Volume 8 Issue 1 (2010)
Note: Linguistic Discovery uses Unicode characters
to represent phonetic symbols. Please see Optimizing Display
for requirements to accurately reproduce this page.
Analyzing Semantic Maps: A Multifactorial Approach
Andrej L. Malchukov
In this paper I argue that semantic similarity is not the only
factor which motivates polysemy patterns cross-linguistically; I also show that
these other factors (markedness, distinguishability, etc) may give rise to
polysemies which are problematic for established semantic maps. Only when these
other interfering factors, both functional and structural, are featured out,
does a semantic network emerge and a “similarity map” reduce to a
There are two general approaches to semantic maps, which are
also represented in this issue. These approaches are sometimes characterized as
traditional vs. statistical approaches, as implicational vs. probabilistic maps
(Wälchli, this issue), or as first- vs. second-generation maps
(Sansò, this issue). In the traditional approach (conceptual)
categories/functions are preselected, and a network of categories is constructed
in such a way that the arrangement reflects formal similarities between adjacent
categories. The connections between categories, represented in the form of
lines, are usually taken as indicative of semantic overlap, which can be
captured in terms of shared semantic components (e.g., features; cf. Zwarts,
this issue). The traditional approach has been developed by Anderson, Croft,
and Haspelmath, among others; the details are well known and need not be
rehearsed here. In the alternative approach (maybe in its clearest form
represented in the work by Wälchli, this issue) the categories are not
preselected, and semantic maps are automatically generated from parallel corpora
through the use of statistical scaling methods.
As noted by Cysouw (2007), the traditional approach faces a number of
problems. First, it cannot represent frequencies of individual polysemy
A related problem is that
as the amount of data increases vacuous maps become more and more widespread
since frequent, rare, and exceptional patterns will all be represented on the
map. Second, the traditional method is overgenerating since not all the
predicted patterns are actually
. The most common response
to the first problem by practitioners of the traditional approach is that
semantic maps reflect most frequent polyfunctionality patterns in a certain
domain; thus, exceptional patterns would be featured out in a larger sample. For
example, Narrog & Ito (2007) do not represent polysemies of (instrumental)
case markers found in less than 10% of cases. There may be more principled
methods to distinguish between recurrent and exceptional patterns; for example,
Rice and Kabata (2007) use Fisher’s exact test to determine which patterns
of case polysemy (involving allative markers) are statistically significant. In
other words, those polysemy patterns that are rare and have been overlooked in
the early work on semantic maps, based on small scale comparison, are likely to
be featured out in a larger sample as statistically insignificant. Indeed, maps
proposed in earlier work are often confirmed in a larger sample. Thus, Narrog
& Ito’s (2007) study largely corroborated Haspelmath’s (2003)
map of instrumental and related functions, and Mauri’s (this issue) map
of coordination is “slightly different” from Malchukov’s
(2004) map of contrast markers as far as same categories are addressed. From
this perspective, exceptional polysemies do not constitute a major problem for
the semantic map approach because they would not make it into the map. While
this is a legitimate approach, it involves data reduction; so the question
arises as to what extent such data reduction is justified. This question can
only be answered by analyzing motivations behind common vs. rare polysemy
patterns. In this paper I will show that rare patterns are often not indicative
of (immediate) semantic relatedness of respective categories, but are due to
other factors. This does not mean that such minority patterns should be
dismissed; on the contrary, they deserve to be analyzed in their own right as
they can provide important insights into motivations behind polyfunctionality.
Moreover, the problem of overgeneration, which has not been addressed in the
literature so far, can arguably also be solved by invoking other factors
inhibiting broader polysemies (see Section 4 for further discussion).
Another response to limitations of the traditional approach was the
introduction of statistical scaling techniques (such as multidimensional
scaling) that can represent relative frequencies of individual polysemy patterns
more readily and involve less data reduction (Wälchli, this issue). Yet,
as pointed out in the literature (Cysouw 2007; Zwarts, this issue, Narrog, this issue), the latter method has its own limitations. First, unlike traditional
maps, the maps generated by multidimensional scaling (MDS) are not
implicational, thus they cannot be used to constrain the data and therefore are
“less interesting” (in this sense, they are more similar to vacuous
maps in the traditional approach), nor can they readily represent a diachronic
dimension (Narrog, this issue). Moreover, automatically generated maps are
often difficult to interpret (which is somewhat ironic, given that they have
been developed as a visualization tool) because the emerging picture is quite
“messy” (this might be partly a consequence of the compression of
multiple dimensions into the two-dimensional space; Cysouw 2007). Thus, it is
often not clear whether there is a certain clustering pattern in the first two
dimensions (cf. question-marked clusters in Sansò’s contribution),
and turning to further dimensions makes identification of clusters still more
problematic (e.g., Wälchli, this issue, finds only the first three
dimensions interpretable in his dataset). Thus, while MDS maps visualize general
semantic dimensions or tendencies in a certain domain, they are less suited for
capturing semantic connections between individual categories. Therefore, unlike
traditional maps, MDS maps cannot directly feed semantic analysis. One reaction
to this is Wälchli’s (2007; this issue) proposal that conventional
semantic analysis (including formal semantics) should give way to
“similarity semantics”. A less radical approach, advocated in this
paper, is that semantic maps (whether constructed in a conventional way or
automatically generated) can provide an important insight into semantics of the
analyzed categories once the interfering factors that introduce
“noise” are featured out. Only at this later stage can the semantic
map feed the semantic analysis, aiming to account for the recurrent polysemy
patterns in terms of shared semantic components. In this way semantic maps can
become an important tool for semantic analysis and not just a visualization
technique (a frequent criticism heard from opponents of the semantic map
In what follows I discuss a number of semantic maps proposed for
different domains in the literature, including my own earlier work. The
discussion will be intentionally programmatic as my goal is not to present the
data, but rather to raise methodological questions relevant to the semantic map
approach. In particular, I will argue that similar encoding may not reflect a
semantic similarity but may be due to other factors, both functional
(markedness, economy, distinguishability), structural, as well as diachronic.
Only when these other interfering factors are featured out does a semantic
network emerge. Thus, it might be useful to distinguish terminologically between
similarity map (e.g., obtained through MDS visualization technique) and
a semantic map, which emerges once the “noise” is reduced. A
similarity map is shaped by the cumulative effect of different factors
contributing to polyfunctionality. A semantic map, on the other hand,
represents, so to say, a semantic residue of the similarity map once the other
factors are featured out.
Baerman et al. (2005) is a recent study of inflectional
syncretism, in general, and of case syncretism, in particular. Since syncretism
depends on a number of factors both functional and phonetic, it is not
surprising that the patterns of syncretism are not consistent across languages.
Thus, Baerman et al. (2005) conclude that only patterns involving syncretism of
core cases are cross-linguistically consistent. Another generalization noted by
Baerman et al. (2005) is that case syncretism shows strong dependency on number
(in fact this is the only case of a strong interaction between inflectional
categories noted for the nominal domain). Not surprisingly, case syncretism is
found in non-singular numbers more frequently when compared to the singular. The
following example from Sanskrit, where syncretism is more pervasive in the
plural than in the singular and is more pervasive in the dual than in the
plural, is representative in that respect.
