Arguing from an ‚online‘ perspective on spoken syntax, I want to show in this presentation how lexical and grammatical categories (and functions) can be dependent on positioning within the turn-at-talk. As a consequence, so I will argue, grammar needs to be position-sensitive. I will discuss several examples which are difficult to handle within a traditional, position-insensitive approach since the lexical or grammatical constituents in question can not be categorized in an unambiguous and empirically justified way. In all cases selected for this talk, positioning before or after sentence, or positioning in the periphery or centre of the sentence in conversational German are at issue. The example for positionally sensitive lexical class membership will be the particle/adverb/discourse marker „nu(n)“, the examples for positionally sensitive grammatical constituents/functions will be ‚matrix clause‘-constructions with verba sentiendi, and ‚appositions‘.
Concentrating on English, German, and Hungarian, I examine instances of grammaticalisation from operators related to clause-typing (hence: C-operators; e.g. relative operators) into complementisers. C-operators may fall into a wide range of syntactic categories (e.g. adverbs, DPs): an essential criterion for their reanalysis into complementisers is the loss of certain category-specific markers incompatible with the properties of complementisers; the process is characterised by gradience (see Traugott and Trousdale 2010). I propose three criteria that can be used for modelling diachronic changes from C-operators into complementisers.
Complementisers define the type of the clause, whereas operators are contingent upon these (overt or covert) C heads. However, operators may be unambiguously associated with a certain clause-type and hence be responsible for the overt marking thereof, leaving the C head zero. Furthermore, in certain clause types operators are obligatory, thus operator movement can be detected even if the operator is zero (e.g. island effects in comparatives, Kennedy 2002). The distinction is especially elusive diachronically since it is precisely the ambiguous status of the element that allows for its reanalysis.
Criterion 1 (syntactic): C-operators may take lexical XPs along, complementisers must not. Hence the co-presence of a lexical XP (e.g. strings like how long in comparative subclauses cross-linguistically) indicates that the element in question cannot be a complementiser. This is not true vice versa: several C-operators do not take lexical XPs for independent reasons, e.g. VP-adverbs, which can hence be reanalysed as complementisers (e.g. hogy ‘how’ and mint ‘how’ in Old Hungarian, see Bacskai-Atkari 2014), while reanalysis is universally blocked for operator + lexical XP strings.
Criterion 2 (morpho-phonological): C-operators have to lose certain features that are incompatible with complementisers in order to grammaticalise. This criterion shows cross-linguistic variation: e.g. person and number features are not permitted on complementisers in many languages, but they are in others (e.g. Bavarian, Fuß 2004). Furthermore, the presence of such features may not be traceable overtly, either because they are irrelevant for certain operators (e.g. adverbials) or because they are not marked overtly in the given language (e.g. the grammaticalisation of that in English). The lack of overt marking of complementiser-incompatible features may facilitate the loss of these features, and hence the grammaticalisation, as will be shown to have been the case for Hungarian complementisers and German comparative complementisers.
Criterion 3 (morpho-phonological): the lack of certain morpho-phonological changes affecting a certain subclass of C-operators indicates that the given element is no longer the member of that subclass, and has been reanalysed as a complementiser. This can be extensively detected in late Old Hungarian: ordinary relative operators started to develop new forms via the cliticisation of a matrix pronominal element (e.g. milyen > a-milyen ‘how’), while this did not happen to elements that were already complementisers (e.g. mint ‘than’ in comparatives). Criterion 3 is useful in diachronic studies especially if Criteria 1 & 2 cannot decide on the exact position in particular examples: Criterion 3 is able to trace the reanalysis process based on longitudinal studies.
Nominal compounding in Germanic raises intriguing issues of category status and change. Intuitively, the demarcation between compounding and derivation seems quite obvious, but this boundary is blurred when taking into account bound morphemes with a more lexical meaning, like neo-classical combining forms (e.g. cyber-, neo-, -logy), or so-called ‘affixoids’. The latter can be defined as compound constituents systematically deviating in meaning from the corresponding free lexeme, being affix-like in that their new, more abstract meaning is bound: the Dutch prefixoid hoofd- signifies ‘main’, but only when part of a compound (as a free lexeme or in ordinary compounds hoofd denotes ‘head’). Affixoids tend to be highly productive and therefore form series, another property reminding of affixes. In addition, some affixes have their origin in compound members that went through an ‘affixoid phase’ (e.g. G. -heit < OHG. heit ‘position, character’).
The notion of ‘affixoid’ used to be heavily disputed (cf. Schmidt 1987), but it seems to have been more or less resurrected in recent years (see, among others, Stevens 2005, Booij 2009, Elsen 2009, Leuschner 2010, Norde & Van Goethem 2014); not as morpheme category in its own right, but as a handy descriptive term for certain lexical items in between compounding and derivation. Under the framework of Construction Morphology (Booij 2010), affixoids can be aptly conceptualised as ‘constructional idioms’ at the word level, i.e. word formation schemata with a lexically filled slot (see also Booij & Hüning 2014). In this way, the resemblance of affixoids to both compound members and affixes is accounted for.
How the notion of ‘affixoid’ is precisely to be defined, however, is another matter. Adopting it as a descriptive means of classification, the first part of my paper aims at discussing the issue of its delimitations. Are the German suffixoids –papst, lit. ‘pope’ (e.g. Wander-papst ‘expert in the field of hiking’), or -sau, lit. ‘pig’ (e.g. Politiker-sau ‘damned politician’), for instance, simply to be seen as lexemes that have undergone metaphorical extension, as suggested by Meibauer (2013)? And does it make sense to examine both prefixoids and suffixoids at the same time?
The second part of this paper will be devoted to analyses of corpus data from Dutch, German and Swedish, including state-of-the-art productivity measures (cf. Baayen 2009). It will be shown that affixoids not only pose a challenge to the demarcation between compounding and derivation, but also to the distinction of compounds from phrases. Heads as well as non-heads of complex words are a constant locus language change, and for that reason clear-cut categories often do not seem appropriate in the highly dynamic domain of word formation. It is argued that the concepts of ‘morphological transcendence’ (Libben 2010) and schematicity’ in a construction-morphological sense and can adequately account for these dynamics.
We intend to present the results of a research project investigating the integration of nominal and verbal properties in nominalizations. We will present evidence from two studies: an ERP study and an acceptability rating study. Like nouns, Polish nominalizations have (some) nominal inflection, occupy“nominal slots”in a sentence, can be modified by adjectives. Like verbs, they might have eventive semantics and an argument structure, be marked for aspect, be negated or modified by adverbs. Psycholinguistic research can provide new evidence regarding the categorical status and processing of nominalizations. As hybrid elements existing on the border of two grammatical categories, nominalizations can be expected to constitute a specific challenge for language users who have to embed some verbal structure in a nominal frame. This means not only a language system adjusting to enable this overlapping of categories (e.g., excluding/including verbal projections in nominalization phrases) but also the parser developing specific processing strategies (e.g., processing nominalizations primarily in neural areas dedicated to verbs, using verbal aspectual distinctions to guide nominal countability).
