Upcoming Events


Nov 09 - 10, 2018

Event Semantics Workshop 2018 (EvSem2018)




This year's Event Semantics Workshop will take place at the Mathematikon at Heidelberg University.



(preliminary version of 17/10/2018)


Thursday 08-11-2018


- pre-conference get together (location to be announced) -

Friday 09-11-2018


- welcome -


Christiane von Stutterheim: Event Unit Formation [invited] abstract


- coffee break -


Dennis Wegner, Marcel Schlechtweg & Holden Härtl: The (early) availability of future and perfect information in verbal clusters: Implications for the compositional semantics of composite tenses abstract


- coffee break -


Dorota Klimek-Jankowska and Joanna Błaszczak: The processing of grammatical aspect in Polish – new experimental evidence abstract


- lunch break -


Curt Anderson: Specification of methods and the semantics of method-oriented adverbs abstract


- coffee break -


Carla Umbach, German wie-complements: Manners, methods and events in progress abstract


- coffee break -


Willi Geuder & Katja Gabrovska : Adverbs of intentionality abstract


- end talks day 1 -


- conference dinner (location to be announced) -


Saturday 10-11-2018


Anette Frank: Resolving Abstract Anaphors in Discourse — Uphill Battles with Neural Networks and Automatic Data Generation [invited] abstract


Martin Schäfer: Unmögliche Interpolationen abstract


- coffee break -


Katja Laptieva & Sebastian Bücking: Towards a compositional analysis of the an-construction in German abstract


Sebastian Löbner: Act-cascades, TTs, and dot types abstract


- coffee break -


Johannes Gerwien: Predicting referents based on structural meaning – The case of the Mandarin Chinese bǎ-construction abstract


N.N. [invited] – confirmation pending


- end of day 2 -



Specification of methods and the semantics of method-oriented adverbs

Curt Anderson (Heinrich-Heine-Universität Düsseldorf)

Much work on the syntax and semantics of adverbs in English, German, and other languages has focused on understanding classes of adverbs such as evaluative adverbs, agent-oriented adverbs, and manner adverbs. One class that has remained understudied is method-oriented adverbs, as described by Schäfer (2013). Method-oriented adverbs are said to specify the method by which some event comes about (examples (1-3)).

(1) The aliens communicated with each other telepathically.

(2) The two nations solved their issues diplomatically.

(3) The semanticist evaluated the data linguistically.

These adverbs are derived from relational adjectives (telepathic, diplomatic, linguistic) and not property adjectives, unlike many manner adverbs. Additionally, they can be distinguished from manner adverbs via syntactic and semantic tests, suggesting that method-oriented adverbs must be given an analysis different from those of manner adverbs.

I clarify the syntactic position of method-oriented adverbs within the clause, and show in what sense these adverbs specify a method. My account is based on a lexical decomposition of the adjective that the adverb is derived from (cf. Anderson & Löbner 2018) and its interaction with the lexical semantics of the modified verb, and makes use of independently motivated primitives for instrumental meaning developed by Koenig et al. (2008).


Anderson, Curt & Sebastian Löbner. 2018. Roles and the compositional semantics of role-denoting relational adjectives. In Uli Sauerland & Stephanie Solt (eds.), Proceedings of Sinn und Bedeutung 22, vol. 1, 91–108.

Koenig, Jean-Pierre, Gail Mauner, Breton Bienvenue & Kathy Conklin. 2008. What with? the anatomy of a (proto)-role. Journal of Semantics 25(2). 175–220.

Schäfer, Martin. 2013. Positions and interpretations: German adverbial adjectives at the syntax-semantics interface. Berlin: De Gruyter Mouton.


Resolving Abstract Anaphors in Discourse — Uphill Battles with Neural Networks and Automatic Data Generation

Anette Frank (University of Heidelberg, Institute for Computional Linguistics)

Abstract anaphora resolution (AAR) is one of the famous Uphill Battle tasks that have been well studied in linguistics without having been solved in a compuational setting. AAR aims to resolve anaphoric expressions that refer to abstract objects such as propositions, events or properties in the (typically) preceding or following discourse. While AAR has been studied extensively in linguistics (Asher 1993), the task is only very recently starting to be addressed in computational linguistics. I will present our ongoing work that is the first to address this task using deep learning methods, drawing from linguistic insights.