Table 1: Sanskrit
a-stem noun (Baerman 2009, citing
Does this particular pattern of case syncretism reflect
semantic similarities between individual cases? Not obviously so—while
nominative and vocative forms may be regarded as constituting a natural class,
there is no clear semantic motivation for the syncretism of the dative and
instrumental, or dative and ablative cases. Moreover, semantic similarities
cannot explain why syncretism is more pervasive in non-singular numbers than in
the singular. Indeed, there is no obvious reason why semantic roles should be
better distinguished for individuals rather than groups.
The main contributing factor to the syncretism pattern exemplified by
Sanskrit is well-known and has been described in terms of markedness. It has
long been noted in the literature that inflectional possibilities of the marked
member of a category are reduced, compared to the unmarked member (Greenberg
1966:27; Croft 1990; Croft 2003:77-80). Croft, following Greenberg, even regards
such skewing as one of markedness diagnostics, which he dubs “inflectional
markedness”. Now, markedness is not an unproblematic notion and can
probably be reduced to other factors such as frequency (as suggested by
); yet effects of
inflectional markedness are pervasive and can be found in different domains. To
cite another example from the nominal domain: recent work on syncretism patterns
in pronominal systems has revealed that neutralization of person categories
(“vertical homonymies”) occur more frequently in the non-singular
forms of both free pronouns and bound pronouns (Cysouw 2003; Siewierska 2004;
Baerman et al. 2005). Similar examples abound in the verbal domain as
well—for example, tense-aspect-mood distinctions as found in (unmarked
affirmative) verbal forms are frequently neutralized in the negative forms (see
Croft 1990, Aikhenvald & Dixon 1998 for more discussion and exemplification
of markedness effects).
The question is what consequences do the effects of inflectional
markedness have for semantic maps. It seems that if semantic maps are designed
to capture similarities in semantic functions, such markedness effects should be
featured out since syncretization of inflectional forms in the marked member is
not driven by semantic similarity in the first place.
3. Other Functional Factors:
Economy and Distinguishability
There is a tradition of representing alignment patterns
(ergative, accusative, neutral) in the forms of diagrams reminiscent of semantic
maps. The first linguist to use these schemes was apparently Charles Fillmore,
but since the 1970’s, in typological studies (by Bernard Comrie, R.M.W.
Dixon, Frans Plank and A.E. Kibrik, among others) it has become conventional to
represent alignment patterns in such a way. Thus typology textbooks (e.g.,
Whaley 1997; Song 2001) include the following representations of different
alignments: accusative (A=S≠P; Figure 1), ergative (A≠S=P; Figure
2), neutral (A=S=P; Figure 3), and tripartite patterns (A≠S≠P;
Originally, these schemes were not conceived of as semantic
maps. For example, Kibrik (1985; cf. Plank 1985), discusses different alignment
patterns from a multi-factorial perspective. While he admits that proper
encoding of semantic roles is an important factor, he also acknowledges the role
of other factors such as economy or distinguishability of arguments. Yet, more
recently the alignment schemes have been explicitly reformulated in terms of
semantic maps. Thus, Croft (2001:137, 147) proposed the following semantic map
for the encoding of core arguments.
Figure 5: Croft’s conceptual space for core arguments
This semantic map conveniently captures the variation of
alignment patterns across languages, both in the monotransitive and the
ditransitive domain. In the latter domain the distinction is made between
languages with a direct vs. indirect object distinction (“indirective
alignment”; schematically, T=P≠R), and languages with a primary vs.
secondary object distinction (“secundative alignment”;
schematically, T≠P=R) (Dryer 1986; Haspelmath 2004; Siewierska 2004).
Given standard assumptions about the well-formedness of semantic maps, this
approach correctly predicts marginality of alignment types which would display
discontinuous segments on the map (such as the anomalous S≠A=O pattern).
Thus the following two maps represent two of the possible alignment types
complying with the semantic map: the accusative “indirective”
language such as German (see (1)), and an ergative language with a
“secundative” alignment such as Eskimo (see (2)).
Ich gab ihm ein Buch
‘I (NOM) gave him (DAT) a book’
Eskimo (West Greenlandic; Fortescue 1984:88)
‘(He) gave Nisi money’
Figure 6: Alignment maps: German
Figure 7: Alignment
More similar alignment schemes can be found in Dryer (2007),
who, however, does not call them maps.
Yet, the alignment map in Figure 5 is not unproblematic since we find
exceptions both in the monotransitive and ditransitive domain. Thus, the
anomalous Double-Oblique pattern (S≠A=O), also called “horizontal
alignment”, is cross-linguistically rare but is attested in a number of
Iranian languages (see (6) from Vafsi below). Another type of violation, this
time in the ditransitive domain, is found in Spanish. As is well-known, Spanish
displays differential object marking (DOM) in the monotransitive
domain—that is, P, if animate, has to be marked by the dative-accusative
a, but remains unmarked if inanimate. By contrast, the
object-theme in the ditransitive domain always remains unmarked (as in (4));
note that marking an animate T by the preposition
a makes the
construction (5) unacceptable:
Spanish (Company 2003:234)
‘The teacher introduced his wife to his pupils’
‘The teacher introduced his wife to his pupils’
Since the intransitive subject is equally unmarked, the
resultant pattern violates contiguity because T shares the same encoding with S,
but not with P. How can one account for such a counterexample? One feature of
the ill-formed pattern in Spanish is that it involves a zero marking. Now,
distribution of zero markers is arguably governed by considerations of economy
and distinguishability and need not reflect a semantic affinity. Note that
marking an (animate) T with the “dative” preposition results in a
pattern where both R and T are introduced by the same preposition (see the
unacceptable pattern in (5)) thus violating distinguishability (see Malchukov
2008 for an optimality-theoretic account of the Spanish pattern along these
lines). Economy considerations are equally important: arguably, absence of
marking is not as strong an indicator of semantic relatedness as overt marking,
and thus zero markers incur violations on the map. Another instructive example
of the role of economy concerns the distribution of the direct case in languages
with a two-term case system, recently discussed by Arkadjev (2005; 2009).
Arkadjev observes that in some of these languages (languages of the
“distributing” type, in his terms) distribution of the direct case
(as opposed to the general oblique case) is problematic for the semantic map
approach. In these languages (e.g., some Pamir languages) the direct case is
found, apart from the subject, also on nouns in peripheral functions such as
temporal or locative, which are not connected to the subject function on his
map. Given that the direct case is typically unmarked, economy provides a ready
explanation for such cases. Indeed, in many languages nouns denoting temporal
notions or place names remain unmarked, but this does not implicate semantic
similarity with the subject; rather it indicates that the respective roles are
recoverable in the absence of marking (see Malchukov & de Swart 2009 for
further discussion of economy and distinguishability effects in case marking;
cf. Aristar 1997). Thus, distribution of zero markers need not reflect a
semantic similarity but may be rather a matter of economy/recoverability of
4. Distinguishability as an
Viewing alignment schemes as semantic maps may be
problematic for another reason as well. In particular, it is not clear whether S
is a unitary category in semantic terms or not, and thus it may be misplaced on
the semantic map of core (macro)roles. Indeed, in semantic terms, it is more
adequate to distinguish between agentive (or, unergative) subjects (Sa) and
patientive (or, unaccusative) subjects (Sp). Given this distinction we can
arrive at the following configuration A – Sa – Sp –P, which
makes more sense semantically. In particular, this configuration allows us to
represent different kinds of split intransitive systems which group together A
with Sa and/or P with Sp (Dryer 2007; cf. Dixon 1994; Song 2001). A more radical
solution is to exclude S from the semantic map altogether since it does not
correspond to one role. Note that this exclusion is also justified for another
reason as well. As noted above, S is usually unmarked, and thus its distribution
will be less informative for the arrangement of functions on the semantic map.
