Das Oberseminar bietet Vorträge von eingeladenen Referentinnen und Referenten oder Kolleginnen und Kollegen aus der Abteilung an. Die Referierenden stellen aktuelle Forschungsergebnisse zu allen für die allgemeine und theoretische Linguistik relevanten Bereichen vor. Jeder ist herzlich eingeladen. Studierende werden besonders ermutigt, teilzunehmen, um forschungsbezogene Vorträge von Spezialisten aus erster Hand zu erleben.
Weitere Informationen zu Zeit und Ort der Oberseminare in diesem Semester finden sich unter den Reitern des jeweiligen Datums.
Lisa Beinborn (VU Amsterdam)
Zeit: Montag, 24. Juli, 14:15 - 15:45 Uhr
Titel: Cross-Lingual Transfer and Cognitive Complexity
Abstract: In my research, I focus on the interpretability of cross-lingual language models from a cognitive perspective. We analyze processes of cross-lingual transfer in humans and models using interpretability methods and cognitive data. I will summarize findings from two recent projects with Charlotte Pouw and Eliza Hobo: We predict eye-tracking signals of human processing complexity across languages, and investigate strategies for text simplification for English and Dutch.
Michael Hahn (Saarbrücken)
Title: Modeling word and morpheme order as an efficient tradeoff of memory and surprisal
A major goal of linguistic typology is to provide explanatory accounts of the systematic patterns constraining the variation between human languages. For instance, whether a language places objects before or after verbs is correlated with whether it has prepositions or post-positions (Greenberg's correlation universals). A long line of research has argued that such patterns are determined by the need for efficient communication under constraints imposed by human cognition (such as the boundedness of working memory). However, it has been hard to test these ideas systematically and at scale, in part because of the difficulty of extracting precise predictions from existing psycholinguistic theories. In this talk, we present an information-theoretic formalization of memory efficiency in human language processing in terms of a tradeoff between memory and surprisal, together with a method for estimating this tradeoff from corpora. On real-world text corpora in more than 50 languages, we find that word orders tend to provide efficient tradeoffs. Optimizing word orders for efficiency leads to the emergence of a prominent set of word order universals: Greenberg's harmonic correlation universals. Applying this idea on the word-internal level further reproduces typological patterns of morpheme ordering. Taken together, these results provide a new way of testing long-standing functionalist ideas about the sources of language universals.
Zeit: Montag, 17. Juli, 14:15 - 15:45 Uhr
Ort: Zoom, https://zoom.us/j/92494564543
Nathaniel Imel (Irvine)
Zeit: Dienstag, 04. Juli, 16 - 18 Uhr
Titel: Evolutionary dynamics lead to the emergence of efficiently compressed semantic systems
Human languages have been argued to support efficient communication. In particular, the world’s semantic category systems appear to approximate solutions to a dual optimization problem of minimizing the costs of mental representation, while maximizing accurate communication. This optimality-based account promises to explain important variation in the world’s vocabularies, with implications for linguistics, psychology and anthropology. It is unclear, however, what general evolutionary dynamics could lead to the emergence of efficient semantic systems. We aim to characterize these dynamics in a general way in order to gain a high-level understanding of how linguistic behavior evolves, abstracting over the details of various agent-based models. We show that a variety of simple population dynamics, applied to a model of imprecise imitation of communicative behavior, lead to the optimization of efficient meaning systems. Our results show that a process of regularization by noise -- irrespective of how it might result from specific cognitive architectures -- can shape the efficiency of language.
Mora Maldonado (Nantes)
Zeit: Montag, 03. Juli, 14 - 16 Uhr
Titel: Understanding the cross-linguistic distribution of person systems: A behavioural perspective
Systems of personal pronouns (e.g.,‘you’ and ‘I’) are used to refer to individuals as a function of their role in the context of utterance (e.g., speaker, addressee). While these systems vary widely across languages, not all ways of partitioning the person space are equally likely cross-linguistically. What are the forces that give rise to this constrained variation?
In the first part of the talk, I will present a series of artificial language learning experiments that investigate learning biases in how we represent the person space. Our results suggest that learners prefer person systems which (a) are based on natural classes, and (b) keep distinct forms to refer to the speaker. We take this to suggest natural class-based similarity and speaker-distinctiveness as two independent forces influencing the learnability of person systems.
In the second part of the talk, I will present some new work where I explore the idea that some of these constraints on person systems (i.e., the two independent forces mentioned above) are grounded on language-independent representations of pronominal referents, and available to speakers in non-linguistic tasks.
I will finish by showing how a similar approach can be extended to investigate the cross-linguistic distribution of other indexical expressions, such as locatives (e.g. here, there).