In our work we characterize abstract anaphora as establishing a relation between the anaphor embedded in the anaphoric sentence and its (typically non-nominal) antecedent. We present the first neural network-based approach to abstract anaphora resolution that overcomes one of the major obstacles for addressing this task computationally, by automatically deriving training material from parsed corpora (Marasovic et al. 2017). The proposed model defines AAR as a mention ranking model, following the entity coreference model of Clark and Manning 2015. To capture the relation between the abstract anaphor and the antecendent in context, we propose an LSTM-Siamese Network architecture that we extend with an adversarial training method. I will describe how we harvested training data from a parsed corpus using a common syntactic pattern consisting of a verb and an embedded sentential constituent. I will present results obtained with the mention-ranking model trained for different types of abstract anaphors - nominal and pronominal - from different corpora, and in increasingly challenging evaluation setups, to raise this computational modeling task stepwise towards a realistic document-level resolution task.


Predicting referents based on structural meaning – The case of the Mandarin Chinese bǎ-construction

Johannes Gerwien (University of Heidelberg)

The characteristics of the Mandarin Chinese bǎ-construction allow to hypothesize an exceptional processing strategy in event comprehension: Besides other functions, the marker bǎ changes the canonical word order S-V-O to S-bǎ-O-V; it marks the noun to which it belongs as the direct object (cf. Yang & van Bergen 2007), and it signals that the corresponding referent must be interpreted as having changed from one state to another (cf. Li & Thompson 1981). The quality of the referent’s initial and resultant state, however, is specified by the sentence- final verb. Thus, the bǎ-construction offers a unique case to study the online comprehension of structural meaning (temporal and causal relations) independent of content meaning. Can referent states be activated before they are qualitatively specified? While listening people use the current linguistic input to predict upcoming discourse (cf. Altmann & Mirković 2009). We hypothesized that the function word bǎ triggers predictions about the referent following it as an affected object.

In a visual world paradigm, we measured saccade-onset times directed to target objects under 2 conditions: In the critical condition, an auditorily presented target noun followed bǎ, whereas in the control condition it followed de, a possessive marker in our context. This yielded sentence pairs such as tā bǎ xiǎoshuō sī huài le/tā de xiǎoshuō bèi sī huài le (‘He bǎ novel rip apart’ / ‘He de (=His) novel was ripped apart’). The visual stimuli always showed three objects, one of which was unambiguously depicted in a resultant state, e.g., a deformed plastic bottle (see Fig. 1). There were 12 bǎ/de stimulus pairs in total and 12 fillers. 26 Mandarin native speakers participated in the experiment. Two experimental lists ensured that every subject encountered only one pair-partner, 6 from each condition (but all fillers). The participant’s task was to click on the object mentioned in the sentence. Our analyses revealed a reliable effect: There were significantly more pre-noun target-saccades in the critical condition (bǎ) compared to the control condition (de). Furthermore, the cumulative proportions of first target-saccades increased significantly more rapidly.

Results of a second experiment confirm these findings and rule out alternative explanations. Only sentences like those in the critical condition in experiment 1 were used (He bǎ noun ...). The nouns matched a visual target that this time was not depicted in a resultant state. Instead an irrelevant visual resultant-state competitor (torn book) was either present (critical), or not (intact book) (control). Participants looked faster to the target if the resultant- state competitor was not present. In addition, more and longer looks were registered to the competitor in the critical condition than in the control condition. 

These findings suggest that processing bǎ activates an abstract, that is, a qualitatively unspecified representation for an affected object in the comprehender’s situation model. This representation interacts with the visual input, and leads to predictions about the linguistic input. We interpret predictions on the basis of such type of structural information as a special kind of incremental processing which has not been reported previously.


Adverbs of intentionality

Willi Geuder & Katja Gabrovska (Heinrich-Heine-Universität Düsseldorf)

Mental-attitude adverbials like English "intentionally" / German "absichtlich" are commonly analysed as predicates of events (e.g. Landman 2000), but usually without discussion of whether they differ from manner adverbs in their relation to the event variable. In this talk, we propose a frame model for representing certain components of intentional action, in particular proposing "action plan" as an attribute of event descriptions. This attribute forms one main branch of event frames, alongside those attributes that can be targeted by manner modifiers. We analyse the German adverb "absichtlich" as stating a matching relation between an action description (including its manners) and an agent's action plan. On the basis of this model, other modifiers from the same lexical field can also be analysed, esp. E. "unintentionally" / G. "versehentlich", as well as the meaning difference between G. "absichtlich" and E. "intentionally".