The latter approach is taken by Malchukov & Narrog (2009), who propose the
following semantic map (conceptual space) for the case roles from the non-local
Figure 8: The semantic map (conceptual space) for major case
relations (non-spatial domain)
In addition to agent (A), patient (P), theme (T), and
recipient (R), as familiar from alignment maps, the map in Figure 8 introduces
new case functions, mostly from a non-spatial domain: instrument (cf.
the dog with a stick
), material/means (cf.
He built the house with
), beneficiary (cf.
I built him a house), possessor (cf.
saw his house
), source (cf.
He returned from the store), goal (cf.
He returned to the store), comitative (cf.
He returned with a
). The upper part of the map in Figure 8 linking agents to instruments
and causes, on the one hand, and to genitives, on the other, represents
different reductions of the agentivity prototype (Grimm 2005). Thus, possessors
and datives share sentience property with agents but lack the feature of
instigation, while instruments are instigating entities but lack sentience. The
lower part of the map is reminiscent of the semantic map for the ditransitive
domain in Figure 5, even though the T role is not conceived as restricted to
ditransitive themes but also pertains to monotransitive themes (they differ from
patients in that they do not necessarily involve a change of state in the course
of a verbal event). Note that in contrast to the alignment maps in Figures 1-5,
the proposed map does not include a separate S function since S does not
correspond to a single (macro)role. The upper and lower parts of the map are
linked by two “routes”, one leading from Instrumental to Theme, and
another from Possessive to Recipient marking. The former route is mediated by
the means/material function (cf.
load the cart with bricks), which shows
similarities to instrument, on the one hand (it may be conceived of as an
instigating entity), and to the theme, on the other (standing in an incremental
relation to the verb). This accounts for the fact that the means/material
function may pattern either with instruments or with themes across languages or
indeed in the same language, resulting in a well-known “spray/load
load bricks on the cart vs.
load the cart with
). Finally, one of the conceptual underpinnings of the
possessive-benefactive-recipient connection may be the “possession as
goal” scheme, as suggested by Heine (1997; see 6.2 for further
The proposed map has both a semantic plausibility and also finds
empirical support in the most frequent polysemy patterns as described in the
contributions to Malchukov & Spencer (eds.) (2009) dealing with individual
cases: the dative-allative polysemy, as familiar from English (recall the
to), is common across languages (Creissels 2009; Næss
2009), the dative-genitive polysemy is attested in many Australian and
Austronesian languages but is also found elsewhere (Lander 2009; Næss
2009); the genitive-ablative polysemy is especially common in languages using
adpositions for these functions (Heine 2009, Lander 2009); the dative-accusative
polysemy is familiar from languages with differential object marking (Næss
2009; Malchukov & de Swart 2009); instrumental-accusative polysemy is
typical for languages with secundative alignment (Kittilä & Malchukov
2009); the instrumental-comitative polysemy is the most frequent polysemy
pattern involving both cases (Narrog 2009, Stolz et al. 2009); finally, the
ergative-instrumental and ergative-genitive polysemies are identified as the two
most frequent polysemy patterns involving ergative case (Palancar 2009). Thus
the proposed map captures the major polysemy patterns of case markers in the
In light of the previous discussion, let us now consider the anomalous
double-oblique pattern (S≠A=O), which is problematic for the alignment
maps. The double-oblique pattern is cross-linguistically rare, yet it is
attested in a number of Iranian languages (Payne 1980; Bossong 1985; Stilo 2004;
Arkadjev 2005). This pattern is illustrated by an example from Vafsi, where A is
marked by the oblique case in the past tense, and (prominent) P is likewise
marked by the oblique.
Vafsi (Stilo 2004:244)
‘The fox carried off (the) chicken’
Given that both A and P are marked (by the oblique case) and
S is unmarked, this pattern incurs a contiguity violation of the alignment maps
in Figures 1-5. Indeed, this pattern seems to be puzzling as it violates all the
functional motivations behind case marking—it does not reflect semantic
similarity, nor does it distinguish between arguments and is not economical.
However, this pattern is conceivable in light of diachronic data. As noted in
the literature (Kerimova & Rastorgueva 1975; Arkadjev 2005), the double
oblique pattern in these languages results from meaning extensions of the
originally polyfunctional dative-genitive case. This syncretic case developed
into the marker of a (prominent) object, on the one hand, and to the ergative
marker in past tenses, on the other hand. Note that both developments are not
unusual. Ergative markers of genitive origin are common in ergative languages
(e.g. in Eskimo), while the DAT to ACC shift constitutes a well-known
grammaticalization path, familiar from languages with differential object
marking (such as Hindi; Bossong 1985; Lehmann 1995). Thus, the double-oblique
pattern is due to a polysemy chain, with individual polysemy patterns
well-attested elsewhere. Schematically:
If ERG = GEN, GEN = DAT, DAT = ACC, then ERG = ACC, resulting in a
Note that although this scenario cannot be represented on
the alignment map in Figure 5, it can be easily captured on the semantic map for
case functions in Figure 8. Thus, though a map violation in this case is
apparent, it disappears once further functions (which are absent on the
alignment map in Figure 5 but present in the case function map in Figure 8) are
taken into account. This however also provides a (partial) explanation as to why
this pattern is less frequent: it involves a “long-distance”
polysemy. Note that in relative terms the languages displaying an
“extended” polysemy should be rarer than languages displaying
“restricted” polysemies contributing to the polysemy chain. Thus, a
polysemy chain x=y=z, will be found in a subset of languages which display
restricted polysemies x=y and y=z. Thus, while the traditional approach to
semantic maps cannot (and is not intended to) capture absolute frequencies of
individual polyfunctionality patterns, it can straightforwardly account for
relative frequencies in extended vs. restricted polysemy chains.
In addition, there are other functional reasons why this polysemy
pattern involving A and P should be infrequent. Note first that semantic
similarity cannot account for such polysemy because semantic similarities
support polysemies of adjacent (connected) categories on the map, while
non-adjacent categories need not share any common semantic features. In our
case, similar encoding of A and P does not reveal any semantic commonality
between these (macro)roles. More importantly, the resultant pattern fails to
distinguish between subjects and objects. Similar effects of distinguishability
as an inhibiting factor for extended polysemies can be found elsewhere. Thus,
while extension of instrumental case marker to As (resulting in an ergative
pattern) and to Ts (resulting in a secundative pattern) is not uncommon,
languages which show both polysemies are virtually unattested. Indeed, extension
of the instrumental marker to subjects and objects simultaneously would violate
distinguishability. In fact, the few languages which combine ergative alignment
for monotransitives with the secundative alignment for ditransitives use
different cases for subjects and objects. Thus, Eskimo (see (2)) uses the
Instrumental (“modal”) case for T arguments but uses genitive
(“relative”) case for As, so no distinguishability violation arises.