Daniel Lassiter (Edinburgh)
Zeit: Monday, June 19th, 14:15 to 15:45
Sprecher: Daniel Lassiter (Edinburgh)
Titel: Minimal semantics & rich pragmatics for conditionals
This talk is an overview of a programme for conditional semantics and pragmatics that I've been developing over the last few years, mostly focusing on indicative/hypothetical conditionals. In contrast to increasingly elaborate modal theories that dominate formal work in this area, I advocate an ultra-slim theory in which the only thing that conditional antecedents do is to modify the discourse context by introducing a temporary assumption. Everything else follows from the interaction of restricted contexts with assertion, presupposition, and operator domains. Since conditionals are complicated, a minimal semantics needs a rich but independently motivated pragmatics. To illustrate, I'll sketch two separate lines of research that converge to explain why felicitous use of a conditional frequently seems to require that the antecedent and consequent are relevant to each other. This result eliminates the motivation for a different kind of elaborate semantic theory of conditionals, the "Inferentialist" approach that has enjoyed recent popularity. While much more remains to be done, there is much reason to hope that the intricate semantic and pragmatic interactions that conditionals display can be derived from interactions of various independent components of meaning with an extremely simple discourse-oriented semantics.
Thomas Roeper, UMass Amherst
Zeit: Montag, 22. Mai, 14 Uhr
Ort: Hörsaal Mineralogie/LMB H102 (Lothar-Meyer-Bau)
Title: Embedding Propositional Attitudes within Propositional Attitudes: Could thought-syntax precede Syntax
Abstract: How exactly does the mind compose recursive PA structures? Is one of these sentences more difficult than the other:
a) John thinks Bill knows Mary is late.
b) John knows Bill thinks Mary is late.
Is embedding an opaque verb inside a factive verb different from the opposite? Children as young as 6yrs say things like: "I know you think I think Easter is 3 days long", among other complex sentences.
We will review a number of such examples and explore the hypothesis advanced by Chomsky that syntax is a direct reflection of thought. The claim raises the question of whether the mind might generate thought representations prior to syntactic ones. We will look at the implications of early remarks by children like "that's easy" and explore its implications for the Meaning First hypothesis advanced by Sauerland et al (2021, 2023). These observations lead to our hypothesis about the nature of what we can call Minimal Interfaces which in turn make predictions about acquisition.
Kenichi Okamoto, University of St. Thomas
Zeit: Monday, February 6, 4.15 pm - 5.45 pm
Meeting ID: 970 9073 3084
Title: Hypothesis testing reticulated linguistic phylogenies: Borrowed, inherited, both or neither?
Abstract: Observed similarities can arise across lineages (linguistic or biological) through (i) random chance, (ii) when lineages inherit a characteristic from a shared common ancestor, or (iii) when a lineage acquires the characteristic from another lineage (i.e., borrowing). Evaluating the relative importance of these mechanisms remains a major challenge in evolutionary biology and historical linguistics alike. Recently, biologists have come to leverage gene trees (phylogenies for a single gene across lineages) to quantify the likelihood that a reticulated, or networked, phylogeny explains observed genetic similarities and differences. I apply this approach to assess proposed language relations, using the vocabularies of so-called "Altaic" language families as a case study. I conclude by discussing how this approach can help us interpret lexical similarities in sprachbunds more broadly.
Zeit: Monday, January 30th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: A Simple Pragmatics for Language Development Research
Abstract: The study of language development presents a number of developmental puzzles. For example, according to an influential view (Tomasello 2008; Scott-Phillips 2015), children’s acquisition of natural languages is a consequence of uniquely human abilities for acting with and attributing communicative intent. These abilities are thought to be uniquely human, because they require uniquely human abilities for thinking about others’ mental states (abilities known as ‘Theory of mind’ or ‘mindreading’).
This view is problematic for a number of reasons. First, in relevant respects, infants’ mindreading abilities are not better than those of great apes. Second, the ToM abilities that they do possess seem to be inadequate for the ability to attribute communicative intent, at least on Gricean analyses. Third, the mindreading skills needed for Gricean communication seem to be developmentally dependent upon language, and so cannot explain language development. Additionally, while infants’ ability to understand informative pointing is often taken to be evidence of uniquely human abilities for understanding communicative intent, dogs also understand pointing. While captive chimpanzees are usually poor at pointing comprehension, enculturated chimpanzees do much better – suggesting that they may also understand communicative intent.
The possibility that a number of species attribute communicative intent remains difficult to reconcile with Gricean analyses of communication. In response to this puzzle, in this talk I sketch a cognitively undemanding account of acting with and interpreting communicative intentions. It is, I argue, suitable for explaining both children’s language development and the communicative interactions of a number of non-human species. It also presupposes no mindreading abilities that are inconsistent with data from developmental research. This simple pragamatics potentially solves a number of puzzles in the developmental literature. However, it requires us to acknowledge that there may be far more continuity between human and animal communication systems than has commonly been believed.
Liviu Dinu (University of Bucharest)
Zeit: Monday, January 23rd, 16:15-17:45
Zoom link: https://zoom.us/j/91089127778
Meeting ID: 910 8912 7778
Titel: Computer-assisted tools for generating and discriminating related words in historical linguistics
Abstract: Natural languages are living eco-systems, they are constantly in contact and, by consequence, they change continuously. Traditionally, the main HL problems (How are languages related? How do languages change across space and time?) have been investigated with comparative linguistics instruments.