Our first experiment, an ERP study on finite verbs, converbs and nominalizations, derived from the same action verbal stems, was based on Kellenbach et al.’s (2002) study. They found that nouns and verbs elicited different ERPs: verbs relative to nouns elicited a posteriorly maximal enhancement of the P2 component between 250 and 300 ms and nouns elicited greater negativity than verbs between 350 and 450 ms. Our research question was whether the two “in-between” categories, i.e., converbs and nominalizations, would be processed distinctly as nouns or verbs (producing clear noun-like or verb-like ERP-signatures), or whether the findings would be more scalar (the elicited ERP-signatures would be partially verbal or nominal). This could be manifested in the presence of both signatures elicited by one stimulus or in the similarities in the topography and amplitude of the elicited ERP-components. The preliminary results of the experiment point to the conclusion that the ERP-signatures elicited for finite verbs, converbs and nominalizations are equivalent, which suggests that, from the processing perspective, the latter two are verb-like.
The aim of the second study was to determine whether verbal aspect affects the countability of Polish eventive argument-supporting nominalizations of the -nie/-cie type. The results of an acceptability judgment questionnaire indicate that verbal aspect can modulate the acceptability of a derived nominalization in typical mass contexts: nominalizations derived from atelic/imperfective verbs are more mass-like than those derived from telic/perfective verbs. This suggests that verbal features like aspect are relevant for interpreting nominalizations and that they can translate to the nominal mass/count properties. We also found that generally all types of nominalization were judged very low in count contexts (plural number, numerals) with only a marginal effect of aspect. This suggests a conflict between some verbal features (particularly perfective aspect) and plural morphology. To the extent that this conflict is not motivated semantically it points to the limits of the operations turning verbs into nouns.
Fillmore’s (1968) list of deep cases (a.k.a. semantic roles) provided generalizations about syntactic realizations of a verb’s arguments, but it was problematic: (1) the ranking of roles does not always yield the same argument realization patterns; (2) roles are not always defined consistently (esp. “theme” and “patient”); (3) semantic generalizations based on role information is problematic. Subsequent approaches to semantic roles such as Dowty (1991) and Van Valin and LaPolla (1997) offer different accounts of semantic roles, but they, too, have problems with delimiting the range of semantic roles and defining their boundaries clearly (see Levin and Rappaport Hovav 2005).
This paper offers solutions to these issues by proposing a more fine-grained list of semantic criteria that help to distinguish semantic roles from each other more straightforwardly. Building on data from FrameNet (http://framenet.icsi.berkeley.edu) and the methodology of frame-syntactic classification developed by Boas (2008), who proposes to introduce semantic features into frame-semantic descriptions, I propose more precise lexical entries that clearly demarcate the differences between target verbs evoking the same semantic frame. This approach is also helpful for determining what types of frame elements (FEs; finer-grained semantic roles, see Fillmore and Baker 2010) can be phrase-internal with respect to other FEs: only FEs exhibiting systematic metonymic and/or metaphorical relationships can be incorporated into FEs of the same frame. Based on these observations, I finally argue that frame-semantic descriptions need to include more fine-grained information that will allow us to clearly distinguish not only the different target words from each other, but also the different FEs from each other.
Starting point of the talk is the question: What is presupposed in analyzing and categorizing linguistic data? As part of the answer I will characterize a network of linguistic theories that are intended to be axiomatic theories at the end. It will be argued for at least three kinds of theories: (1) General linguistic theories that provide notions (terms) for analyzing arbitrary languages; (2) theories of one or more individual languages that describe and explain their specific properties in terms of a general linguistic theory; and (3) methodologies that guide the development of theories of the second kind. Usually, theories of the second kind are called “grammars” and contain identificational statements on categories like word classes. From a realistic point of view (cf. Neef 2014, Lieb 1983) categories can be reconstructed as set theoretical entities that are endpoints of systems of classifications (cf. Lieb 1993, 2005). As a rule, grammatical categories of a language S can be identified by statements like
(1) Adjective-of-S = the set of words of S that …
where “…” is a language specific characterization of formal properties of adjectives in this language. Identificational statements of this kind are the core of a language’s grammar connecting it with the presupposed general linguistic theory: In (1), the term “adjective” is used presupposing its introduction whether as an axiomatic primitive or by definition. For well known formal reasons (cf. Suppes 1957), (1) is excluded as a definition of “adjective”, in fact, as a definition in the logical sense of “definition” at all. On the other hand, notions like “adjective” can be defined in a general linguistic theory relying on functional notions. As a consequence, language specific categories (if identified properly) are comparable in principle. It will be argued that identificational statements on categories in a language’s grammar and definitions of underlying notions in the presupposed general linguistic theory follows opposite directions:
----------------------------------------------------------------------> language specific identification of categories
Phonology Morphology Syntax Sentence semantics
<----------------------------------------------------------------------- general linguistic definition of terms
It will be argued that neither overlapping of categories nor a prototype structure need to be ruled out. Hence, well known facts of, for example, language change can be accounted for in a natural way. – The main field of examples will be the parts-of-speech system, but other grammatical categories will be discussed, too.
Adverbials have been described by Hasselgård as “a rag-bag category in the linguistic system” (2010: 3) which is mostly defined in terms of not being a verb or not having a participant function in the clause. Although a commonly agreed-upon standard classification is difficult to come by, most linguists would probably agree that the elements grouped together under that name are far from uniform regarding their form and syntactic behavior in English. We find a diversity of realization forms (adverb phrases, prepositional phrases, noun phrases, clauses) and adverbials that may be obligatory, near-obligatory or optional elements. What holds this group together is a semantic approach. This renders different categories of adverbials, for instance adverbials of time which locate situations in time, or give information on the frequency or duration of events. I would like to explore the appropriateness of such a semantic classification of adverbials to describe an English word order pattern.
In English, adverbials display a high degree of flexibility regarding their placement. Thus, adverbials may appear in sequences immediately adjacent to one another within one clause, which is referred to as clusters. Their order within such clusters is often described in terms of their semantic classification. For instance, a general ordering preference for Present Day English has been described by Quirk et al. (1985: 565) for the end position as respect – process – space – time – contingency.
Hasselgård comes to the conclusion that what we see here is an interaction of various word order principles at work. Some of the factors discussed by her are obligatoriness, scope, lexical proximity to the verb, or weight (2010: 143ff.). Conversely, Hawkins (2000) questions accounts that are based on a semantic classification of adverbials, suggesting that this order is motivated by a more general principle of processing efficiency instead.
In a diachronic corpus-based study of clusters of adverbials of time and place in English, I have analyzed whether there is a preferred order of these adverbials in Old English, Middle English and Early Modern English, and which factors might motivate the respective ordering preferences. Against the backdrop of this study, I would like to discuss whether the ordering of the elements in these clusters is actually best described using a semantic classification such as place-before-time. What does the semantic classification of the adverbials contribute to the analysis overall? Is it appropriate to assume one broad category of ‘adverbials’ given the inherent variability of elements within that category?
To explore this point, I suggest we take a closer look at the data, in particular at a subset of nonobligatory adverbials only. These clusters make up about a third of the data and the multifactorial analyses of these data are rather inconclusive. However, their order follows a semantic pattern established by the majority of the data. Based on this observation, I will argue for a semantic classification of these elements within one overarching category of adverbials.