Instrumental polysemy involving the agent and comitative functions might provide
another instructive example. Both polysemy patterns of instrumentals are widely
attested (Stolz, 2009, Narrog, 2009), but the combined polysemy involving all
three functions is not found (Stolz 1996; Palancar 2009) although it is
perfectly compatible with the map and thus predicted to occur.
Distinguishability or recoverability of functions is again a likely explanation
for this restriction. Indeed, in a language showing such extended polysemy, an
animate NP in the “instrumental” case would be regularly ambiguous,
allowing both agentive and comitative interpretation. Note that such ambiguity
is not found with instruments since the latter are typically inanimate. Similar
distinguishability effects can be found in the domain of object encoding as
well. Note that while extension of the dative case to P is common, further
extension of the same marker to T is disfavored as it leads to ambiguity (recall
the discussion of the Spanish examples (4)-(5)). It is to be expected that
extended polysemies in other domains would also be disfavored due to other
factors such as distinguishability
Thus, invoking other functional factors can not only account for unexpected
polysemy patterns, but also explain why certain (extended) patterns permitted by
the map are not attested (recall the problem of overgeneration, raised by
5. Structural Factors and
Semantic maps can be produced not only for grammatical
markers but also for verbs types of different valency (cf. Wälchli, this issue, for motion verbs), or, more generally, for different constructions (cf.
Cristofaro, this issue). Malchukov, Haspelmath, and Comrie (forthcoming, henceforth
MHC) propose the following semantic map (conceptual space) for the domain of
ditransitive constructions. In addition to the ditransitive recipient theme
I gave him a book), it includes a number of other
three-argument constructions (see Margetts & Austin 2007 for an overview):
the malefactive theme construction (cf.
I stole a book from him), patient
beneficiary construction (cf.
I built him a house), internal possessor
I saw his house), external possessor construction (cf.
On mne slomal nogu lit. ‘He broke me the leg’),
theme goal construction (cf.
I put/threw a newspaper on the table), and
patient instrument construction (cf.
I hit the dog with a stick), among
The claim embodied in the semantic map in Figure 9 is that categories
adjacent on that map will share a similar construction (valency pattern), and
the same map can constrain diachronic extensions of certain constructions. In
Figure 9, some of the extensions of the basic ditransitive constructions in
three languages (Finnish, Eskimo, and Jamul Tiipay) are represented, in order to
illustrate basic alignment types: (i) indirective alignment (cf. the extension
of the allative case in Finnish), (ii) secundative alignment (cf. the extension
of the instrumental case in Eskimo), and (iii) neutral alignment (cf. the domain
of the double object construction in Jamul Tiipay):
Figure 9: Basic ditransitive constructions in Jamul Tiipay,
Finnish, and Eskimo
In the present context it is interesting to consider
extensions of the double object construction across languages. Kittilä
(2006), which is one of the very few studies of lexical variation in
ditransitive constructions, concluded that ‘give’ is one of the
verbs which is most likely to be found in a double object construction. There
are good reasons for this preference. As argued by Kittilä,
‘give’ scores high on Hopper & Thompson’s (1980)
transitivity parameters. For example, unlike ‘send’,
‘give’ implies a successful transfer; hence, not only the theme but
also the recipient is affected by the action. Furthermore, issues of
distinguishability/recoverability seem to be relevant as well (as argued in
MHC). Insofar as T is typically inanimate and R animate, the respective roles
are easily recoverable and need not be distinctively marked. Note that verbs
involving two animate participants like ‘introduce’ are less
commonly used in a double object construction cross-linguistically.
Kittilä (2006) notes some counterexamples to the preferential use
of ‘give’ in a double object construction. Some of these can be
explained in terms of the same functional factors invoked above. Thus, verbs
taking a malefactive source, like ‘steal’, arguably outrank
‘give’ verbs in affectedness (of the third participant); therefore,
they are likely to be found in the double object construction (see the domain of
double object constructions in Jamul Tiipay) and even may be preferably eligible
for this construction, compared to ‘give’ verbs (as, for example, in
Mandarin Chinese; see MHC for discussion). Some other exceptions, however, are
puzzling and defy a functional explanation. Thus, in Malayalam,
‘give’ takes a dative construction, while “less canonical
ditransitives” (Asher & Kumari 1997:205) like ‘entrust’
and ‘feed’ take a double object construction:
Malayalam (Asher & Kumari 1997:205)
‘The child gave me the pen.’
‘I fed the cows grass.’
One would be hard pressed to provide a functional
explanation for the eligibility of these particular verbs for the double object
construction in Malayalam, yet this can be easily explained in structural terms.
As it turns out, both ‘entrust’ and ‘feed’ are causative
verbs, and causatives (of transitives) take a double object construction in
Malayalam. While cross-linguistic preference of causatives for the double object
construction in its turn might need a separate explanation (see MHC for some
proposals), it should be admitted that in the case of Malayalam the choice of
the (double object) pattern is due to structural factors (morphological
structure) and is not directly related to semantic similarities (between
‘feed’ and ‘entrust’ to the exclusion of
‘give’). So while evaluating semantic maps such structural factors
should be featured out in order to account for apparent exceptions.
Another interfering factor discussed in MHC concerns verbal polysemy and
pattern inheritance. Consider the case of Chechen (Nakh-Daghestanian), where
‘hit’ unexpectedly takes the dative-allative pattern, which it
shares with caused motion verbs.
Chechen (Bickel & Nichols 2009)
‘Father stabbed the sheep with the knife.’
Note that this pattern causes a discontinuity on the map in Figure 9
since the allative strategy is found both with ‘hit’-verbs and
‘throw/put’-verbs, but is not found with the intermediate type of
‘load’-verbs (which frequently align either with the former class or
with the latter class, or with both, which leads to a familiar “spray-load
alternation”). Now, as is clear from the glosses, the verb in the
‘hit’-construction is also used as a contact verb (‘strike
at’) elsewhere. This latter use naturally provides a motivation for the
use of the allative-dative pattern, which is frequent with contact and caused
motion verbs. Similar cases where the pattern of ‘affect’-verbs
(like ‘hit’) is modeled on caused motion verbs are attested
elsewhere. For example, in Ewe, the verb
da, which is used for
‘hit’, is used as a caused motion verb ‘throw’ elsewhere
(Essegbey 1999:166). In one
this extension of
the allative strategy into the instrumental domain is accompanied by a semantic
shift (e.g., from strike’/‘throw’ to ‘hit’; see
Malchukov 2005 for a general discussion of pattern polysemy and pattern
inheritance in case marking). Thus polysemy of individual verbal lexemes can
introduce inconsistencies into the semantic map because the case frame would be
motivated only by one of the meanings of the verbal lexeme. Such semantic shifts
are problematic for semantic maps as they may incur a contiguity violation. In
our case, either one needs to recognize a violation or postulate an extra link
connecting ‘hit’ and ‘put’ verbs. While this is always
possible (but undesirable since additional connections lead to less predictive
maps) it also loses the generalization that this connection obtains when the
verb itself is polysemous. Again, this violation is due to an interfering factor
manifesting the pressure for an analogical syntactic behavior on the part of
polysemous items. This factor can be factored out on the map by checking for
polysemies of individual verbs.
