Fred Weerman (Amsterdam): The rise and fall of grammatical gender

Although the way in which an abstract property like grammatical gender develops in IE languages may be more or less in accordance with a 'grammaticalization cline', this is hardly explanatory: it remains unclear why some (but not all) languages develop this characteristic, and why some (but not all) languages lose it again. Until recently it was only possible to speculate on these questions, but I will show that combining work from several linguistic sub disciplines (in particular theoretical linguistics, L1/L2 acquisition, typology, microvariation, historical linguistics) offers insightful answers, focusing on Germanic (and Dutch more in particular).

1. Gender features are not very likely candidates for a 'universal list'. I will assume that there is a discovery procedure that leads the language-learning child from lexical, distributional evidence to abstract features that feed the computational system. In fact, the lexical lexical and a more abstract 'rule based' side are clearly visible in gender systems. Although all kinds of (sub) regularities may play a role, it has to be stored that a particular noun or root in Germanic is, for instance <+neuter> or <+feminine>. And if grammatical gender does not go beyond determiner-noun agreement, it may in fact be argued that the entire phenomenon can be described in terms of rather superficial lexical rules. This becomes, however, much less likely as soon as features of grammatical gender pop up in several types of agreement (adjectival agreement, possessives, articles, determiners, ...). It can be shown that in Dutch L1 acquisition, the lexical side is relatively time consuming: it takes monolingual children over six years to find out which Dutch words are neuter (Modern (standard) Dutch has a twoway gender system: common versus neuter). It can be shown, however, that the abstract agreement rules (even the ones that are relatively opaque) are available in the same children at a very early age (cf. Blom, Polisenska & Weerman 2008 a.o.). Children are, in this respect, "little inflection machines" (Wexler 1998). Crucially, this is rather different in both adult and child L2 acquisition of (Dutch) gender. Although adults do make progress on the lexical side of gender, the type of abstracts features that are apparently relatively easily accessible in monolingual L1 acquisition, turn out to be hardly learnable in adult L2 acquisition. As a result, late learners of the Germanic gender system typically keep making overgeneralizations (in Dutch: common forms instead of neuter forms).

2. A first diachronic consequence of this state of affairs is that gender systems should not be simplified on the rule based side as long as their development is mainly dependent on monolingual transmission from one generation to another (since the rules do not seem to be problematic in any way for L1 children). Another prediction is that (robust) language contact will lead to changes on the rule based side of gender rather directly, in particular when language contact leads to (massive) L2 acquisition of the language in question. In addition, we predict which forms may survive in such a situation (namely the ones that are overgeneralized in the process of L2 acquisition). I will show that this is supported by evidence from diachronic developments in Dutch, English and (semi) creoles related to Dutch, like Afrikaans, Negerhollands and Berbice Dutch.

3. There is another diachronic prediction: if children are indeed "inflection machines" and are capable of acquiring abstract features much more easily than late learners, we may expect that they play an eminent role in the coming into being of a gender system. The implementation of abstract generalizations in complex spell-out rules may be due to their acquisition process. If this is the case a condition for the appearance of a gender system is the undisturbed successful transmission in (many) successive generations of L1 learners, where children reanalyse input and introduce abstract generalisations. That such a correlation makes sense for complex morphology in general is argued by Lupyan and Dale (2010), based on a corpus of over 2000 languages (their work also supports our conclusion in 2 that also the reverse holds, namely that contact leads to loss of inflectional properties). Focussing on the development of relatively new determiners (cf. Van der Velde 2011 and Weerman & Van der Velde 2014), I will show how indeed the interplay between early and late L1 acquisition is crucial in this process.

4. A scenario as just sketched does not directly shed light on the question what features may have triggered learners to form gender classes. It is probably impossible to reconstruct the relevant stages, but recent work on acquisition and variation in modern Dutch grammatical gender shows how semantic notions plays a role in modern varieties (cf. Audring 2009, Hulk & Cornips 2010). More precisely, Audring, for instance, argues that masculine and common gender pronouns in modern Dutch are used for referents with a high degree of individuation and neuter gender pronouns for referents with a low degree of individuation. Following Kraaikamp (2012), I will argue that the agreement pattern observed in Dutch pronouns today does not reflect a new interpretation of the pronouns, but rather a surfacing of a semantic interpretation of the genders that has always existed. In other words, these same semantic notions may have triggered the original gender categorization. Once gender is stored for a noun separately and formally expressed on agreeing elements via complex spell-out rules, the originally semantics-based agreement may be overruled by this new system.

5. In other words, I claim that the original semantic basis of the gender system is literally grammaticalised by successive generations of L1 learners, who, as "inflection machines", introduce generalizations based on abstract, purely formal features. Such a system may be (extremely) difficult to acquire by adults, leading to a dismantling of this system in situations of language contact. As a side effect, however, the semantic generalizations may surface more clearly again.

References
Audring, Jenny (2009). Reinventing pronoun gender. Utrecht: LOT dissertation.

Blom, Elma, Daniela Polišenská & Fred. Weerman (2008). Articles, adjectives and age of onset: the acquisition of Dutch grammatical gender. Second Language Research, 24(3), 297-331.

Hulk, Aafke & Leonie Cornips (2010). The role of gender and count features in the acquisition of het as a pronoun. In J. Costa, A. Castro, M. Lobo & F. Pratas
(Eds.), Language Acquisition and Development, proceedings of GALA 2009, Newcastle: Cambridge Scholars Publishing, 229-240.

Kraaikamp, Margot (2012). The Semantics of the Dutch Gender System. In Journal of Germanic Linguistics 24(3), 193-232.

Lupyan Gary and Rick Dale (2010) Language Structure Is Partly Determined by Social Structure. PLoS ONE 5(1): e8559. doi:10.1371/journal.pone.0008559.

Wexler, Kenneth (1998). Very early parameter setting and the unique checking constraint: A new explanation of the optional infinitive stage. Lingua 106: 23-79