Since Aristotle, researchers have debated the relation between metaphor and simile. In figurative English copular constructions, (1), either the metaphor or the simile form may be used; when the figuratively interpreted noun modifies a verb (2) or an adjective (3), only the simile form is allowed.
(1) John is (like) a snake.
(2) Mary behaves *(like) a baby.
(3) Kim is as strong as an ox/*strong an ox.
A very different pattern is observed in sign languages. When (1-3) are translated into, e.g., Israeli Sign Language (ISL), the simile form is not possible in (1’), and is dispreferred in (2’-3’), where the metaphor form is possible, in fact preferred.
(1’) JOHN (*LIKE) PIG
(2’) MARY BEHAVES (?LIKE) BABY
(3’) KIM STRONG (?LIKE) OX
Why these differences?
Theoretical considerations (Croft 2003), experiments (Gentner & France 1988) and corpus studies (Huang 1994, Sullivan 2009), all indicate that, when given a choice, verbs, adjectives, or adverbs, rather than nouns, tend to be interpreted figuratively.
But what if the intended meaning requires nouns to be interpreted figuratively, as in (1-3)?
In English, the noun is simply interpreted figuratively after all. But sign languages have another option: reinterpret the noun as a different category. Words in many sign languages are often multicategorial and can be interpreted as nouns, verbs, adjectives, or adverbs (Meir 2012).
The use of a figuratively interpreted noun is extended to a different category. Specifically, the translation of (1) is (1’), where PIG is interpreted as an adjective, which readily receives a figurative interpretation. The simile form is not allowed since, syntactically, LIKE cannot precede an adjective,
Evidence that the figuratively interpreted words really are adjectives comes from the fact that they can occur with degree adverbs meaning 'very' (2a) like other adjectives (2b), while the comparable nouns cannot (3):
(2) a. HE CATFigurative 'VERY'. 'He is very catty/sly.'
b. HE STRONG 'VERY'. 'He is very strong.'
(3) *HE CATLiteral 'VERY'. 'He is very cat.'
However, some constructions require, in addition to categorical reinterpretation, logical type-shifting, which is costly and only applies as a last resort (Rooth and Partee 1983, Chierchia 1998, Krifka 2003).
Specifically, when the figuratively interpreted noun modifies a verb (2’) or an adjective (3’), its function must, syntactically, be extended to an adverb. But this involves shifting its type, from a first order property (noun) to a property modifier (adverb). Usually, the preference to avoid type-shifting is weaker than the tendency to avoid figurative interpretations of nouns, which is why the metaphorical form is preferred. But sometimes the ISL user may prefer to avoid type-shifting by interpreting the noun figuratively. The syntactically required adverbial is produced by preceding the noun with the preposition LIKE, leading to the simile form.
Thus, the puzzling distribution of metaphors and similes in spoken and sign languages is explained by the greater categorical flexibility of sign languages, and the interaction of two constraints: (1) avoid figurative interpretations of nouns (2) avoid type-shifting.
The intensive research on word classes (WoC henceforth) carried out over the last decades has shown that several WoC can be split into subclasses according to specifiable criteria. One of the finest categories that can be isolated on the Verb-Noun continuum (Ross 1972; Sasse 2001; Simone 2006) is that of the Nouns of Once: a kind of Process Noun characterized by aspectual features [- processual] [+ telic], that encodes a very brief, dot-like event (i.e. Ar. farabat ‘hit’).
Described in the traditional Arabic grammar as ‘ismu al marrat, Nouns of Once are recognizable in Romance languages, where they can be realized morphologically, syntactically or lexically (cf. for example Sp. fren-azo ‘(individual act of) braking’). One puzzling problem is that of determining whether this class may be supposed to exist in any language, in particular in those where the distinction between WoC (including the major ones) is blurred. In this paper we focus on Modern Standard Chinese, a language that lacks a general distinction between Process Nouns and the corresponding verbs: i.e. bàofā refers to both the verb ‘burst’ and the Process Noun ‘explosion’.
We suggest that Modern Standard Chinese has specific tools for generating Nouns of Once. In particular, four different syntactic or lexical strategies can be isolated, all involving a marked use of the classifiers. In general, as well known, nominal classifiers are a typologically widespread grammatical means for categorizing nouns (Aikhenvald 2003), broadly comparable to the system of gender (Corbett 1991) and measure words. However, when they appear in a specific pattern ([V + yī ‘one’ + NClass + N]), nominal classifiers can change their semantic format so as to refer to a dot-like performance of an action. For example, in (1), classifier kǒu, normally selected by nouns referring to people, gets the reading of the Noun of Once „sip‟:
(1) tā yǐn yī kǒu jiǔ
3SG drink one CL wine
‘He drinks a sip of wine.’
The same phenomenon can be observed for verbal classifiers, a class of words used to count the number of times an action occurs. If they usually follow the pattern [V + Num + VClass], verbal classifiers can appear in marked structure [yī ‘one’ + VClass + V], getting the meaning of a Noun of Once. For example, in (2), verbal classifier shēng, used for repetition of sounds, appears followed by verb qiāojī ‘beat’, getting the meaning of ‘a beat’:
(2) měi yī sheng qiāojī de jiànxì dōu shì yīmóyīyàng
each one CL beat DE interval all be exactly.alike
‘All the intervals of the beats are exactly alike.’
Finally, two lexical strategies will be discussed: the first involving the type of classifiers called Activity Measure Words (e.g. bù ‘step’) and the second involving lexical Nouns of Once (e.g. pēntì ‘sneeze’).
As a contribution to the general theory of language, the data presented in this paper corroborate the claim that Discourse Operations (as Nominalization, from which Process Nouns are derived) are a pervasive, possibly universal, phenomenon.
Num = numeral
NClass = nominal classifier
VClass = verbal classifier
In refining our understanding of language, we naturally search out areas where analysis is difficult, often as a result of indeterminate boundaries between phenomena. And progress may be hampered by problems of definition. One area where the situation has become serious is that of categorization systems (gender and classifiers). To stimulate progress here, I offer a canonical view of the phenomena being discussed, since ‘the canonical approach breaks down complex concepts in a way that clarifies where disagreements may lie between different linguists and theoretical frameworks.’ (Nikolaeva 2013: 100). This approach to typology involves analysing and defining phenomena which are subject to variability (across and within languages), extracting the various scales along which we characterize variability, and identifying the criteria to establish the logical end point of these scales. By integrating these scales, we construct theoretical spaces of possibilities, and only then do we investigate how this space is populated with real instances. To be a canonical instance of a phenomenon – that is, a clear and indisputable instance – means matching a full set of criteria. It then follows that such instances are likely to be infrequent or even non-existent (in accord with the ‘Anna Karenina Principle’, Diamond 1998: 157). This is therefore an axiomatic approach, which aims to ensure that we are aware of the full range of the phenomena we wish to account for, and that we have a metalanguage to describe them. It is justified entirely by utility and results.
I will report on work with Sebastian Fedden, in which we set up a canonical ideal for gender. We work through the consequences of the view that - in the canonical world - each noun has exactly one gender value. We show the types of phenomena which vary from this ideal: they do so to different extents, and so our canonical ideal serves as a yardstick against which our real examples can be calibrated. In contrast, we suggest that there is no coherent and useful view of a canonical classifier, since the various items that have been given this label do not share a common core. They are located in different areas of the theoretical space that we can establish, on the basis of the criteria for canonical gender.