We propose here, for Romance languages, computer-assisted methods for cognates identification, for discriminating between cognates and borrowings and between inherited and borrowing words, and for protoword reconstruction. Firstly, we introduce a method to automatically determine if a pair of words (u,v) are cognates or not, and we apply our method in the tasks of discriminating between cognates and borrowings and between inherited and borrowing words. Further we developed a methodology to automatically reconstruct the Latin proto-words from which the modern words evolved.
1. Alina Ciobanu, Liviu P. Dinu, 2019. Automatic Identification and Production of Related Words for Historical Linguistics. Computational Linguistics, vol. 45, No. 4, 667-704.
2. Alina Maria Cristea, Liviu P. Dinu, Simona Georgescu, Mihnea-Lucian Mihai and Ana Sabina Uban, 2021. Automatic Discrimination between Inherited and Borrowed Latin Words in Romance Languages In: Proc. EMNLP 2021(Findings), Punta Cana, 2021
3. Alina Maria Ciobanu and Liviu P Dinu, (2018). Ab initio: Automatic latin proto-word reconstruction. In Proc. COLING 2018, Santa Fe, New Mexico, USA, August 20-26, 2018
4. Alina Maria Ciobanu and Liviu P Dinu, 2014. Automatic detection of cognates using orthographic alignment. In Proc. ACL 2014, June 22-27, 2014, Baltimore, MD, USA
5. Alina Ciobanu and Liviu P Dinu, 2015. Automatic discrimination between cognates and borrowings. In Proc. ACL 2015, July 26-31, 2015, Beijing, China
6. Alina Ciobanu, Liviu P. Dinu, Laurentiu Zoicas, 2020. Automatic Reconstruction of Missing Romanian Cognates and Unattested Latin Words. In Proc. LREC 2020 Marseille, France, 2020
Zeit: Monday, January 16th, 16:15-17:45
onsite: Room 1.13, Wilhelmstraße 19
Titel: Signaling without Saying: The semantics and pragmatics of dogwhistles
Abstract: A dogwhistle is a piece of language that sends one message to an outgroup while at the same time sending a second (often taboo, controversial, or inflammatory) message to an ingroup. We propose an analysis of dogwhistles in the setting of social meaning games that treats them as signaling the persona of the speaker, and in some circumstances enabling an enrichment of the conventional meaning of the expression. We compare this account with views in terms of conventional implicature, invited inference, and classical gricean implicature. We further show how this formal framework allows, not just a account of dogwhistles, but opens up a way to analyze a variety of sociopragmatic phenomena involving trust, reliability, ideology, standpoints, etc. from a probabilistic game-theoretic perspective.
Yushuan Li (Zhejiang University)
Zeit: Friday, January 13, 10ct-12
Ort: Seminar room 1.13 in the SfS (Wilhelmstr. 19)
Titel: The Corpus of German Learners in China (CDLK) and Analyzing Syntactic Complexity Development in Writing of Adolescent German Learners
Abstract: My presentation will be divided into two parts, the first part being an introduction to the self-built German learner corpus of our team at Zhejiang University in China. German language learning has been booming in China. According to the survey in 2020, there are 20% more learners of German in all levels, i.e., school, university and adult education, than five years ago. Globally, only a few learner corpora (Falko, AleSKo, Kobalt, and BFSU corpus) contain written texts of Chinese Learners of German. The first three were constructed in Germany, containing only a few texts and only from advanced students (39, 43, and 20 texts at the B2 level, respectively). The BFSU corpus from China consists of exam essays written by Chinese German students, but it is not publicly available. It can be seen that there are limited text types and not comprehensive German levels in the current corpora. The CDLK (The Corpus of German Learners in China) is a multidimensionally annotated (Error annotation, lexical annotation, syntactic annotation) learner corpus that covers all levels or
learning stages (from beginner to intermediate level) and contains the texts of both high school and college students. In addition, texts
from the same learners are collected regularly over a three-year period.
With these premises in mind, the second part of my presentation will be a CDLK-based study of the development of syntactic complexity in
the written language of adolescent learners of German in China. Chinese adolescent learners of German are a growing group in China, and are likely to continue to increase. However, there is insufficient research on the acquisition characteristics of this group. By proposing a set of indicators that measure syntactic complexity, I would like to study the developmental characteristics of the syntactic complexity in writing of Chinese adolescent learners of German in order to understand their acquisition characteristics and to make suggestions for teaching German.
Alex Warstadt (ETH Zürich)
Zeit: Monday, January 9th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: The Poverty of the Stimulus from the Perspective of Neural Language Models
Neural language models have shown a remarkable ability in recent years to generate complex and grammatically well-formed texts, all through self-supervised learning on raw text data. These results call for a re-examination of long-standing debates about the poverty of the stimulus: How strongly does the linguistic input to a learner support the kinds of systematic grammatical generalizations observed during human language acquisition? I examine how neural language models generalize when presented with ambiguous or impoverished inputs. Results suggest that these models acquire a preference for generalizations based on hierarchical syntax and other grammatical features as their exposure to raw text increases, even when trained on impoverished data. I discuss the implications for our understanding of human language acquisition. Language models are far from perfect models of humans, but there are clear ways to improve their cognitive plausibility. Furthermore, they have advantages over human subjects in terms of ethics, expense, and expanded possibilities for experimental design, which enable us to tackle previously difficult-to-test hypotheses about the effects of the learner's environment on grammatical generalization.