In this section we consider some diachronic factors which
may cause violations of the contiguity of semantic maps. It should be noted that
although the previous discussion has been mostly couched in synchronic terms, it
is well known that semantic maps can be used to represent a diachronic dimension
as well (for discussion of “dynamicized” semantic maps see Van der
Auwera & Plungian 1998; Haspelmath 2003; Narrog, this issue). Viewed
diachronically, polysemies arise from meaning extensions of individual
categories. In most cases, such meaning extensions are gradual and proceed
stepwise along the network of functions on a map without incurring a contiguity
violation (Croft et al. 1987). However, there are more complex diachronic
scenarios, to be discussed in this section, which might be problematic for
semantic maps. We will discuss three such cases: polygrammaticalization of
lexical items, reinterpretation through reanalysis, and gram replacement.
effects on semantic maps
Klamer (fc.) discusses grammaticalization of motion verbs in
the Papuan languages Teiwa and Kaera—these verbs are in the process of
developing into the “oblique” case marker. Thus, in Teiwa, the
ma (cf. Kaera
mi) is used to mark goals (as in (10)) and
instruments (as in (11)).
Teiwa (Klamer fc.)
‘The people went to do (that) then put (it) in a
‘Someone cuts wood with a machete.’
Such polysemy is quite unusual typologically and is likely
to incur a violation on a semantic map for case functions. Note that the goal
and instrument are not contiguous on the map for syntactic case functions in
Figure 8 above (or on similar maps of case functions proposed by Haspelmath 2003
and Narrog, this issue). The origin of this marker provides a straightforward
explanation for this puzzling polysemy: as shown by Klamer, this marker
represents early stages of grammaticalization of the verb
‘come’. Note that at early stages, grammaticalization is likely to
be more context-dependent (Hopper & Traugott 1993:89), and lexical items
involved in grammaticalization can develop into different markers in different
constructions (a process sometimes referred to as
“polygrammticalizaton”; Craig 1991). In this particular case, the
motion verb has grammaticalized into an oblique marker when combined with
inanimates, while in a construction with animates it is rather used as a modal
marker or a conjunction (cf. the first use of
ma in (10) above). It is
clearly futile to try to find a common denominator, or even shared components,
for all these disparate functions. Thus, results of polygrammaticalization of
lexical items are more likely to cover regions inconsistent with the map than
those which represent further meaning extensions (“lateral” meaning
shifts). One approach to handling these counterexamples would be to exclude from
consideration cases where a grammatical item derives from a lexical one. At
least in some cases this exclusion seems to be motivated as it is still possible
to argue that we are dealing with different contextual uses of a lexical item.
In fact, this is a position ultimately adopted by Klamer (fc.), who argues that
these disparate functions of
ma in Teiwa are better conceived of as
different contextual uses of the single intransitive motion verb with an
unspecified argument position.
Thus, early grammaticalization is likely to incur violations of the
established semantic maps. A legitimate question is why we do not encounter such
cases more frequently, which might eventually undermine the semantic map
enterprise. The reason is that polygrammaticalization of a lexical item is
construction sensitive and requires a number of preconditions. Thus, for Kaera
and Teiwa, Klamer notes the following factors which allowed
polygrammaticalization: productiveness of verb serialization; non-distinction of
finite and non-finite verbs; lack of three-valent verbs. Given that such
preconditions are likely to be areally/genetically restricted,
polygrammaticalization effects would not produce a consistent pattern across
6.2 Diachronic factors:
In a paper discussing the connections between benefactive
and possessive domain, Daniel & Malchukov (forthcoming) propose the
following map (simplified here), which can be seen as an elaboration of the map
in Figure 8 from Malchukov & Narrog (2009) (cf. also Haspelmath 2003;
Malchukov, Haspelmath & Comrie (forthcoming); Narrog, this issue for maps for the
Figure 10: The possessive-benefactive
This map “zooms in” on the
possessive-benefactive domain of case relations—additional functions,
which do not appear on the map in Figure 8, include purpose (cf.
We went out
) and external possessor/experiencer (as in Figure 9); the latter
functions will not concern us here. In the present context it is important that
goal and possessor are not adjacent on that map; rather, the connections are
mediated through the intermediate functions of recipient and beneficiary (the
other route involving external possession is ignored here). The extension from
goal marking through the recipient (or purpose) function to beneficiary and
further into the domain of attributive possession is not exceptional across
languages. Heine (1997) regards it as one of the major encoding patterns
(“schemas”) for encoding possessors—the so-called goal schema
(Y exists for/to X > X has/owns Y). The goal schema is more frequent in the
domain of predicative possession (cf. French:
Ce chien est à moi
‘This dog is mine (lit. to me)’), but is also occasionally extended
to attributive possession as well (Heine 1997).
It is important to keep in mind that the term “goal schema”
is potentially misleading since Heine’s goal, comprises recipients and
beneficiaries as well. In accordance with the semantic map, partial polysemies
between adjacent categories should be more widely attested. Indeed, we find many
cases where the same marking is used for goal and recipient (cf. English
to), or for recipient and beneficiary (e.g. Russian dative), or for
beneficiary and possessor (e.g. -
paj in Imbabura Quechua). Occasionally,
we find “longer” polysemy chains, as when the allative marker in
Finnish extends to recipients and beneficiaries (cf. Figure 9 for an extension
of the allative strategy with different verbs types) or when possessive
classifiers in Austronesian are extended to beneficiaries and recipients (Song
2005). In (Classical) Persian the purposive postposition
rādiy ‘for the sake of’) spread to beneficiaries and
possessors, on the one hand, and to recipients and further to patients, on the
other hand (Bossong 1985:60 ff.; Hopper & Traugott 1993:158). Importantly,
all these configurations are compatible with the semantic map in Figure 10 (see
Narrog, this issue, for more
). Occasionally, we find
still broader polysemies which include goal, recipient, beneficiary, and
possessor. One example is Budukh, a Daghestanian language where the
“alienable genitive” is used to mark recipients and goals.
Budukh (Authier, fc.)
‘He gave me the knife (temporarily)’
Authier (fc.) argues that the “alienable
genitive” originally had a locative function (better preserved in the
genealogically related Kryz), and its extension to the adnominal domain followed
from a reinterpretation of ditransitive constructions. The important point here
is that semantic evolution/extension is gradual: allative extends first to
dative function, then dative extends to genitive. As can be readily seen, these
extensions are in accordance with the map in Figure 10.
In Khwarshi (another Daghestanian language) a seemingly similar pattern
is attested (Khalilova, 2009). Also in Khwarshi the same case in
lo is found both on possessors and goals, yet the situation is
more complex in two respects. First, the use of this case in the goal function
is restricted to some contact verbs (like ‘touch’,
‘strike’), and is used with animate goals exclusively (with
inanimate goals another case, contessive, is used). Second, the use of
lo in the genitive function is also restricted insofar as it is used to
modify oblique rather than absolutive arguments (Khwarshi, as other Daghestanian
languages, is ergative). A third puzzling point is that in Khwarshi, the same
case is found on (some) goals and (some) possessors, but neither of the
intermediate functions (recipient, beneficiary) seems to be involved. How does
this polysemy come about? This is clearly not an extension of the allative
marker, as proposed for Budukh. Indeed, its goal use is very restricted, so the
possessive use is more likely to be the source function. One obvious explanation
which accounts for peculiarities of this construction is that the
“oblique” genitive is used in the allative function due to the
omission of the head noun referring to a body-part (cf.
body‑CONT in (15)).