To move forward here it proves helpful to look at languages which are claimed to have both a gender system and some form of classifiers. The existence of such double systems is usually stated rather than argued for; again we argue that a canonical approach is valuable here. I will show what canonical single systems and double systems look like, and how languages which are challenging, that is where the analysis of gender and/or classifiers has been difficult, can be fruitfully tackled given this typology of single/double systems.
This talk is concerned with the problem of categorizing the behaviour of words in inflection. Inflection classes are intra-morphological categories of disputed status regarding their psychological reality (cf. e.g. the past tense debate). Their delineation has been subject to decisions of grammarians and morphologists with different theoretical backgrounds and aims (scholarly/didactic etc.) since the first grammatical descriptions of German in the 16th ct. I take a brief diachronic look at classifications through time and then concentrate on comparing different present-day approaches towards delineating inflectional classes in different theoretical frameworks, asking:
- What counts (how much) as a class delineating criterion? What criteria are drawn on
- assuming a "main"/ a "minor" divide?
- How many and what kind of borderlines? Binary? Gradual? Categorical? Fuzzy?
- Is multiple classification of one item possible?
- May criteria be overridden? Default inheritance?
I show that some decisions in drawing borders within the verbal and nominal lexicon are quite stable (e.g. strong/weak, even if labelled differently), while other divides are highly variable and depend on the theoretical aims and claims of the person undertaking the classification. Moreover, the level of granularity varies to a high degree, i.e. the models cope differently with respect to subclasses. Last not least, I also ask about common principles and differences in classifying the inflectional behaviour of nouns versus verbs.
The outcome should be a typology of what linguists do in classifying inflectional behaviour, which should increase our sensitivity towards the explicit and implicit assumptions linguists make in categorizing.
In Luxembourgish, personal pronouns exist in two distinctive forms, i.e. du / de (pro.2ndpers.sg). The same holds for most determiners: een / en (det.indef.sg). There have been several attempts to label the different forms as ‘full’, ‘emphasized’, ‘reduced’, ‘weak’ or ‘clitic’ (cf. Bruch 1955, Krier 2002, Schanen & Zimmer 2012). Many authors make use of a rather inconsistent terminology leading to fuzzy boundaries of the functional concepts of these forms. The specific features of each pronoun are linked to different linguistic levels and some of the listed labels can be traced back to those respective levels: full vowel forms – forms containing a schwasound (phonetics), stressed – unstressed (phonology), strong – weak (morphology), full – clitic (syntax). Besides, there are strong semantic restrictions for anaphoric pronouns that have not been taken into consideration until now. For the Luxembourgish pronouns of the 3rd person singular, different semantic restrictions apply: the so-called ‘full’ form hien can only refer to animate or defined objects like persons, animals or the Nobel Prize, whereas the supposedly ‘reduced’ form en can be used for any object with masculine gender, although it is prosodically restricted (as it cannot be stressed in the sentence). In contexts in which the masculine pronoun needs to be stressed for prosodic reasons, but is at the same time semantically blocked (i.e. Feierdag ‘feast day’), an equivalent d-pronoun comes into play (i.e. deen ‘this’).
On the syntactic level, Weiß (in print) provides empirical evidence from German dialect data for a three-fold pronoun system (full – reduced – clitic) that applies for two morphological forms, creating a sort of mismatch. The Luxembourgish data will show if this tripartite distinction can be implemented for the word order of Luxembourgish pronouns and how the semantic restrictions operate in the sentence structure. The empirical work is based on a large corpus of oral and written Luxembourgish (> 62 mio. tokens).
This research on the particular features of Luxembourgish pronouns focuses on the syntactic and semantic properties and is to find answers to the following research questions:
- What is the ‘basic form’ or the ‘descriptive norm’ for pronouns?
- Which terms are compatible with all the specific features and how can the functional concepts be described adequately?
- Should a form be called ‘full’ if it has a limited range of anaphoric reference possibilities?
- What is the specific role of semantics in this context? How can the animacy hierarchy be implemented for the anaphoric reference of Luxembourgish pronouns?
The introduction of systematic acceptabiliy rating studies as a method of syntax research has not eliminated the problem of divergence in the assessment of the acceptability of certain constructions. The penalty for crossing a wh-subject in German sentences such as was hat wer gesagt (‘what has who said’) turned out to be large in some experiments, while it was non-existent in others. Likewise, the size of the crossing penalty differs substantially in experiments in English, Russian, and Czech. We will report a number of experimental studies that try to explain such effects in terms of differences in the modality of presentation and in the construction of the material
In my talk I will broadly consider what the role is of analogy in the way we learn language (which naturally includes grammatical categories), what its place is in the (biological) system of grammar that each individual adult language user develops in the course of his/her life ─ and how this, in turn, affects the theoretical model(s) of language that linguists develop ─ in order to answer the question whether and to what extent analogy helps to explain what happens in language use and language change (cf. e.g. Moravcsik 2011).
Foundational in my thinking about analogy are a number of assumptions that are generally acknowledged in the greater part of the linguistic community, such as (i) the idea that the principles and/or constraints that operate in language acquisition also operate in language change; (ii) that linguistics should be concerned with the production and comprehension of language use by speakers, and that, therefore, linguists should concentrate on language processing, and on language output (which of course first and foremost provides our data) as a product that is not separate from processing; (iii) that all human beings are born with a propensity for language, whatever that propensity entails.
With respect to (iii), since the idea of an innate biological grammar, as proposed by the generative school to solve the problem of the so-called ‘poverty of the stimulus’, is difficult to put to the proof at this stage of our (neuro-)linguistic knowledge, and since there are many arguments that have been put forward against this idea (cf. e.g. Pullum and Scholz 2002, and other articles in The Linguistic Review 19, i-ii (2002) devoted to this topic, and my discussion in Studies in Language, 2003), I will concentrate on analogy because it is a domain-independent (cf. e.g. Faarlund 1990, Bybee 1988, 2010) general, cognitive principle (cf. e.g. Hofstadter 1995, 2001) that may provide more in terms of explanation as to how we learn, process and change language.
As a historical linguist and syntactician, I will concentrate on how analogy may help explain how categories emerge and change, noting advantages as well as disadvantages for this approach on the way. Unlike Bybee and Becker (2014: 506), who write that “all [changes] fall under the umbrella of changes in categorization” and then add that “a separate mechanism of change dubbed ‘analogical’ is not necessary”, I believe that analogy is the very basis on which categorization rests. I conceive categorization broadly, including not only the traditional parts of speech but also higher, more abstract structures, with interaction possible on different linguistic levels (including discourse), as is done for instance in construction grammar. I will illustrate the possible usefulness and uselessness of thinking in analogical terms by discussing some previous case studies that came to the conclusion that the ‘analogical turn’ did not help in providing an explanation, and other case studies showing the opposite.
The so-called second cognitive revolution has substantially changed the way we think about categories and categorization in all cognitive domains, including language. While research paradigms such as cognitive and functional linguistics have provided ways of theoretically modeling linguistic categorization (Taylor 2003), there has now been a focus shift to the question how categories and processes of categorization can be identified in an empirically driven, bottom-up way (e.g. O’Keeffe 2004; Hilpert 2013).