Zeit: Monday, December 19th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: Neural networks and historical words
Abstract: In historical linguistics, cognates are words that descend in direct line from a common ancestor, called their proto-form, and therefore are representative of the evolution of their respective languages through time. These historical words allow linguists to better determine all manners of synchronic and diachronic linguistic relations. They are usually identified by discovering and matching the patterns which link them. Since neural networks are well known for their ability to latently learn interesting patterns in the data they see, what can they bring to historical linguistics and historical words reconstruction?
Zeit: Monday, December 5th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: Towards an RSA model of enriched compositionality: The case of logical metonymy
Abstract: First, the talk will focus on the semantic phenomenon of logical metonymy (the interpretation of sentences such as "John began the book" is enriched with an activity - reading - that is not made explicit in the sentence). This phenomenon has received attention because it is a clear case of mismatch between the syntactic structure of the sentence and its semantics, and suggests the influence of world knowledge in the process of interpretation. We will show how experimental data regarding the factors that drive a speaker to use logical metonymy can be predicted by a Bayesian model of communication inspired by RSA theory. The Bayesian model that will be presented contains some innovations from classical RSA theory, particularly concerning the way in which meanings are represented and construed by the literal listener. The proposed model will provide an opportunity to reflect on some possible ways to extend the RSA model in order to account for the incrementality of meaning construction and its sensitivity to world knowledge, and thus make it possible for the model to reason about a wider set of possible meanings and utterances.
Zeit: Monday, November 28th, 16:15-17:45
Ort: Wilhelmstraße 19, room 1.13
Meeting ID: 965 6472 7294
Titel: Exceptive-additive besides
Besides is similar to exceptive constuctions like except/but. Exceptive and exceptive-additive constructions behave similarly when they occur with universal quantifiers.
(1) Every girl besides/but/except Ann came.
The exceptive inference:
⇝ Ann did not come.
Besides differs from except/but. Exceptive constructions are known to be incompatible with non-universal quantifiers (Horn 1989). Exceptive-additive constructions can occur in such contexts and they give rise to an additive inference.
(2) Some girl(s) besides/*but/*except Ann came.
(3) (Exactly) one girl besides/*but/*except Ann came.
(4) At least/more than one girl besides/*but/*except Ann came.
(5) Fewer than/at most two girls besides Ann came.
The additive inference:
⇝ Ann came.
In this work we offer a unified semantic account of exceptive and additive uses of besides by:
- Providing a unified treatment for every and exactly n cases in terms of domain subtraction and Exh, where Exh negates the alternative formed by substitution of besides by including.
(6) [IP2 Exh [IP1 [every/exactly one girl [besidesF Ann]] came ]]
- Explaining why in the first case the inference is negative and in the second it is positive.
- Extending this account to all other cases of modified numerals and existentials by adopting a decomposition account of them.
(7) [IP3 at least one [IP2 1 [IP1 Exh [[ exactly d1 girl besidesF Ann ] came]]]]
Zeit: Monday, November 7th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: Motivating decoding from language generators using characteristics of human communication
Abstract: Today's probabilistic language generators fall short when it comes to producing coherent and fluent text despite the fact that the underlying models perform well under standard metrics, e.g., perplexity. This discrepancy has puzzled the language generation community for the last few years. In this talk, we’ll take a different approach to looking at generation from probabilistic language generators, pulling on concepts from psycholinguistics and information theory in the attempt to provide insights into some observed behaviors of these models, e.g., why high-probability texts can be dull or repetitive. Humans use language as a means of communicating information, aiming to do so in a simultaneously efficient and error-minimizing manner; in fact, psycholinguistics research suggests humans choose each word in a string with this subconscious goal in mind. We propose that decoding from probabilistic language generators should attempt to mimic these behaviors. To motivate this notion, we’ll look at common characteristics of several successful decoding strategies, showing how their design allows them to implicitly adhere to attributes of efficient and robust communication. We will then propose a new decoding strategy, with the explicit aim of encoding these human-like properties of natural language usage into generations.
Zeit: Monday, October 24th, 16:15-17:45
Zoom link: https://zoom.us/my/michael.franke.tuebingen
Titel: Exploring a common computational framework to study the evolution and development of human communication
Abstract: Human communication has been described as a contextual social inference process: listeners use utterances and social-contextual information to make inferences about speakers’ underlying intentions. Research into great ape communication has been inspired by this view to look for the evolutionary roots of the social and cognitive processes involved in human communication. This approach has been highly productive, yet it is often compromised by a too-narrow focus on how great apes use and understand individual signals. I will present a computational framework that formalizes great ape communication as a multi-faceted social inference process. This model makes accurate qualitative and quantitative predictions about real-world communicative interactions between semi-wild-living chimpanzees. When enriched with a pragmatic reasoning process, the model can be used to explain repeatedly reported differences between humans and great apes in the interpretation of ambiguous signals (e.g. pointing gestures). Importantly, the same modeling framework can be used to study word learning in young children. Taken together, our approach provides a new tool kit for studying the evolution of human communication. It illustrates some deep similarities between the ways in which humans and great apes communicate, but also specifies in what ways human communication might be unique.