Khwarshi (Zaira Khalilova and Raisat Karimova, p.c.)
‘I colored her (with the paint).’
‘I colored her leg (with the paint).’
Thus, the genitive-goal construction as in (14) developed
from the construction in (15) where the goal is in the contessive case and the
possessor predictably takes the oblique genitive (as it modifies an oblique
argument). A similar explanation has been proposed by Comrie et al. (2007) for
Tsez where the oblique genitive can also encode goals under similar conditions.
Thus, unlike what we observed in Budukh, the goal-possessor polysemy in
Khwarshi is not due to a meaning extension but rather results from reanalysis.
In such cases polyfunctionality need not comply with the map. Yet as in the case
of polygrammaticalization discussed above, such changes are likely to be less
common since they presuppose certain conditions. In the case of Daghestanian
languages (Khwarshi and Tsez), availability of a special genitive case used for
oblique arguments arguably made it easier for the body part to be omitted
because its meaning is still recoverable. Note also that not all cases of
reanalysis are problematic for semantic maps. Thus, reanalysis of the
recipient-beneficiary to possessor involving “internalization” of
beneficiaries into an NP (as posited for Budukh by Authier) or reanalysis in the
opposite direction involving “externalization” of possessors (as
posited for Austronesian by Song 2005), do not violate the map.
6.3 Semantic map violations due
to gram replacements
Sadanobu and Malchukov (forthcoming) discuss the evolution of the
aspecto-temporal forms in Japanese, in particular, the perfect-continuous form
teiru and the past form in
ta. One puzzling point that has
attracted much attention in the literature (see Sadanobu and Malchukov (forthcoming) for
an overview) is that
ta can perform two seemingly opposite functions:
apart from the general past meaning as in (16), it can also be used for emphatic
(or “mirative”) present, as in (17):
‘Yesterday it rained for three hours.’
‘Oh, look, there is a wild boar over there!’
at the sudden sight of a wild hog during an excursion in the
While the polysemy confined to past tense and present
mirative meanings is virtually unattested across languages, it is conceivable in
light of the diachronic data. Historically
ta derives from the
tari, which was subsequently reanalyzed as a general
past form. Yet, it has retained the residual mirative functions related to its
original resultative function. Note that although
ta is related to
perfect historically, synchronically, it does not qualify as a perfect marker.
As shown in (16) it can freely combine with time adverbials (a context more
characteristic of imperfects cross-linguistically) and is furthermore banned
from some contexts characteristic of perfects
. The perfect
function is currently performed by
teiru (which also has continuous uses
elsewhere). Thus the puzzling polysemy of
ta in Modern Japanese can be
attributed to the fact that the perfect meaning has been partially
teiru, which is a “young perfect”
replacing the “old perfect” in
ta in its core function. This
scenario is depicted in the Figures 11 and 12 below representing the
hypothesized semantic evolution of the
These figures may be viewed as a representation of a
grammaticalization path, as familiar from the grammaticalization literature (see
Bybee et al. 1994 for grammaticalization of perfects-anteriors) but also as a
semantic map incorporating a diachronic dimension. Note that the map in Figure
12 is ill-formed since the domain of
ta-marking is no longer contiguous
on the map. This illustrates a well known case of contiguity violation
conditioned by gram replacement. Similar examples have been cited in the
literature for different grammatical domains, including voice (Croft et al.
1987), mood (Van der Auwera & Plungian 1998), tense (Haspelmath 2003), and
clause combining (Malchukov 2004).
As noted in the literature, such processes of gram replacement present a
challenge for semantic maps because, viewed synchronically, they violate
contiguity of semantic maps (see also Zwarts, this issue). In my view, such
exceptions can be dealt with on the assumption that the new grams replacing the
old ones are structurally “heavier”, that is are morphologically
more complex. This is obviously true for Japanese but seems to hold for most
other cases discussed in the literature (e.g. renewal of reflexive markers in
the middle domain discussed in Croft et al. 1987, renewal of the present case in
Turkish, discussed by Haspelmath 2003:236, and renewal of the concessive
conditional function in Russian, discussed in Malchukov 2004). Thus even in the
absence of historical evidence, we can account for such exceptions by a simple
metric comparing structural complexity of the “offending” category
to the adjacent discontinuous grams (the discussion of complexity metrics is
discussed in another connection by Mauri, this issue).
Again a question arises as to why such violations are not more common
cross-linguistically given that grammaticalization is a pervasive process, and
grams are often driven through the same grammaticalization paths. Yet it is
important to keep in mind that such renewal processes do not necessarily lead to
a contiguity violation. First, such violations can occur only if a
grammaticalization path is “branching” in a certain domain. Second,
at early stages, extension of a new gram usually yields an overlap rather than
total replacement. On the other hand, at a later stage of grammaticalization,
the “new” gram will replace the “old” one in the
extended functions as well so that no contiguity violations arise.
Thus, in general, semantic maps are better suited to represent the most
common scenarios involving meaning extensions of particular grams, while more
complex scenarios, involving gram interaction are more likely to yield
contiguity violations. Yet, as has also been observed with regard to
polygrammaticalization and reanalysis, the latter scenarios are more complex,
because they require more preconditions. Therefore such violations are predicted
to be rare and can be featured out on a large scale semantic map.
In this paper I have argued that semantic similarity is not
the only factor which motivates polysemy patterns cross-linguistically; I have
also shown that these other factors (markedness, distinguishability, etc.) might
give rise to polysemies problematic for the established semantic maps (i.e.
induce contiguity violations). It should be noted (as pointed out by S.
Cristofaro, p.c.) that it is not the case that these other factors always yield
contiguity violations on the established semantic maps. Also, I am not claiming
that patterns originating from these other factors should always contradict a
particular semantic map. In some cases a contiguity violation can be remedied by
introducing further functions and/or connections (cf. the discussion of the
agent-patient polysemy pattern found in some Iranian languages in Section 4).
Yet, although the correlation is not absolute, it is certainly true that
ill-formed (discontinuous) patterns on semantic maps are most often determined
by factors other than semantic affinity. Note that an extended polysemy in
itself challenges the general assumptions behind the semantic map approach
insofar as (non-adjacent) categories involved in extended polysemies need not
share a common semantic component (recall again the discussion of the double
oblique pattern in Iranian
On the other hand,
although the effects of other factors need not always be problematic for an
established semantic map, they do not consistently support the configuration
established on the basis of semantic affinities either.
Thus, semantic affinity is one of the factors which conditions
cross-linguistically recurrent polysemies, but is not the only factor. The other
factors which may give rise to identical encoding include markedness, economy,
distinguishability, as well as structural and diachronic factors. Yet, compared
to semantic affinity, the effects of other factors are less pervasive and less
consistent for a number of reasons. Thus, markedness effects, by definition, are
confined to neutralization in the marked category, while economy considerations
regulate distribution of the zero-marked grams. The role of distinguishability
as a factor contributing to polyfunctionality is also limited to specific
contexts (recall the discussion of Spanish ditransitives in Section 3), while
elsewhere it rather favors different encoding of categories which might be
otherwise confused (recall the inhibiting function of distinguishability
discussed in Section 4). The effects of structural factors and polysemy
patterns, as observed in individual languages, are not totally random, yet they
do not easily generalize cross-linguistically and therefore do not yield a
consistent pattern. Finally, violations of the semantic maps can occur under
certain diachronic scenarios (polygrammaticalization of a lexical item,
reanalysis, gram replacement), but these scenarios are more complex—they
require certain preconditions and are therefore expected to be more rare as
compared to the gradual meaning extensions consistent with the map.