In this paper, we discuss two morphological constructions which have been treated as “interface phenomena” in traditional, modular approaches to grammar, namely the English gerund in -ing and German nominalization with the suffix -ung. While going back to the same root, their diachronic development shows remarkable similarities as well as revealing differences. The English verbal gerund developed out of determinerless uses of the nominal gerund (such as in reading of your letter, cf. Fonteyn et al. 2015). A similar construction allowed German ung-nominalization to be used in a more verb-like fashion in Early New High German (e.g. in Lesung des Briefes ‘in reading the letter’). However, while the English gerundive construction served as basis for the development of a decidedly verbal construction, i.e. the verbal gerund (Fanego 2004), the ‘verby’ variants of its German counterpart fell out of use (Hartmann 2014). Thus, the evolution of German ung-derivation can be described as a process of “nominalization with ‘nominalization’ taken literally” (Demske 2002).
However, the implications of our study regarding the diachronic investigation of categories in grammar reach far beyond the notions of ‘nouniness’ and ‘verbiness’. Our study touches upon the much-discussed issue of the relation between lexicon and grammar, which are conceptualized as poles on a continuum in usage-based approaches (e.g. Boas 2010; Broccias 2012). While the morphological constructions themselves have very abstract meaning (and can therefore be seen as ‘grammatical-procedural’ constructions in the terminology of Traugott & Trousdale 2013), the resultant word-formation products are independent lexemes which can assume idiosyncratic semantic content through lexicalization. In this sense, the constructions under discussion are in-between “lexicon” and “grammar”. Importantly, however, abstract (‘grammatical-procedural’) constructions arise through abstractions and generalizations over their concrete instantiations (Goldberg 2006). In other words, they arise through processes of categorization. Drawing on corpus analyses based on the Penn-Helsinki corpora for English and the GerManC corpus for German, we will argue that investigating the changing usage patterns of -ing- and -ung-forms allows for a bottom-up description of diachronic category shifts.
In line with usage-based network models of language (Diessel forthc.), we argue that part of what makes linguistic categories so elusive is the fact that they are constantly in flux: paraphrasing Hopper’s (1987) much-cited slogan, our data imply that in language, ‘there are no categories, there’s just categorization.’ However, corpus-linguistic methods can help identify meaningful patterns of abstraction and generalization that language users make over instances of actual language use – in other words, they can help identify constructions, which themselves represent categories (Croft 2001: 27), and give clues to the functional motivations for their diachronic emergence, change, and loss.
In my talk I will deal with a problematic case of verbal morphosyntax in German. Specifically, it concerns the peculiar phenomenon that certain verbs do not adhere to the requirement that their finite forms occupy the second structural position (V2) in main clauses. Sentences like (1a) or (1b) are generally dismissed as not grammatically acceptable. However, in verb-final position (VL), i.e. in subordinate clauses (1c), the occurrence of these verbs is unproblematic.
(1) a. * Der Regisseur uraufführt das Stück.
the director original.on.put.3SG.PRESV2 the piece
b. * Der Regisseur führt das Stück urauf.
the director put.3SG.PRESV2 the piece original.on.
‘The director premières the show.ʼ
c. ... ob der Regisseur das Stück uraufführt.
whether the director the piece original.on.put.3SG.PRESVL
‘… whether the director premières the show.ʼ
Looking at a number of phonological, morphological, syntactic, and information-structural properties of these verbs I will discuss various factors that might account for their idiosyncratic behavior, thereby raising the question whether any criteria can be established at all that warrant the existence of a special “NonV2” verb class or, if so, the inclusion of a particular verb in that class.
There is an obvious difference between describing and/or analyzing an individual language and comparing languages thereby searching for cross-linguistic similarities, and this difference has important repercussions for linguistic terminology and categorization. Haspelmath (2010), for instance, argues that we must strictly differentiate between language-specific categories (descriptive categories) and terms used for cross-linguistic comparison (comparative concepts) and that there is no taxonomic relationship between these two levels. However, such a strict separation seems to be extreme in the light of the fact that many typologists are also experts of individual languages and that typological findings have a great impact on fieldwork and vice versa.
In this talk, I will analyze how the distinction is dealt with in current approaches to linguistic typology. How do language-specific categories link with cross-linguistic categories and how well do our comparative concepts fit the data from specific languages? In recent years we have seen at least two opposite approaches:
1. bottom-up: from data/languages/structures to comparative concepts/cross-linguistic categories
2. top-down: from comparative concepts/ cross-linguistic categories to data/ languages/ structures
The first approach is explicitly advocated by Bickel & Nichols (2002), Bickel (2010), Bickel et al. (2014) under the name of Autotypologizing or multivariate analysis, etc. The second is promoted by Canonical Typology (e.g. Corbett 2005, Brown, Chumakina & Corbett 2013). Both approaches rely on the assumption that categories are complex items that can be dissected into smaller features (variables) and possible values that the features can have.
The aim of the talk is to discuss the two approaches with respect to their very different premises and methodologies (e.g. for Bickel frequency and statistical analysis are inherent parts of the typological enterprise while Canonical Typologies relies on our ability to a priori identify canonical instances even in cases where these are rare and could, at least in theory, even be non-existent) and with respect of how exactly language-specific and cross-linguistic terms/categories/concepts/data are linked. In a second step, I will discuss what these approaches tell us about the nature of categories in linguistics, how we arrive at them and what problems and questions recur over and over again.
While research in linguistic categories often highlights the need for abstract grammatical rules that structure and organize these linguistic categories, the present work follows a different line: disambiguate the perceptual fringes of incoming input that mark the limits of what can be categorized (an acoustically shaped boundary of a category will block features outside its range, and all such boundaries are likely not to incorporate features outside the human hearing range, crudely speaking), and locate physiologically grounded maxima for acoustic interaction to occur. While the hearing range of humans, the speed of sound and the speed of auditory perception are trivially included in the canon of every phonetics class, little is known about the interplay between physiological prerequisites and grammatic constructs. In this fashion, not a single discriminant feature, but rather a multitude of different acoustic parameters have a significant influence on all levels of linguistics.
Single tones and arrangements of tones at the same time are a convenient stimulus candidate to test the specific settings involved in auditory decoding, since they are void of many levels of meaning and social information that speech invariably transfers. This domain of research facilitates to uncover the upper and lower boundaries of acoustic input in its relation to language, as well as possible perceptual bottlenecks (e.g. the detection of sound in noise) and early category formation. A question that has been notoriously ignored in the early days of linguistic research is thus getting more deserved attention: Is categorical and conceptual knowledge governed by grammatic categories or does our body have a word in this debate, too?
