UN
and S are marked the same way while P is marked differently), ergative (P and S have the same
marking while A is marked differently), tripartite (UN, S, and P are all marked differently), e
split-S (more ‘agentive’ S arguments are marked the same way as A arguments, while more ‘pa-
tientive’ S arguments are marked the same way as P arguments). (See Comrie 1978, Dixon 1994,
Palmer 1994, Coon et al. 2017.) The relative prevalence of each alignment type among the
world’s languages is suggested by the numbers in (3). The N values in the second column of the
table are taken from the WALS sample of languages with morphological marking of core case
roles (Dryer and Haspelmath 2013). In the third column these N values have been converted
into approximate whole-number percentages (for the sake of simplicity, languages with align-
ments other than accusative, ergative, tripartite, and split-S are excluded). Using these percent-
età, we can construct a pie chart where each slice represents a different alignment type, COME
shown in Figure 1. This chart is attached to the wheel, the wheel is spun, and whichever slice of
the pie the pointer lands on determines which alignment the conlang will have.
(3)
nominative-accusative (A+S vs. P) 52
32
ergative-absolutive (A vs. S+P)
4
tripartite (A vs. S vs. P)
4
split-S (A+SA vs. P+SP)
TOTAL 92
N approx
%
57
35
4
4
100
Alignment
Tripartite Split-S
ERG-
ABS
NOM-
ACC
Figura 1. Spinner wheel for determining core case alignment
6
The probability that the conlang will end up with a given feature as a result of a random spin is
determined by the relative size of the corresponding pie slice. In the case of the alignment pa-
rameter, the wheel in Figure 1 ensures a greater chance that the pointer will land on the accusa-
tive setting rather than the ergative setting, and a greater chance that it will land on the ergative
setting rather than the tripartite or split-S settings. In questo modo, the probability that the conlang
will end up with a given alignment roughly mirrors the probability that a randomly chosen
natural language will have that alignment.
Fondamentalmente, as each parameter of the conlang is set, the probabilities governing subse-
quent decisions are adjusted based on known statistical correlations. In altre parole, each spin
of the wheel determines which pie charts will be used for subsequent spins, providing a graphic
illustration of how parameter settings in natural languages are interdependent. For example,
suppose that previous spins of the wheel have determined that the conlang has a default order
for verb and object in basic transitive clauses, either OV or VO, and that it has an adposition
category. Will the conlang have prepositions or postpositions? Dryer (1992) and others have
shown that, overwhelmingly, OV languages are postpositional while VO languages are preposi-
tional. The table in (4) gives data supporting this generalization, with N values taken from
WALS (Dryer and Haspelmath 2013). The number and percentage of OV languages in the
WALS database with postpositions (Postp) is vastly larger than the number and percentage
with prepositions (Prep); conversely, the VO languages with prepositions vastly outnumber
those with postpositions.
(4)
OV and postpositions
OV and prepositions
TOTAL
VO and postpositions
VO and prepositions
TOTAL
N
472
14
486
42
456
498
approx
%
97
3
100
8
92
100
The strong statistical correlation between verb-object order and adposition type can be ex-
pressed as a pair of biconditional universals: OV « Postp and VO « Prep.
As a consequence of this correlation, the pie chart for determining adposition type in the
conlang will look very different depending on how the OV/VO parameter had been set by pre-
vious spins of the wheel. If the conlang has OV order, the pie chart for determining adposition
type would look like Figure 2 below, resulting in a 97% chance that a random spin of the wheel
will yield postpositional order. D'altra parte, if the parameter has been set to VO, the pie
chart for determining adposition type would look like Figure 3. Qui, a random spin of the
wheel would have a 92% chance of landing on the preposition setting.
7
OV
Prep
Postp
Figura 2. Spinner wheel for determining adposition type (language is OV)
VO
Postp
Prep
Figura 3. Spinner wheel for determining adposition type (language is VO)
A somewhat different type of word order correlation is exemplified by the table (5), showing
the relationship between OV/VO order and the position of a relative clause with respect to the
noun that it modifies (N values from Dryer and Haspelmath 2013). COME (5) shows, VO languages
overwhelmingly place the relative clause after the noun (N-Rel). OV languages, on the other
hand, show no such correlation: they are roughly equally likely to have either order (N-Rel or
Rel-N) (cfr. Dryer 1992).