Thus, if we conceive of a semantic map approach as a method for
uncovering semantic relationships between categories, these interfering factors
should be considered as “noise” which should be featured out before
the semantic map can feed the semantic analysis. Below we have outlined certain
heuristics as to how this can be achieved:
- Consider polysemy patterns of the unmarked member to check
for effects of “inflectional
- Semantic maps should be
restricted to cases of overt marking; distribution of zero-markers may be driven
by economy/recoverability rather than by
- Check (code) for structural factors
to reduce “noise”;
- Check for
polysemy patterns of verbs because a similar syntactic pattern can be due to
analogical extension (pattern
- Exclude items having both
grammatical and lexical functions to check for polygrammaticalization effects
which can lead to discontinuities;
- Check for
“heaviness” (structural complexity) of “offending”
markers, compared to the adjacent discontinuous grams to check for possibility
of gram replacement.
Once the effects of interfering factors are featured out, a
similarity map (as obtained through multidimensional scaling, for
example) reduces to a
semantic map, representing a semantic residue of
the similarity map. At the next stage, the semantic map can feed the semantic
analysis, aiming at identifying common semantic components of (adjacent) related
functions. On the other hand, minority patterns introducing “noise”
should be also analyzed in their own right as they can provide important insight
into the different factors contributing to polyfunctionality. To conclude:
further advances in the semantic map approach crucially depend not only on
developments of new representation (visualization) techniques but also on
advances in the analysis of the established polyfunctionality patters (see also
Zwarts, this issue, Cristofaro, this issue, for related proposals).
Aikhenvald, A.Y. and R.M.W. Dixon. 1998. Dependencies between
grammatical systems. Language 74.56-80. doi:10.2307/417565
Aristar, A.R. 1997. Marking and hierarchy types, and the
grammaticalization of case-markers. Studies in Language 21.313-68. doi:10.1075/sl.21.2.04ari
Arkadjev, P. 2005. Dvuxpadezhnye sistemy v indoiranskix jazykax:
tipologicheskaja perspektiva (Two-term case systems in Iranian languages in a
typological perspective). Indoiranskie jazyki i tipologija jazykovyx situacij,
ed. by N. Kazanskij. St. Petersburg.
-----. 2009. Poor (two-term) case systems: limits of neutralization. In Malchukov & Spencer (eds.) Handbook of Case: OUP, 686-700.
Asher, R.E. and T.C. Kumari. 1997. Malayalam. London:
Authier, G. to appear. Eléments de la langue kryz, dialecte
d’Alik. Paris: Société de Linguistique de
Baerman, Matthew. 2009. Case syncretism. In Malchukov & Spencer (eds.), 219-231
Baerman, M., D. Brown and G. Corbett. 2005. The syntax-morphology
interface: A study of syncretism. Cambridge: Cambridge University Press.
(Cambridge Studies in Linguistics 109).
Bickel, B. and J. Nichols. 2009. Case marking and alignment.
Malchukov and Spencer (eds.), 304-322.
Bossong, G. 1985. Differenzielle Objektmarkierung in den
neuiranischen Sprachen. Tübingen: Narr.
Bybee, J., R.D. Perkins and W. Pagliuca. 1994. The evolution of
grammar: Tense, aspect and modality in the languages of the world. Chicago:
University of Chicago Press.
Company-Company, C. 2003. Transitivity and grammaticalization of
object: The struggle of direct and indirect objects in Spanish.Romance objects:
Transitivity in Romance languages, ed. by G. Fiorentino, 217-261. Berlin:
Comrie B., M. Polinsky and R. Rajabov. 2007. Tsezian Languages.
Unpublished manuscript, Max Planck Institute for Evolutionary
Craig, C. 1991. Ways to go in Rama: A case study in
polygrammaticalization. Approaches to grammaticalization, ed. by E.C. Traugott
and B. Heine, vol. 2, 455-493. Amsterdam: Benjamins.
Creissels, D. 2009. Spatial cases. Malchukov and Spencer (eds.),
Croft, W., H. Shyldkrot and S. Kemmer. 1987. Diachronic semantic
processes in the middle voice. Papers from the 7th international conference on
historical linguistics, ed. by A.G. Ramat, O. Carruba and G. Bernini, 179-192.
Croft, W. 2001. Radical Construction Grammar. Oxford: Oxford
Cysouw, M. 2003. The paradigmatic structure of person marking.
Oxford: Oxford University Press
-----. 2007. Building semantic maps: The case of person marking.
New challenges in typology: Broadening the horizons and redefining the
foundations, ed. by M. Miestamo and B. Wälchli, 225-247. Berlin: de
Daniel, M. & A. Malchukov. in preparation. Possessive encoding
of beneficiaries: A cross-linguistic study.
Dixon, R.M.W. 1994. Ergativity. Cambridge: Cambridge University
Dryer, M. 1986. Primary objects, secondary objects, and antidative.
Language 62.808-845. doi:10.2307/415173
-----. 2007. Clause types. Language typology and syntactic
description, 2nd edition, ed. by T. Shopen, vol. 1, 224-275. Cambridge:
Cambridge University Press.
Fortescue, M. D. 1984. West Greenlandic. London. Croom
Grimm, S. 2005. The lattice of case and agentivity. MA Thesis.
University of Amsterdam.
Haspelmath, M. 2003. The geometry of grammatical meaning: Semantic
maps and cross-linguistic comparison. The new psychology of language, ed. by M.
Tomasello, vol. 2, 211-243. Mahwah, NJ: Erlbaum.
-----. 2004. Explaining the ditransitive person-role constraint:
A usage-based account. Constructions 2. Available online at
-----. 2006. Against markedness (and what to replace it with).
Journal of Linguistics 42/1.25-70 doi:10.1017/s0022226705003683
Heine, B. 1997. Possession: Cognitive sources, forces and
grammaticalization. Cambridge: Cambridge University Press.
-----. 2009. Grammaticalization of cases. Malchukov and Spencer
(eds) , 458-470.
Hopper, P.J. and E.C. Traugott. 1993. Grammaticalization.
Cambridge: Cambridge University Press.
Kerimova, A. A. and V.S. Rastorgueva. 1975. Kategorija padezha [The
category of case]. Opyt istoriko-tipologicheskogo issledovanija iranskix
jazykov [A Diachronic-Typological Study of the Iranian Languages], ed. by V.S.
Rastorgueva, vol. 2, 117-199. Moscow: Nauka.
Kibrik, A.E. 1985. Toward a typology of ergativity. Grammar inside
and outside the clause, ed. by J. Nichols and A. Woodbury, 268-324. Cambridge:
Cambridge University Press.