In spoken German, a limited set of elements is used exclusively in the end position of an utterance where they signal its potential completion. These elements, such as final particles (e.g. dann, sogar, ja, jetzt, Haselow 2015), response elicitors (e.g. oder?), or general extenders (e.g. und so), are not integrated into the morphosyntactic structure of a clause and thus “outside” the clause in traditional syntactic analysis, or “extra-clausal constituents” in Dik’s theory of Functional Grammar (Dik 1997: 380). Yet, they are linked to a syntactic unit (e.g. a phrase or a clause) in a variety of other ways, e.g. in terms of prosody, scope and discourse function. These kinds of final elements provide methodological problems for grammatical description as they raise the question of how to deal with units that have no constituent status (they are in no dependency relation with any constituent of the host unit) and whose use is largely restricted to interactive speech production. They also challenge traditional ways of categorization, which are based on constituency in “well-formed” syntactic units and betray a strong written bias. Final particles, for instance, neither fulfil the morphosyntactic criteria that define adverbs, nor are they analyzable as post-posed modal particles. Similarly, a response elicitor like oder? is not analyzable as a conjunction. The categorization of such elements is complicated by the fact that they have heterosemes in other (non-final) positions, where they are used with a different meaning. In short, they represent an ideal testing ground for alternative categorization models, based on the integration of the syntax of spoken language produced “online” into our modeling of syntactic categories.
In this talk I will present an approach to grammatical categorization that distinguishes two cognitive serialization principles according to which speakers build up a unit of talk in real-time speech production, microgrammar and macrogrammar, and thus two different “category pools” from which speakers choose constituents. These two concepts have been inspired by the notions of microsyntax and macrosyntax introduced in the works of Alain Berrendonner, Claire Blanche-Benveniste, and their colleagues. However, the model presented here encompasses more than syntax, namely the establishment of structural relations beyond a single utterance, and it is based on the cognitive principles of on-line speech production and processing (Auer 2009). Microgrammar is a serialization principle that refers to the formal means employed by speakers to structure a unit of talk based on internal hierarchization, embedding, constituency, and dependency relations. Macrogrammar refers to relational functions outside phrase-, clause- and sentence-internal dependency relations, and is based on serialization principles that rest upon aspects such as information structure, textual coherence and processibility.
Based on an empirical study of their contexts of use and their function in the German Datenbank für Gesprochenes Deutsch, it will be shown that final elements as one particular manifestation of macrogrammar form sets of grammatical categories serving the structuring of speech produced on-line and should therefore not be thrown into a wastpaper bin category such as “pragmatic markers”. Moreover, I will argue that elements in the final field are “external” to a structural unit from a “microgrammatical” perspective, but an integral part of the structural unit they accompany from a “macrogrammatical” perspective.
If linguistic competence involves implicit knowledge of grammatical categories, this knowledge must be acquired during the early years of language development. There are essentially two views regarding the question of how it is acquired. In generative approaches, children have innate knowledge of grammatical categories, and the process of language acquisition involves mapping the language children hear onto the categories of a predetermined universal set. In constructivist approaches, grammatical categories are actively constructed from the input data. Common to both generative and constructivist theories of language acquisition is the assumption that category identification/formation relies on analysis of co-occurrence patterns; that is, children acquire grammatical categories on distributional grounds.
Generative and constructivist models of language acquisition further agree that children’s early linguistic input includes use of ‘formulaic routines’, defined as multi-word strings that are learnt and produced as unanalysed wholes (e.g. ‘What’s that?’). However, formulaic routines are accorded a very different status in the two frameworks. In constructivist accounts, the child progresses from formulaic routines that correspond to high frequency constructions in the input (e.g. ‘What’s that?’) to lexically-specific schemata (e.g. ‘What’s [X]?’) that become increasingly abstract (e.g. [WH-WORD] [AUX] [SUBJ] [VERB]). Constructivists thus postulate continuity between the use of formulaic routines and the production of novel syntactic structures. By contrast, in generative approaches, which lack the concept of schema, there is no gradual development from formulaic routines to productive syntax through schema abstraction. Early utterances that are stored and retrieved as idiomatic, holistic chunks are ‘pre-syntactic’; they form part of the child’s lexicon but not her grammar. Formulaic routines then essentially provide a dead-end for syntactic acquisition, to be abandoned as the child acquires the final-state grammar. This view of formulaic routines is unsatisfactory in that it does not ascribe any impact of formulaic routines on language acquisition, raising the question why children should identify and store linguistic expressions that do not straightforwardly map onto the categories of their pre-determined set.
This paper proposes a re-evaluation of the role of formulaic routines within a generative approach to language acquisition. More specifically, it considers the acquisition of wh-questions in English and German, and argues that formulaic routines provide the starting point for the acquisition of wh-categories by allowing children to identify the [+ wh]-feature in wh-routines through pattern analysis. The proposal is as follows: The earliest instances of (non-holophrastic) wh-questions appear to be ‘wh-identity questions’ of the form ‘What/Where is …?’, which, even if unanalysed by the child and hence to be considered as wh-routines, must be associated with a functionally coherent communicative intention (e.g. ‘Tell me the name/location of…’). Wh-routines emerge at a stage of language acquisition at which the child also uses non-wh-routines, such as ‘That’s a...’. Given the functional (and prosodic) contrast between the two types of entrenched expression, and children’s well-attested ability to recognise distributional patterns, the first step toward the identification of wh-categories may be the extraction of a wh-feature from the (otherwise unanalysed) wh-routine, cf. (1).
(1) a. [FP That’s a …]
b. [FP What’s a …]? < [wh [FP that’s a …]].
The paper provides empirical support for the early emergence of wh-identity questions and the co-occurrence of wh- and non-wh routines, and gives a novel, concrete account of how wh-category formation might proceed. Moreover, the suggestion that wh-routines have the rudimentary syntactic structure in (1b) offers an explanation of the absence of wh-displacement errors in early child language as an effect of ‘spell-out driven movement’.
I will focus on subcategories within the nominal and verbal domain: (1) the mass/count distinction for nouns, e.g. gold vs. ring, and (b) the activity/accomplishment/achievement distinction with verbs, e.g. run vs. recover vs. arrive. While these distinctions are fairly well established by syntactic and semantic tests, we find that particular lexical items can be astonishgly flexible — e.g., there are count noun uses of mass nouns, as in three waters, and mass noun uses of count nouns, as in there’s apple in the salad.
In this talk I will discuss the idea of coercion: systematic semantic changes that can adapt a lexeme to the needs that arise with a particular use. Assuming coercion lead to potential problems — e.g., why should we assume lexical count nouns at all, knowing that there are coercion operators that may change a mass noun meaning to a count noun meaning. In this talk, I will discuss the promises and the problems of the coercion approach to fuzzy category distinctions.
An intruiging topic with regard to category status and change is word order, particularly in Germanic where word order normally distinguishes between clause types such as polar interrogatives (verb first = V1), declaratives (verb-second = V2 or, in English SVO) and in part also subordinate clauses (V-late in German, Dutch etc.). Assuming that such distinctions are organized in a prototype-like fashion and therefore not necessarily watertight, I will focus on one particular case of overlap, viz. V1-conditionals:
(1) Scheitert der Euro, (dann) scheitert Europa. (Merkel)
fails the Euro then fails Europe
Since the V1-protasis in (1) is structurally identical to a polar interrogative, it is often assumed that such protases arose from polar interrogatives in a dialogue type of sequence (adapted from Jespersen 1940: 374):
(2) A: Will you come? (– B: Yes. –) A: Then we can start at once.