(5)
OV and Rel-N
OV and N-Rel
TOTAL
VO and Rel-N
VO and N-Rel
TOTAL
N approx
%
54
46
100
1
99
100
132
113
245
5
416
421
8
In the case of adposition type, the correlation with verb-object order is bidirectional. If a lan-
guage has OV order it is highly likely to have postpositions, e viceversa, if it has postposi-
tions it is highly likely to have OV order; likewise for VO order and prepositions. In the case of
relative clauses, Tuttavia, the correlation with verb-object order is only partial. If a language has
VO order, it is overwhelmingly likely to have postnominal relative clauses (N-Rel), whereas no
prediction can be made regarding the placement of relative clauses in OV languages. On the
other hand, if a language has prenominal relative clauses (Rel-N), it is overwhelmingly likely to
also have OV order, while no prediction on the order of verb and object can be made for lan-
guages with postnominal relatives. This can be expressed as a pair of unidirectional implica-
tional universals: VO ® N-Rel and Rel-N ® OV. In both cases the converse implication does
not hold.
Figures 4 and 5 below show the pie charts, derived from the data in (5), which would be
used when spinning the wheel to determine the order of noun and relative clause in the con-
lang. If the conlang is OV, there is a roughly equal chance that the pointer will land on N-Rel
(46%) or on Rel-N (54%). If the conlang is VO, Tuttavia, the probability that the pointer will
land on N-Rel is overwhelming (99%).
OV
N-Rel
Rel-N
Figura 4. Spinner wheel for determining order of noun and
relative clause (language is OV)
VO
Rel-N
N-Rel
Figura 5. Spinner wheel for determining order of noun and
relative clause (language is VO)
9
This exercise gives students the opportunity to investigate language universals in an original
and hands-on manner. Typological research has shown that natural languages do not vary arbi-
trarily in their structure, but can instead be grouped into types (word order types, morphologi-
cal types, ecc.) based on shared formal properties. By exploring how the setting of each structur-
al parameter in their ‘naturalistic’ conlang influences subsequent parameter settings, students
learn how the statistical tendencies described by implicational universals give a language its
overall grammatical ‘shape’. Inoltre, they are invited to speculate about why certain
grammatical features tend to cluster together in natural languages while others do not—
whether this reflects external functional constraints on language design, arbitrary properties of
Universal Grammar (UG), or other factors.
4. Fleshing out the conlang
A language is more than just a set of categories and parameter settings: it also has a vocabulary,
a cultural context, eccetera. It is in working together to flesh out of these aspects of the conlang that
students are able to exercise their collective creativity.
When we carried out this exercise at UCLA, grammatical morphemes were typically in-
vented in class, as needed. For example, if successive spins of the wheel determined that the
language had ergative and absolutive case suffixes, the instructor would solicit suggestions for
the phonological shape of the suffixes from the students. We would also coin lexical roots as
needed during class sessions so that we could practice the features of the conlang by translating
example sentences from English.
To further expand the vocabulary, students were required to propose five new roots per
week (with definitions) and submit their suggestions to the instructor. Back when this exercise
was developed, we asked students to include a list of roots as part of their weekly homework
assignments. Nowadays, to avoid students suggesting different lexical items for the same con-
cept, vocabulary suggestions could be uploaded to an online platform, either a course website
or a wiki. For each round of submissions, proposed roots are reviewed by the instructor, or a
designated student, to make sure they conform to the phonological constraints of the language
(per esempio., to make sure that only permitted phonemes and syllable types are used). Approved roots
are then added to the dictionary for the language, which is updated and distributed to students
once a week.
In addition, instructors can assign students to invent derivational morphemes (or deri-
vational processes such as reduplication and ablaut), and practice creating new lexical items
from pre-existing roots. Including productive derivation allows the vocabulary to expand
quickly by giving students and instructors the means to coin new words on the fly.
Finalmente, students can be invited to expand the grammar and cultural background of the
conlang through term paper projects. Per esempio, students may be required to invent and de-
scribe some grammatical feature of the conlang which was not specified as part of the group
project (the description could take the form of a sample chapter from a reference grammar of
the language, which explains the feature concisely and illustrates it with glossed example sen-
tences). In alternativa, students could be asked to construct a sample text in the language (per esempio., UN
folk tale, dialogue, ecc.), including a free English translation, interlinear glosses, and supple-
mental vocabulary. The need to write texts in the conlang, or translate texts from English, may
lead students to propose and discuss additional grammatical constructions not covered in class.
10
5. Riepilogo
I have described a collaborative project where students work together to invent a typologically
plausible language. The phonological, morphological, and syntactic features of the language are
determined at random by spinning a wheel to decide among a set of alternatives (per esempio., for core
argument alignment, the alternatives might be accusative, ergative, tripartite, and split-S). In each
case the alternatives are weighted in accordance with their prevalence among the world’s lan-
guages, as extrapolated from numerical data taken from the WALS database and other large
language samples. In this way, the probability that the conlang will end up with a particular
grammatical feature as a result of a random spin approximates the probability that an arbitrari-
ly chosen natural language will have that feature.