Zeit: Monday, July 25th, 4.15 pm - 5.45 pm
Ort: SfS, room 1.13 & Zoom (hybrid event)
Zoom link: https://zoom.us/j/99299406988
Meeting ID: 992 9940 6988
Title: Co-evolution of Tool Use and Language: An experimental archaeological approach integrating the use of EEG and EMG
Abstract: Technological innovation played a crucial role in the biocultural evolution of our human lineage. The earliest emergence of lithic tool technologies represents one of the major subjects of debate, with many researchers speculating that the early hominins who produced and used these early artifacts may had already evolved advanced manipulative and cognitive abilities, including the capacity for higher-level communication. In this regard, a series of neurological studies have recently demonstrated that tool use/making and language present functional similarities that rely on largely common brain activation patterns, providing support to the hypothesis that manual praxis may have provided the scaffold for complex forms of communication. Even though previous research has addressed the biomechanical and cognitive requirements for tool manipulation as well as its neural correlates with language communication, the interaction between these two has never been investigated together under real-time experimental conditions. Moreover, no experimental study to date has focused on the potential neural correlations between early stone tool use and gestural communication. In this framework, the Junior Research Group “Co-evolution of tool use and language” is developing and employing a novel experimental approach for identifying patterns of brain-hand coordination using electroencephalography (EEG) and electromyography (EMG), involving tasks of early hominin stone tool use and basic tool-related communication (both gestural and verbal).
Jennifer Hu (MIT)
Zeit: 16.15 - 17.45 Uhr
Access link: https://zoom.us/my/michael.franke.tuebingen
Title: Investigating ad-hoc scalar implicatures
Abstract: Scalar implicature (SI) is traditionally considered a hallmark of generalized conversational implicature (Levinson, 2000). This view has been challenged by recent studies demonstrating variance in SI rates within and across scales (e.g., Doran et al. 2009; Degen 2015; van Tiel et al. 2016), suggesting that SI depends on context and the structure of the scale itself. However, these studies focus on scales ordered by entailment (e.g., <some, all>, <warm, hot>), while little attention has been given to SIs arising from ad-hoc ordering relationships (Hirschberg 1985), which are common in naturalistic communication. In this ongoing work, we test the hypothesis that SI rates — across both entailment-based and ad-hoc scales — depend on the listener’s confidence in the underlying scale, separately from the relationship between the weak and strong scalar items. First, we use an artificial language model to test whether uncertainty about the scale predicts SI for a set of entailment-based scales (van Tiel et al., 2016). Next, we conduct a pilot behavioral experiment using novel materials featuring ad-hoc scales in naturalistic sentences. Our preliminary results suggest that scalar uncertainty may predict SI for this broad range of scales. This suggests a potential unified mechanism underlying entailment-based and ad-hoc SI, further challenging the distinction between generalized and particularized conversational implicatures.
Colin Twomey (University of Pennsylvania)
Zeit: 16.15 - 17.45
Access link: https://zoom.us/j/95896132814
Meeting ID: 958 9613 2814
Title: What we talk about when we talk about colors
Abstract: Names for colors vary widely across languages, yet color categories are remarkably consistent. Shared mechanisms of color perception help explain these shared patterns and have been the focus of past work. But the mappings from colors to words are far from identical across languages, which may reflect differences in communicative needs – how often speakers must refer to objects of different colors. A link between compression and categorization in natural language gives us a new way to look for the key factors shaping color vocabularies in 130 languages around the world. We introduce a new approach to inference using this link, and reveal a hidden diversity in communicative needs across linguistic communities. We show that the extensive variation in needs can be explained in part by differences in geographic location and local biogeography, while commonalities in the color regions of greatest need are correlated with the colors of salient objects, including ripe fruits in primate diets. Our work reconciles opposing theories of color naming, while opening new directions to study cross-cultural variation in communicative needs and its impact on the cultural evolution of color categories.
Elisa Kreiss (Universität Stanford)
Zeit: 19.15 - 20.45 Uhr
Access link: https://zoom.us/my/michael.franke.tuebingen
Title: Towards Context-Sensitive Image Descriptions with a Purpose
Abstract: Images are pervasive across the Web, and they are generally tightly connected to natural language that is intended to serve a variety of purposes -- e.g., the articles and tweets they appear in, or the language read out by screen readers to make those images non-visually accessible. We use the term image-based Natural Language Generation to summarize the efforts to artificially generate this class of image-based texts. While these different kinds of texts share a connection to images, summarizing them as a single task as is suggested in much of the "image captioning'' literature is highly misleading. In this work, we argue that for developing models that can usefully generate this variety of texts in practice, image-based Natural Language Generation needs to be guided by two central considerations: (1) The communicative purpose the specific type of text needs to serve, and (2) the context the image is embedded in. In this talk, I'll describe a new publicly available Wikipedia-based dataset called Concadia and first model results showing the benefit for considering both components with a special focus on image descriptions for accessibility. Finally, I'll discuss the opportunities of integrating Questions Under Discussion (QUDs) in the description generation process and the challenges that come with such an approach.