The processing of grammatical aspect in Polish – new experimental evidence

Dorota Klimek-Jankowska and Joanna Błaszczak (University of Wrocław)

Earlier studies examining the processing of aspect focused mainly on English or German, that is languages in which aspectual information is computed primarily on the basis of lexical aspectual information. Relatively little attention has been paid to the processing of aspect in languages in which aspectual meanings arise from the interaction of grammatical aspect and lexical aspect. In order to fill this gap, we intend to present the results of a series of our experiments: a self-paced reading and a related eye-tracking during reading study and an ERP study on the processing of perfective and imperfective aspect in Polish. The results of these experiments shed a new light on such questions as: (i) what is the domain of aspectual interpretation (V, VP or end of sentence) in languages with grammatical aspect; (ii) what are the neural underpinnings of the process of aspectual mismatch detection and resolution in languages which have a grammatical perfective/imperfective diachotomy?; (iii) is grammatical aspect lexically encoded as part of a verbal lexical entry or is it computed only in morphosyntax?; (iv) Is the semantic specificity asymmetry between imperfective and perfective aspect (imperfective aspect being more underspecified) reflected in the way these aspects are processed?      


Towards a compositional analysis of the an-construction in German

Katja Laptieva (IDS Mannheim), Sebastian Bücking (University of Tuebingen)

In German, some transitive verbs allow their direct object to be expressed as the internal argument of a prepositional phrase headed by an ‘at, on’. In contrast to the transitive structure in (1-a), the an-construction receives an atelic interpretation (Engelberg, 2007). The an-sentence in (1-b) is compatible with a scenario where only a part of a song was created and which does not specify whether the song was finished later on: (1) a. Paul Paul hat has ein-en a-acc Song song.acc geschrieben. written ‘Paul has written a song.’ b. Paul Paul hat has an at.prep ein-em a-dat Song song.dat geschrieben. written ‘Paul was writing a song.’ It has been usually assumed that the an-construction is a derived structure and that the an-phrase is a partitive variant of the theme argument of a transitive verb (Filip, 1999). That is, the verb meaning and the relationship between the verb and its (prepositional) object were taken to be identical in both argument structures. There are several arguments against the derivational analysis. First, the an-construction and the transitive structure are not always interchangeable, with the transitive expression being ungrammatical (am Aufstieg basteln ‘try to enter a better league’ vs. *den Aufstieg basteln). Second, obligatory transitive verbs like verzehren ‘consume’ or analysieren ‘analyze’ can not be used in the an-construction (die Banane verzehren ‘consume the banana’ vs. *an der Banane verzehren). Third, the corpus data reveal severe differences in the selectional preferences of the verbs in the two argument structures. For example, the direct object of basteln ‘craft, make’ usually refers to a concrete entity like a mask or a lantern, whereas in the prepositional variant the head noun of the an-phrase mostly denotes an abstract entity like a concept or an idea. In our approach, we depart from the modeling of the an-phrase as an alternative syntactic realization of the theme argument of the verb. Instead, we analyze the an-phrase as a modifier that combines with the atelic verb variant and introduces a contact relationship between its own internal argument and an entity that is derived by conceptual means from the meaning of the verb.


Engelberg, Stefan (2007). ” Konstruktionelle Varianten zwischen W¨orterbuch und Grammatik“. In: Germanistische Mitteilungen 66, S. 11–27.

Filip, Hana (1999). Aspect, eventuality types, and nominal reference. Outstanding dissertations in linguistics. New York: Garland Pub.


Act-cascades, TTs, and dot types

Sebastian Löbner (Heinrich-Heine-Universität Düsseldorf)