This apparently attractive account does not readily fit the facts of English, however, where V1-protases are introduced by either should, had or were and are much harder to associate directly with polar interrogatives than in German (Van den Nest 2010). Although ostensibly developed as a scenario for historical origins, (2) therefore raises intriguing synchronic issues concerning the category status of V1-conditionals as well. As I will argue in the first part of my paper, the most promising response is constructional, as suggested for V1-conditionals in English by Kim (2011, with references). From this perspective, V1-conditionals must be close sisters to V1-interrogatives in the constructional network of German, whereas the respective relationship in English is more like that of estranged cousins.
As to the diachronic perspective, one can of course claim that the account in (2) is not intended to refer to the present-day languages at all and suggest instead that V1-conditionals developed out of their assumed interrogative origins in an asynchronous fashion in different languages, emancipating themselves faster from interrogatives in English than in German (König 2012: 8f.). V1-conditionals thus constitute a classic case of "Sprachwandelvergleich" (Fleischer/Simon, eds., 2013), but as I will show using data from Van de Nest (2010), the asynchronicity hypothesis is only partially true. Despite some similarities, V1-conditionals in earlier English never resembled their German counterparts to the expected extent at any time in their history, nor did they show a particular affinity to polar interrogatives even in Old English. Part of the problem, I argue, is therefore the fundamental assumption implied by (2) that one V1-construction was necessarily the historical source of the other. In fact, V1 was not a marker of a clearly delimitated set of clause types in ancient Germanic; rather, it was associated with special, in part language-specific types of information structure (Petrova/Solf 2008) which apparently lent themselves readily to, inter alia, non-assertions. In taking this approach forward, I adopt a distinction by Traugott/Trousdale (2013) to argue that V1- constructions originated in slightly different processes of constructionalization in ancient Germanic and then followed divergent paths of constructional change depending on whether V1 remained in productive opposition to V2 (cf. German) or ended up in a weaker, niche-like position vis-à-vis SVO (cf. English).
Old English has inherited from Germanic a number of argument structure constructions that do not contain a traditional subject. Not only can the nominative argument generally be dropped when recoverable from the context; there are also a number of phrasal patterns that feature oblique experiencers or adjuncts in topical position and are perfectly grammatical without any nominatives. Old English syntax, in other words, allows subject-less argument structures – if we equate subject with nominative argument.
There have been a number of publications arguing for the existence of oblique subjects in the Germanic languages (for OE, see Allen 1995: 50). This obviously entails a definition of subjecthood based on syntactic rather than morphological criteria. (Barðdal mentions first position in declarative clauses, inverted position in questions, first position in subordinate clauses, subject-to-subject raising, subject-to-object raising, conjunction reduction, control infinitives, clause-bound reflexivization, and long distance reflexivization; cf. 2009: 123, FN). The problem is that, in the absence of an explicit discussion of what a subject is, the diagnostic test is tacitly assumed to define the category and its value is not assessed. Since the different criteria yield different results within languages as well as cross-linguistically (cf. Croft 2005: 45), Croft argues against syntactic categories except as elements of constructions. In his Radical Construction Grammar, syntactic categories are assumed to be language and construction specific and what universals there are are considered cognitive, not linguistic in nature. In other words, a subject in English does not have the same characteristics as a subject in German or Icelandic, and the subject of an intransitive construction is not the same as the subject of a transitive construction (Croft 2001: 48).
The question I would like to debate is under which conditions the notion of subject is then useful at all.
Most theories of case assume that the functions of case markings (nominative, accusative, etc.) are syntactic relations (subject, direct object, etc.), and that it is only via these syntactical relations that case morphology indicates semantic roles (agent, patient, etc.). For the description of the Old English case system, however, it seems that syntactic labels can be dispensed with. Does it not suffice to specify that construction C includes the elements AGRCASE1, AGRCASE2, AGRCASE3, and that these arguments encode PARTICIPANTROLE1, PARTICIPANTROLE2, and PARTICIPANTROLE3, respectively? If subject and object are (only) generalizations over elements of argument structure constructions, and if, at least in Old English, subject and object properties can coincide in elements, does the dichotomy make sense?
There is one other possible definition of subject besides the morphological and the syntactic, namely a discourse-functional one: the subject as the topic about which a predication is made. From such a perspective, it may well turn out that Old English does indeed have oblique subjects, and that the syntactic commonalities shared by nominative and oblique subjects are in fact derivative of their discourse-structural functions.
The proposed presentation is focused on research into reduplication as a word-formation process which was studied in a genetically and areally stratified sample of 106 languages. The research combines the approach of canonical typology with a large-scale typological research of reduplicative constructions across the sampled languages. The objective of the presentation is twofold.
(i) Reduplication has often been assessed as ‘a difficult phenomenon’ since it “is one of those phenomena which appear to be interwoven with so many related phenomena that they do not readily let themselves to a clear-cut definition” (Stolz 2012: 9). In order to circumvent the “Correspondence Problem” (Corbett 2012), the initial theoretically-oriented part of the presentation adopts the approach of canonical typology since it allows to identify the domain of research by means of “constructing a theoretical space of possibilities in which a particular phenomenon can be calibrated” (Bond 2012: 20). Through a set of converging criteria associated with the base (two sets of criteria were identified: formal and semantic), the clearest or canonical instances of reduplication were identified and other examples of reduplication were calibrated out from the canonical.
(ii) The second part of the presentation aims to provide preliminary results of a cross-linguistic analysis of reduplication from the perspective of the onomasiological theory of word-formation.
Adopting the onomasiological approach to word formation enables to find out how onomasiological (cognitively grounded) categories are linguistically represented through reduplication and its various formal subtypes (complete, partial preposing, partial postposing, partial infixing, echo reduplication, reduplication accompanied with stem modification, suffixation and compounding) in contrast to other selected morphologically marked word-formation processes. Following Štekauer et al (2012) the analysed semantic categories are divided into four basic groups: nominal (agent, patient, instrument and location), evaluative (augmentatives and diminutives), verbal (causative, intransitive and transitive, frequentative and intensive) and word-class changing. The presentation thus offers two ways of looking at the same morphological category, one of them dealing with the theoretical aspects of defining reduplication and the other one focusing on results of analysis of reduplication in terms of semantic categories that it conveys in comparison to other morphological processes selected as a statistical baseline for comparison.
This paper discusses the semantic categories of animate–inanimate and human–non-human. Grammatical descriptions rely on these distinctions when explaining the use of impersonal constructions (e. g. ISK 2004), personal pronouns (e. g. Blanche-Benveniste 1978, Mikkola & Laitinen 2013) and expressions of intentional action (e. g. Heine 1995) or empathy (Langacker 1991). The present study shows that the constructions considered as reserved to human reference in Finnish are used for animal reference as well, in a context where the speakers seek to gain understanding.
The data come from a radio program where listeners call in to ask a council of wildlife experts questions arising from their nature observations. The discussions typically consist of a caller’s narrative, reporting the human-wildlife encounter, and an exchange of questions and answers between the caller and the experts.
The paper focuses on the deontic and dynamic modal expressions, as well as three types of open reference constructions (the zero person, the generic 2nd person singular and the passive-impersonal). It addresses the question of animals as linguistic agents and persons, showing that the clear-cut distinction between the semantic categories of human and animal is contextually and culturally constructed.