Zeit: 16.15 - 17.45
Access link: https://zoom.us/j/95896132814
Meeting ID: 958 9613 2814
Title: Blue and green, or grue? The color lexicon at the intersection of the environment, genetic and culture
Abstract: The great diversity of the world’s language is reflected in the way people label the color spectrum. As an example, while some languages have separate words for “green” and “blue”, others lump the two together (“grue”). Earlier proposals suggest that the diversity of the cultural and physical environments in which populations live, as well as individual differences in color perception, may cause this variation. However, the exact role of these factors and their interaction has remained unclear. During this talk, we present the results of analyzing 142 populations, showing that color lexicon is affected by several aspects of the environment: cultural complexity, distance to water, and - of particular importance - the amount of ultraviolet light. Ultraviolet light negatively affects the lens of the eye cumulatively across the lifespan, resulting in less blue light reaching the retina. Across generations, this ontogenetic effect may generate negative selection against the genetically-determined abnormal red-green vision. Together, this shows that languages can only be understood in the context of their cultural, biological, and physical environments. Also, it put forward the idea that differences in physiology might generate biases that, amplified by the repeated use and transmission of language, may contribute to shaping languages.
Sina Zarrieß (Universität Bielefeld)
Zeit: 16.15 - 17.45 Uhr
Access link: https://zoom.us/j/99368857478
Meeting ID: 993 6885 7478
Title: Linguistic Variability and Communicative Effectiveness: Challenges in Neural Language Generation
Abstract: When speakers converse, their utterances are remarkably diverse and, at the same time, remarkably precise and effective. For instance, in a widely used corpus of human descriptions of images showing common objects, Devlin et al. (2015) find that 99% of the image captions are unique. Other work that has collected such descriptions in interactive, game-based settings found that speakers often only need a few words to unambiguously refer to objects or generally make themselves understood in a rich, communicative context.
Handling the variability of utterances that speakers are able to use with such a high degree of effectiveness is still a central challenge in conversational systems that generate natural language. In this talk, I will discuss recent attempts in NLG (natural language generation) at making systems more diverse or more effective, showing that these objectives are often at odds. I will discuss some of our recent work on decoding methods for neural NLG that aims at overcoming trade-offs between quality, diversity and communicative effectiveness.
Zeit: 16.15 - 17.45 Uhr
Access link: https://zoom.us/j/95896132814
Meeting ID: 958 9613 2814
Title: The emergence of words from iterated sound imitations
Abstract: There has been an ongoing debate concerning the modality in which language has emerged (cf. Wacewicz and Żywiczyński 2020). The gestural scenarios of language origins are given more prominence, with experimental studies typically focusing on the role of visual iconicity in the evolution of gestural communication and sign languages (Armstrong 2007; Fay et al. 2014; Goldin-Meadow 2016; Zhang 2016; Silva et al. 2020; Fröhlich 2019).
The current research contributes to the language origins debate as it addresses an empirically underexplored vocal-auditory modality and the power of imitation in the development of first words and follows the novel line of research on spoken languages and sound symbolism (cf. Ćwiek et al., 2019; Edmiston et al., 2018; Perlman et al., 2015; Pernis & Vigliocco, 2014; Imai & Kita, 2014).
Following Edmiston et al. (2018), we use iterated learning paradigm to scrutinize the emergence of words from uninstructed repeated one another’s vocal productions of environmental sounds, e.g., glass breaking, clock ticking (cf. the children’s “Telephone game”). Our project aims at understanding the mechanism of the evolution of the sound lexicon; thus, answering the question: Is the emergence of words through iterated vocal imitations of environmental sounds universal, that is, language-independent?
The currently available data show that vocal imitations may stabilize in form and function (as measured by an increase of acoustic and orthographic similarity). Yet, we cannot generalize over other language groups than investigated English language speakers. The populations tested in our study are Polish, German, and Chinese native speakers which will allow us to verify whether the emergence of words through iterated vocal imitations of environmental sounds is a universal phenomenon.
Armstrong, D. F., & Wilcox, S. E. (2007). The gestural origin of language. Oxford University Press.
Brown, S. (1999). The ‘Musilanguage’ model of language evolution. in The Origins of Music, eds. S. Brown, B. Merker, and N. L. Wallin. 271–300. Cambridge, MA: MIT Press.
Blasi, D. E., S. Wichmann, H. Hammarström, P. F. Stadler & M. H. Christiansen. (2016). Sound–meaning association biases evidenced across thousands of languages. Proceedings of the National Academy of Sciences, 113(39), 10818–10823.
Ćwiek, A., Draxler, Ch., Fuchs, S. Kawahara, S., Winter, B. & Perlman, M. (2019) Comprehension of Non-Linguistic Vocalizations across Culture Presentation at ICLC 2019, Japan.