In cases where different grammatical features correlate with one another, the result of a
given spin will determine how the alternatives for subsequent spins are weighted. Per esempio,
since adposition type correlates with the default order of verb and object, the probability that
the conlang is assigned prepositional (vs. postpositional) order as a result of a random spin of
the wheel is adjusted based on whether the conlang is OV or VO, as determined by a previous
rotazione. Through this exercise, typology students are able to study implicational universals
firsthand by seeing how those universals play out in constraining the development of a ‘natu-
ralistic’ conlang grammar.
Riferimenti
Andrews, UN. 2007. The major functions of the noun phrase. Language Typology and Syntactic De-
scription (2a ed.), vol. IO: Clause Structure, ed. T. Shopen, pp. 132-222. Cambridge: Cambridge
University Press.
Panettiere, M. 1996. The Polysynthesis Parameter. Oxford: la stampa dell'università di Oxford.
Blevins, J. 1995. The syllable in phonological theory. The Handbook of Phonological Theory, ed. J.
Goldsmith, pp. 206-244. Oxford: Blackwell.
Comrie, B. 1978. Ergativity. Syntactic Typology, ed. W.P. Lehmann, pp. 329-394. Austin: Universi-
ty of Texas Press.
Comrie, B. 1989. Language Universals and Linguistic Typology (2a ed.). Chicago: University of
Chicago Press.
Comrie, B. e n. Fabbro. 1977. Lingua descriptive studies: Questionnaire. Lingua 42: 1-72.
Coon, J., D. Massam, and L. Travis, eds. 2017. The Oxford Handbook of Ergativity. Oxford: Oxford
University Press.
Corbett, G. 2000. Number. Cambridge: Pressa dell'Università di Cambridge.
Croft, W. 2002. Typology and Universals (2a ed.). Cambridge: Pressa dell'Università di Cambridge.
Dixon, R.M.W. 1994. Ergativity. Cambridge: Pressa dell'Università di Cambridge.
Dryer, M. 1992. The Greenbergian word order correlations. Language 68: 81-138.
Dryer, M. 2007. Word order. Language Typology and Syntactic Description (2a ed.), vol. IO: Clause
Struttura, ed. T. Shopen, pp. 61-130. Cambridge: Pressa dell'Università di Cambridge.
Dryer, M. and M. Haspelmath, eds. 2013. The World Atlas of Language Structures Online. Lipsia:
Max Planck Institute for Evolutionary Anthropology. [http://wals.info]
11
Greenberg, J. 1963. Some universals of grammar with particular reference to the order of mean-
ingful elements. Universals of Human Language, ed. J. Greenberg, pp. 73-113. Cambridge, MA:
CON Premere.
Harley, H. and E. Ritter. 2002. A feature-geometric analysis of person and number. Language 78:
482-526.
Hawkins, J. 1983. Word Order Universals. New York: Academic Press.
Keenan, E. 1985. Relative clauses. Language Typology and Syntactic Description (1a ed.), vol. II:
Complex Constructions, ed. T. Shopen, pp. 141-170. Cambridge: Pressa dell'Università di Cambridge.
Maddieson, IO. 1984. Patterns of Sounds. Cambridge: Pressa dell'Università di Cambridge.
Moravcsik, E. 2013. Introducing Language Typology. Cambridge: Pressa dell'Università di Cambridge.
Nichols, J. 1986. Head-marking and dependent-marking grammar. Language 62: 56-119.
Palmer, F.R. 1994. Grammatical Roles and Relations. Cambridge: Pressa dell'Università di Cambridge.
Payne, T. 1997. Describing Morphosyntax: A Guide for Field Linguists. Cambridge: Cambridge Uni-
versity Press.
Schachter, P. and T. Shopen. 2007. Part-of-speech systems. Language Typology and Syntactic De-
scription (2a ed.), vol. IO: Clause Structure, ed. T. Shopen, pp. 1-60. Cambridge: Cambridge
University Press.
Shopen, T., ed. 1985. Language Typology and Syntactic Description. Cambridge: Cambridge Uni-
versity Press.
Shopen, T., ed. 2007. Language Typology and Syntactic Description (2a ed.). Cambridge: Cam-
bridge University Press.
Velupillai, V. 2012. An Introduction to Linguistic Typology. Amsterdam: Giovanni Benjamin.