Khalilova, Z. 2009. Grammar of Khwarshi. PhD
Kittilä, S. 2006. The anomaly of the verb ‘give’
explained by its high (formal and semantic) transitivity. Linguistics
Kittilä, S and A. Malchukov. 2009. Varieties of accusative.
Malchukov and Spencer (eds.), 549-562.
Klamer, M. forthcoming. One item, many faces: ‘Come’ in
Teiwa and Kaera. Typological and Areal Analysis: Contributions from East
Nusantara, ed. by M. Ewing and M. Klamer. Leiden: KITLV Press.
Lander, Y. 2009. Varieties of genitive. Malchukov and Spencer
Lehmann, C. 1995 . Thoughts on grammaticalization.
Malchukov, A. L. 2004. Towards a semantic typology of adversative
and contrast marking. Journal of Semantics 21/2.177-198. doi:10.1093/jos/21.2.177
-----. 2005. Case pattern splits, verb types, and
construction competition. Competition and variation in natural languages: The
case for case, ed. by M. Amberber and H. de Hoop, 73-117. Oxford:
-----. 2008. Animacy and asymmetries in differential case
marking. Lingua 118.203-221. doi:10.1016/j.lingua.2007.02.005
Malchukov, A. and P. de Swart. 2009. Differential case marking and
actancy variation. Malchukov and Spencer (eds.), 339-356.
Malchukov, A. and H. Narrog. 2009. Case polysemy. Malchukov &
Spencer (eds.), 518-535.
Malchukov, A., M. Haspelmath and B. Comrie. forthcoming.
Ditransitive constructions: A typological overview. Studies in ditransitive
constructions, ed. by A. Malchukov, M. Haspelmath and B. Comrie. Berlin:
Malchukov, A. and A. Spencer (eds.). 2009. The Oxford handbook of
case. Oxford: Oxford University Press.
Margetts, A and P. Austin. 2007. Three-participant events
in the languages of the world: Towards a crosslinguistic typology. Linguistics
Næss, Å. 2009. Varieties of dative. Malchukov and
Spencer (eds.), 572-581.
Narrog, H. 2009. Varieties of instrumental. Malchukov and Spencer
Narrog, H. and S. Ito. 2007. Reconstructing semantic maps: The
comitative-instrumental area. Sprachtypologie und Universalienforschung
Ogihara, T. 1999. The semantics of tense and aspect in Japanese.
The Handbook of Japanese Linguistics, ed. by Natsuko Tsujimura, 326-348. Oxford:
Palancar, E. L. 2002. The origin of Agent markers. Berlin:
Akademie-Verlag. (Studia Typologica 5).
-----. 2009. Varieties of ergative. Malchukov and Spencer
Payne, J.R. 1980. The decay of ergativity in Pamir languages.
Lingua 51.147-186. doi:10.1016/0024-3841(80)90005-4
Plank, F. 1985. The extended accusative/restricted nominative in
perspective. Relational typology, ed. by F. Plank, 269-311. Berlin:
Rice, S. and K. Kabata. 2007. Crosslinguistic grammaticalization
patterns of the allative. Linguistic Typology 11.451-514 doi:10.1515/lingty.2007.031
Sadanobu, T. and A. Malchukov. forthcoming. Evidential extensions of
aspecto-temporal forms in Japanese from a typological perspective. Cahiers
Chronos (Proceedings of the Chronos conference, Antwerp, 2006).
Siewierska, A. 2004. Person. Cambridge: Cambridge University
Song, J.J. 2001. Linguistic typology: Morphology and syntax.
-----. 2005. Grammaticalization and structural scope increase:
Possessive-classifier-based benefactive marking in Oceanic languages.
Linguistics 43/4.795-838. doi:10.1515/ling.2005.43.4.795
Stilo, D. 2004. Vafsi folk tales. Wiesbaden:
Stolz, T. 1996. Some instruments are really good companions: On
syncretism and the typology of instrumentals and comitatives. Theoretical
Linguistics 23/1-2.113-200. doi:10.1515/thli.1997.23.1-2.113
Stolz, T., C. Stroh and A. Urdze. 2009. Varieties of comitative.
Malchukov and Spencer (eds.), 601-609.
van der Auwera, J. and V.A. Plungian. 1998. Modality’s
semantic map. Linguistic Typology 2/1.125-139. doi:10.1515/lity.19188.8.131.52
Whaley, L. 1997. Introduction to typology: The unity and diversity
of language. Thousand Oaks, CA: Sage Publications.
Wälchli, B. 2006/7. Constructing semantic maps from parallel
text data. Unpublished manuscript.
Acknowledgement: I am
grateful to Sonia Cristofaro for useful comments on the draft version of this
article. The usual disclaimers apply.
Note that although the
traditional approach as it stands cannot represent
of individual polysemy patterns, it does capture
relative frequencies of
extended vs. restricted patterns (see Section 4 for further discussion).
It is important to keep
in mind that once the configuration of the semantic map is established it gives
rise to predictions which go beyond the dataset on the basis of which the
semantic map has been established. It is exactly this feature that makes
(traditional) maps theoretically interesting and warrants the term
“implicational maps” (Haspelmath 2003). Yet, this advantage turns
into a disadvantage if certain patterns predicted to be possible (i.e. which are
compatible with a semantic map) never turn up, even in extended samples.
Thus one could argue
that fewer distinctions are made in the marked category because marked
categories are less frequent compared to unmarked and are therefore less
resistant to analogical leveling leading to syncretism.
See also Wälchli,
this issue, for a similar observation.
The patterns of
syncretism in person paradigms can provide another example of the role of
distinguishability as an inhibiting factor. While in the person paradigms both
“horizontal” syncretism patterns (involving number neutralization)
and “vertical” syncretism patterns (involving person neutralization)
are widely attested (Siewierska 2004; Cysouw 2003), these patterns are in
complementary distribution. Siewierska (2004: 100) attributes complementarity of
horizontal and vertical homophonies to the need to maintain a certain level of
explicitness within person paradigms.
interpretation (suggested by B. Bickel, p.c.) such cases still represent caused
motion or contact verbs, like ‘strike’, rather than affect verbs,
like ‘hit’. In this approach, no contiguity violation arises since
the verb in (9) would not qualify as ‘hit’. Yet, it is not clear
whether this latter interpretation can be upheld either in this particular case
or in general as it seems to exclude the possibility of a semantic shift. In the
case of Nakh-Daghestanian languages, for example, it may be relevant that in
Chechen the sentence in (9) denotes a successful, not an attempted, stabbing (Z.
Molochieva, p.c.). The same is true of the closely related Ingush (J. Nichols,
This map is largely
compatible with a diachronic map for dative-related functions in Narrog (this issue) as far as the same categories are concerned, but it is more restricted
in two respects: it does not postulate either a direct recipient-purpose
connection (which is not supported by the data) or a direct goal-benefactive
connection (which seems to be mediated by the recipient function in languages
cited as supporting evidence: English, Greek, Ik).
Thus, unlike perfect
forms in other languages
ta is not compatible with the adverb
‘yet’ (Ogihara 1999: 332).
Joost Zwarts (this issue) suggests that common semantic features may be lacking only in polysemy
chains on semantic maps with “cycles”. In my view, however, any
(non-adjacent) categories involved in a polysemy chain may lack common semantic
components, irrespective of the configuration of the semantic map.