Edmiston, P., Perlman, M., & Lupyan, G. (2018). Repeated imitation makes human vocalizations more word like. Proceedings of the Royal Society B, 285, 20172709.
Fay et al. (2014). Creating a communication system from scratch: gesture beats vocalization hands down. Frontiers in Psychology, 5(354). doi: 10.3389/fpsyg.2014.00354.
Fröhlich, M., Sievers, C., Townsend, S. W., Gruber, T., & van Schaik, C. P. (2019). Multimodal communication and language origins: integrating gestures and vocalizations. Biological Reviews, 94(5), 1809-1829. https://doi.org/10.1111/brv.12535
Goldin-Meadow, S. (2016). What the hands can tell us about language emergence. Psychonomic Bulletin & Review, 24(1), 1–6.
Imai M, Kita S. (2014) The sound symbolism bootstrapping hypothesis for language acquisition and language evolution. Phil. Trans. R. Soc.B 369, 20130298.
Perniss P, Vigliocco G. (2014) The bridge of iconicity: from a world of experience to the experience of language. Phil. Trans. R. Soc. B 369, 20130300.
Perlman, M., Dale, R., & Lupyan, G. (2015). Iconicity can ground the creation of vocal symbols. Royal Society Open Science, 2(8), 150152–16.
Wacewicz, S., & Żywiczyński, P. (2020) Pantomimic Conceptions of Language Origins. In A. Lock, C. Sinha, N. Gontier (eds.), Oxford Handbook of Human Symbolic Evolution, 2nd edition, Oxford University Press, in press.
Zhang, E. Q. 2016. Why did vocalizations come to predominate gestures in the evolution of language? Ducog 2016, Dubrovnik, Croatia.
Steven Moran (Université de Neuchâtel)
Zeit: 10.15 - 11.45 Uhr
Access link: https://zoom.us/j/93972792983
Meeting ID: 939 7279 2983
Title: Human speech sounds: three evolutionary timelines
Abstract: In order to understand the evolutionary origins of speech, and to perhaps shed light on the origins of language, we need to identify the biological and cultural factors that shape it. In my presentation, I will discuss how we can investigate the biological and cultural pressures on speech sound production with respect to three evolutionary timelines. First, what parallels exist between human and nonhuman vocalizations? Second, how can we investigate which speech sounds extinct hominins, such as Neanderthals, could have produced? Third, which speech sounds today are determined by our biology and which are due to cultural pressures? Since very little is known about the origin of language, including whether it evolved suddenly or gradually during the evolution of our species, any insights from studying the evolution of speech may help pinpoint when and how it evolved.
Zeit: 10.15 - 11.45 Uhr
Access link: https://zoom.us/j/97040596620
Meeting ID: 970 4059 6620
Title: Grammatical restructuring and population dynamics in northwestern Bantu gender systems
Abstract: This talk investigates animacy-based semantic restructuring in Bantu gender systems and the non-linguistic factors that may favor it. Our focus is on variability in patterns of gender agreement, and we present the results of two studies based on a sample of 179 northwestern Bantu (henceforth NWB) languages. Our dataset consists of information on the kind of marking (syntactic, animacy-based, both, or none) that is found on a set of fifteen agreement targets (such as adnominal modifiers, various kinds of pronouns, and predicates). We first present a bottom-up typology of gender systems in NWB. We find that highly eroded gender systems can be explained as a result of the evolutionary dynamics by which animacy-based semantic agreement rises and spreads in more conservative languages. Our data also confirm that animacy-based semantic agreement is more likely to first appear on predicates and independent personal pronouns before spreading to different types of adnominal modifiers, which is in line with the predictions of the Agreement Hierarchy (Corbett 1979, 2006). The second part of the talk focuses on testing the hypothesis that animacy-based agreement and heavily restructured gender systems in NWB can be explained as a function of population history – at least in part. We find that animacy-based gender systems are more common (1) in NWB languages of wider communication (animacy-based agreement is favored in virtue of its higher learnability) and (2) in NWB languages that are closely situated to non-Bantu languages (animacy-based restructuring is due to intense language contact and shift). To our knowledge, this is the first quantitative cross-linguistic study that confirms the oft-repeated claim that situations of intense language contact favor the restructuring and erosion of grammatical gender.
Zeit: 10.15 - 11.45 Uhr
Access link: https://zoom.us/j/97560603153
Meeting ID: 975 6060 3153
Title: Contact-tracing in cultural evolution: a Bayesian mixture model to detect geographic areas of language contact
Abstract: When speakers of different languages interact, they are likely to influence each other: contact leaves traces in the linguistic record, which in turn can reveal geographic areas of past human interaction and migration. However, other factors may contribute to similarities between languages. Inheritance from a shared ancestral language and universal preference for a linguistic property may both overshadow contact signals. How can we find geographic contact areas in language data, while accounting for the confounding effects of inheritance and universal preference? To approach this issue we developed sBayes, an algorithm for Bayesian clustering in the presence of confounding effects. The algorithm learns which similarities are better explained by confounders, and which are due to contact effects. Contact areas are free to take any shape or size, but an explicit geographic prior ensures their spatial coherence. In a first study, we tested sBayes on simulated data and applied it in two case studies to reveal language contact in South America and the Balkans.