From a human perspective, ‘animacy’ appears as a gradient category. Following this, it has been argued that certain animals, such as dogs and other pets, are “anthropomorphized” and thus considered by humans as more prototypically animate than wild animals, especially poikilotherms (e. g. fish, insect). The animals conceived as remote on the cognitive animacy scale could be brought closer to the most central member of the category (the human self) only in fictive or figurative language use (see the discussion in Yamamoto 1999).
The data analyzed in the present paper does not support the idea of anthropomorphizing animal participants on the basis of their perceived closeness to humans. This is illustrated by the example (1), containing passive-impersonal verb forms. The speaker explains the capacity of the fish to orient themselves under water.
(1) tällä tavalla kala pystyy aistimaan pimeässä tai syvällä [...] että
miten päin ollaan että miten päin orientoidutaan siinä ympäristössä
Q way be.PASS COMPL Q way orient.PASS DEM.INE environment.INE (13062012)
‘this way the fish can sense in the dark and deep […] in which position one is and how to orient oneself
in that environment’
The Finnish passive-impersonal implies an unspecified, plural agent, said to be human (Helasvuo 2006). Yet, the idea of a collective action allows, in this extract, the interlocutors to put themselves in the situation of the fish.
By using modal expressions and open reference constructions, the speakers show that they can identify with the animal experience, even when the situation described is not typical of humans. Consequently, the semantic category of human does not appear as a determining one when describing the use of these constructions, but rather the idea of a shared experience. Even when the animal participants’ behavior, environment or life span are incomprehensible to the speakers and motivate their questions, there are interests common to all animals, beyond this otherness: e. g. the desire to survive, to protect one’s progeny and to avoid physical pain.
The knowledge of grammatical categories is considered to be at the core of the ability to produce and comprehend language. Studies on the production of brain-damaged subjects give evidence that grammatical information is an important organizational principle of lexical knowledge in the brain (Hillis & Caramazza 1995). The question is whether grammatical categories should be seen as abstract features independent of semantics and phonology, or whether they emerge as a higher-order regularity from semantic and phonological properties of words (Mirković et al. 2008). This issue has been at the centre of a long-term debate about the way grammatical categories are distinguished from one another, and further relates to the question of how children deal with the grammatical categorization of language data.
In this talk I discuss types of cues that facilitate the identification of grammatical categories in language acquisition, both in naturalistic and in connectionist network contexts. Whereas there is general consensus on the role of distributional cues (frequent syntactic frames) and phonological cues (mainly stress pattern and number of word’s syllables) (Gómez 2002, Mintz 2002, Kelly 1992, Durieux & Gillis 1999, among others), the role of semantic cues is still less clear-cut.
In order to investigate the contribution of semantic cues to the acquisition of grammatical categories, I analyse the acquisition of grammatical gender as a concrete study case. First, I discuss data from Serbian in the context of connectionist network learning (cf. Mirković et al. 2008) as well as from artificial language learning in adults (cf. Mirković et al. 2011). Second, I compare these data to experimental studies of the acquisition of grammatical gender in Russian monolingual children and Russian bilingual children with English, German, Finnish, or Hebrew as L2 (cf. Schwartz et al. 2014).
The data analysis shows that distributional and phonological cues to the acquisition of grammatical categories have a primary role, and serve as a basis for early category assignment, whereas semantic cues are found to have a secondary role, and are not used across-the-board. This may be taken as an indication of the ranking of cues to grammatical categories learning with distributional and phonological cues being strong grammatical categories predictors and semantic cues being weaker ones, as these may but must not always play a role, depending of the specific category to be acquired. In general, this gives us a hint that semantic information is more specific and language dependent as predictor of grammatical categorization, whereas distributional and phonological information has more universal role for the identification of categories.
The German structure [werden + infinitive] presents an almost classical problem of categorization: it has been categorized as a tense (Klein 1994: 116f), as an aspect (Bartsch 1995: 159), as a mood (Vater 1975) or as some combination of these (Leiss 1992: 191ff). Indeed, semantically, a sentence like (1) seems to oscillate between all of these categories:
(1) constructed example
Er wird schlafen.
he FUTAUX? sleep
‘He will sleep (in the future)’
‘He is in the pre-state of sleeping’
‘He is likely to be sleeping’
Things get even more complicated when taking into consideration a non-finite variant of [werden + infinitive], which appears to be emerging currently (Reiner 2014). For this form possibly combines an aspectual and a relative temporal meaning in a very special way, cf. examples (2) and (3).
(2) Mannheimer Morgen, September 19th, 2008, p. 17
Erfreut darüber, nun künftig gerade am Wochenende mehr Zeit mit der Familie verbringen zu können, etwas enttäuscht darüber, künftig nicht mehr ganz so viel Kontakt zu den Menschen in Viernheims Vereinen und Organisationen
haben zu werden.
have.INF INFPART FUTAUX.INF?
‘Pleased that from now on she will be able to spend more time with her family, especially in the weekends, a bit disappointed that from now on she will be less in touch with the people in Viernheim’s clubs and organisations.’
(3) Neue Kronen-Zeitung, March 4th, 2000, p. 2
Dass ich in meinem Leben auch noch für den Opernball
sein werden müsse,
be.INF FUTAUX.INF? must.1SG.QUOT
hätte ich mir nicht gedacht.
‘I wouldn’t have thought that one day I would even have to be in favour of the Opernball.’
Here, I will argue, aspectual and relative temporal meaning are combined inasmuch as the infinitive of [werden + infinitive] seems to express the following relationship, stated in the terms of Klein 1994:
(4) TTsuperordinate verb < TSitinf.
However, while complicating things semantically, the existence of such a non-finite variant might help solve the issue of categorization morphologically. In this regard, I will present arguments based on Abraham 2001 and Rothstein (2013: 117), indicating that in light of examples like (2) and (3), [werden + infinitive] has to be classified as a tense eventually. So, finally, rather clear-cut morphological arguments will have to be weighed against inconclusive semantic criteria.
As a side note, I would also like to address another categorization issue related to [werden + infinitive]: may this structure, as a multi-word expression, be treated as part of morphology at all or does it rather belong to syntax? Here, a canonical definition of periphrasis proves helpful (Brown et al. 2012). Additionally, a canonical approach (Corbett 2012) has the advantage of explicitly distinguishing between categories (e.g., tense) and their respective subcategories (e.g., past, present, and future) in terms of features and their values. Concluding, I will discuss why it might still be problematic to capture the whole structure [werden + infinitive] by means of canonicity.
Examples are taken from the DeReKo, release 2014-II, searchable at https://cosmas2.ids-mannheim.de/cosmas2-web/faces/investigation/archive.xhtml.
 Hawkins (2000: 232)
 In line with the approach of Štekauer et al. (2012), the term word-formation is used as synonymous with morphological derivation or derivational morphology. Reduplication as a word-formation process can thus be characterized as “a process with semiotic and cognitive foundation which serves the purpose of formation of new linguistic signs” (Štekauer 2012:237).
 Onomasiological approach studies the ways how speakers/hearers construe and then represent cognitive conceptual domains in individual languages.
 Although there is a hint to a finer-grained correspondence between semantic categories and grammatical gender so that gender is not altogether semantically arbitrary (cf. the studies in Corbett 1991 and Zubin & Köpcke 1986).