There are two theoretical approaches under discussion that deal with multiple categorization (or type assignment): The theory of "dot objects" / "dot types" (Pustejovsky 2009, Asher 2011) and Cascade Theory (Löbner, under review) based on Goldman's (1970) theory of level-generation. The two approaches share the attempt to deal with things of apparently multiple types (or categories), where the dot object approach was mainly applied to types of objects, and Goldman's and cascade theory to verbs and types of events. Still, the relationship between the two approaches is quite unclear. I will argue in this talk that there are close connections of cascade theory to dot type theory and that cascade theory may help to clarify certain foundational questions of the dot type approach, such as: (i) some of the ontological questions connected to it and (ii) the relationship between the types that belong to one dot type. It will not be claimed that all phenomena to which the dot object approach was applied can be captured by Cascade Theory, but some can – if cascade theory is properly extended as to apply to objects and events in general. One principal point of departure is the proposal recently brought forward for act cascades in Löbner (under review) to consider both theories as concerning not just ontological entities as such, but ontological entities-as-tokens-of-a-type, "TTs" for short. This is a step that is quite natural for semantic analysis since, whenever we talk about "things in the world" we talk of them as things-under-a-given-type-description: the words we use for referring to things and the contextual knowledge about the situation of utterance inevitable provide a type description / categorization of the things we refer to. Realizing this fundamental fact about language (as well as, more generally, about cognition) helps to overcome certain confusion concerning the ontology of dot objects or cascades (act-trees): the question whether we are dealing in such cases with one or more objects can be resolved straightforwardly if we relate it to TTs rather than just bare objects.


Asher, Nicholas (2011). Lexical Meaning in Context: A Web of Words. Cambridge: Cambridge University Press. 

Bücking, Sebastian (2014). 'Elaborating on events by means of English by and German indem.' In Christopher Piñón (ed.), Empirical issues in syntax and semantics 10. pp. 19-36, http://www.cssp.cnrs.fr/eiss10/ 

Goldman, Alvin I. (1970). Theory of human action. Princeton NJ: Princeton University Press. 

Löbner, Sebastian (under review). 'Cascades. Goldman's level-generation, multilevel categorization of action, and multilevel verb semantics.' https://semanticsarchive.net/Archive/2Y4YjdkM/ 

Pustejovsky, James (2009). 'The semantics of lexical underspecification.' Folia Linguistica, 32(3-4), 323–348. https://doi:10.1515/flin.1998.32.3-4.323


Unmögliche Interpolationen

Martin Schäfer (Universität Tübingen)

Parallelen zwischen Modifikation im Nominalbereich und Modifikation im Verbalbereich spielen eine prominente Rolle in derEreignissemantik. Mismatches zwischen diesen Bereichen spielen dagegen eine geringere Rolle. Bücking & Maienborn (in Begutachtung) werfen die Frage auf, warum die Interpretation von 'eine schnelle Zigarette' als 'eine schnell gerauchte/gedrehte Zigarette' durch die Interpolation eines Ereignisses möglich ist, gleiches aber in prädikativer Verwendung ausgeschlossen ist ('?Die Zigarette war schnell'). In meinem Vortrag möchte ich erstens einen Überblick über die entsprechende Datenlage im Englischen geben, und zweitens unterschiedliche Erklärungsansätze diskutieren.


Bücking, S. & Maienborn, C. (in Begutachtung). Coercion by modification - The  adaptive capacities of event-sensitive adnominal modifiers.


Event Unit Formation

Christiane von Stutterheim & Johannes Gerwien (Heidelberg University)

Events are fundamental units in human perception and cognition. The definition of what event units are varies across disciplines. A core criterion is related to the fact of ‘quality change over time’. This criterion, however, leaves room for variation across the dimensions of quality as well as time. Both dimensions are involved in what has been called ‘level of granularity’. While earlier studies on event unit formation have shown that subjects select different levels of granularity when segmenting the continuous stream of perception (Newtson’s experimental work 1973, Zacks & Radvansky 2014) it still is not clear which criteria subjects draw on, when forming event units. In the presentation we will argue for the hypothesis that language is a major factor in event unit formation. Languages differ with respect to the conceptual categories to which they attribute cognitive prominence on the basis of either grammaticalization or lexical differentiation. A language which requires obligatory marking of phases of an event by aspectual verbal categories, for example, forces its speakers to attend to these differences. A language which differentiates object specific features in the verbal lexicon of position verbs (sitzen, stehen, liegen) forces its speakers to attend to features of a visual input which provide the basis for selecting the respective verb. Cognitive processes which are motivated by linguistic structure in this sense are highly automatized, deeply entrenched in the course of first language acquisition.  

Earlier cross linguistic studies on event construal and verbal representation of events have shown that speakers of different languages a) segment visual input at different break points into  event  units  and  b)  select  different  components  of  the  visual  input  for  verbal representation (v. Stutterheim et al. 2012) So far these two aspects of event construal have not been investigated in their interrelation. A situation in the world is complex, in that different quality changes can take place at a given time interval, such as a leaf which is falling and rotating at the same time or a person who is walking and approaching a goal (Bennett 2002). In order to form an event unit an observer has to select a layer of the complex composition of different qualities which entails the relevant criteria for identifying break points. This is where language comes into play.  Experimental cross linguistic studies in the domain of motion events are taken as evidence for the role of the linguistically packaged conceptual categories in event unit formation. Results will be presented from both verbal and non-verbal tasks. Speakers of four languages which vary with respect to typological features in the domains of spatial and temporal cognition (French, Tunisian, German and English) saw short real world video clips. In one experiment they  were  asked  to  segment  the  input  non-verbally  (button-press-method),  in  a  second experiment they were asked to verbalise the scenes. Our hypothesis is supported by two findings: a) the cross linguistic differences found in segmentation  patterns  converge across  the  verbal  and  the  non-verbal task. b) The cross linguistic differences in the selection of information representing the scenes correspond to typological differences at the level of grammaticalization and lexicalization patterns. Speakers select different layers of the scenes depicted for forming event units.   

In conclusion the findings will be discussed in the context of theories of event cognition  
focusing on the role of language.


Bennett, Jonathan (2002): What events are. The Blackwell guide to metaphysics 43-65.

Newtson, Darren (1973): Attribution and the unit of perception of ongoing behavior. Journal of Personality and Social Psychology, 28, 28.

Radvansky, Gabriel A/Zacks, Jeffrey M (2014): Event cognition. Oxford University Press.  

Stutterheim, Christiane von/Andermann, Martin/Carroll, Mary/Flecken, Monique/Schmiedtová, Barbara (2012): How  grammaticized  concepts  shape  event  conceptualization  in  language  production:  Insights  from linguistic analysis, eye tracking data, and memory performance. Linguistics, 50, 833-867.


German wie-complements: Manners, methods and events in progress

Carla Umbach (Leibniz Zentrum Allgemeine Sprachwissenschaft / Universität Köln) 

In German, complement clauses embedded by the wh-word 'wie' ('how'/'like') have two different readings. The first one is a manner reading expressing a manner or method of doing something. The second reading is called "eventive" in the talk because it expresses an event in progress instead of a manner. Assuming that the 'wie' ('how'/'like') in the two readings is the same word, the question arises of why use a manner word to express an event in progress. 

The basic semantic hypothesis in the paper is that the wh-word 'wie' expresses similarity (as it does in e.g., similes). The paper starts from the observation that in the manner reading 'wie' has a base position next to the verb and is a modifier of the event type. In the eventive reading, on the other hand, it has a position above VP thereby adding information about the event token. The analysis includes two components: First, manners are considered as sets of similar events (instead of primitive objects). Methods, in particular, are considered as sets of similar sequences of subevents. Secondly, events in progress are seen as initial sequences which have a set of similar sequences as their continuations. From this point of view, an event in progress is structurally equivalent to a method built from sequences of subevents that share the same initial part. 

This analysis provides a semantic interpretation explaining the common role of the wh-word 'wie' in the two readings.


The (early) availability of future and perfect information in verbal clusters: Implications for the compositional semantics of composite tenses

Dennis Wegner, Marcel Schlechtweg & Holden Härtl (U Wuppertal / U Kassel)

German exhibits two configurations in which the canonical clause-final positioning of a finite element in embedded clauses may be disrupted: verbal clusters headed by future werden and perfect haben (dass Max den Vogel wird/hat singen hören). This correlation suggests that temporal auxiliaries may be preposed in order to reduce the processing load of a computationally challenging verbal structure by virtue of providing the relevant items as early as possible. The present paper discusses this hypothesis against the backdrop of evidence gained from a reading-time study which investigates the consequences of preposing for processing on the basis of a reaction-time-based forced-choice task. While the temporal relevance of the future auxiliary is trivial, our data show that something similar holds for the perfect auxiliary. In fact, the perfect auxiliary haben is bound to introduce relevant perfect semantics unless the auxiliar autonomously expresses perfectivity, which is restricted to predicates that exhibit specific event-structural properties and in turn gives rise to the sein-perfect.

For past events, visit the Events Archive >>>

This website uses cookies to ensure you get the best experience on our website.privacy policy