Ryan Cotterell (ETH Zürich)
Zeit: 16.00 - 17.30 Uhr
Access link: https://zoom.us/j/97572048759
Meeting ID: 975 7204 8759
Title: Meaning to Form: Measuring Systematicity as Information
Abstract: My will focus on a longstanding debate in semiotics centers on the relationship between linguistic signs and their corresponding semantics: is there an arbitrary relationship between a word form and its meaning, or does some systematic phenomenon pervade? For instance, does the character bigram ‘gl’ have any systematic relationship to the meaning of words like ‘glisten’, ‘gleam’ and ‘glow’? We offer a holistic quantification of the systematicity of the sign using mutual information and recurrent neural networks. We employ these in a data-driven and massively multilingual approach to the question, examining 106 languages. We find a statistically significant reduction in entropy when modeling a word form conditioned on its semantic representation. Encouragingly, we also recover well-attested English examples of systematic affixes. We conclude with the meta-point: our approximate effect size (measured in bits) is quite small—despite some amount of systematicity between form and meaning, an arbitrary relationship and its resulting benefits dominate human language.
Gerrit Bauch (Universität Bielefeld)
Zeit: 10.15 - 11.45 Uhr
Access link: https://zoom.us/j/92671301312
Meeting ID: 926 7130 1312
Title: Effects of Noise on the Grammar of Voronoi Languages
Abstract:We study a signaling game of common interest in which a stochastic noise is perturbing the communication between sender and receiver. Despite this inhibiting factor efficient languages still exist. In any equilibrium, sender uses a tessellation consisting of convex cells while receiver uses Bayesian estimators as interpretations. Low levels of error that respect the distance between words lead to concise interpretations in the decoding process. Comparative statics for increasing noise describe robustness of different grammatical structures. Evolutionary modeling approaches converge to equilibria, but not every equilibrium is stable.
Shravan Vasishth (Universität Potsdam)
Zu diesem Oberseminar gibt es einen Mitschnitt.
Zeit: 11.00 - 12.30 Uhr
Access link: https://zoom.us/j/96320858663
Meeting ID: 963 2085 8663
Title: Individual differences in cue-weighting in sentence comprehension: An evaluation using Approximate Bayesian Computation
Abstract: Cue-based retrieval theories of sentence processing assume that syntactic dependencies are resolved through a content-addressable search process. An important recent claim is that in certain dependency types, the retrieval cues are weighted such that one cue dominates. This cue-weighting proposal aims to explain the observed average behavior. We show that there is systematic individual-level variation in cue weighting. Using the Lewis and Vasishth cue-based retrieval model, we estimated individual-level parameters for processing speed and cue weighting using data from 13 published reading studies; hierarchical Approximate Bayesian Computation (ABC) with Gibbs sampling was used to estimate the parameters. The modeling reveals a nuanced picture about cue-weighting: we find support for the idea that some participants weight cues, but not all do; and only fast readers tend to have the predicted cue weighting, suggesting that reading proficiency might be associated with cue weighting. A broader achievement of the work is to demonstrate how individual differences can be investigated in computational models of sentence processing using hierarchical ABC.
Johann-Mattis List (MPI für Menschheitsgeschichte Jena)
Zeit: 11.00 - 12.30 Uhr
Access link: https://zoom.us/j/94224749668
Meeting ID: 942 2474 9668
Title: Cross-linguistic language technologies. Data, Methods, and Analysis
Abstract: In the past two decades the amount of digitally available linguistic datasets has been constantly increasing and many new methods have been proposed in order to analyze large cross-linguistic datasets. Unfortunately, however, only a small fraction of the digitally available data is also integrated in the sense that the data for one particular problem or the data from one particular source can be directly compared with data compiled for different problems or from different sources. In order to deal with this problem, new cross-linguistic language technologies are needed. These technologies help to integrate linguistic datasets across many different languages by providing (a) detailed standards for cross-linguistic data formats, (b) new methods with which data can be converted into the required data formats, and (c) new analyses that target cross-linguistically integrated data. In the talk, I will briefly introduce what has been done so far with respect to data integration and cross-linguistic language technologies and then discuss chances and challenges for future work.
Michael Franke (Universität Osnabrück)
Zu diesem Oberseminar gibt es einen Mitschnitt.
Zeit: 11.00 - 12.30 Uhr
Access link: https://zoom.us/j/92255632073
Meeting ID: 922 5563 2073
Title: Theory-driven probabilistic modeling of language use: a case study on quantifiers, logic and typicality [joint work with Bob van Tiel (Nijmegen) and Uli Sauerland (Berlin)]
Abstract: Theoretical linguistics postulates abstract structures that successfully explain key aspects of language. However, the precise relation between abstract theoretical ideas and empirical data from language use is not always apparent. Here, we propose to empirically test abstract semantic theories through the lens of probabilistic pragmatic modelling. We consider the historically important case of quantity words (e.g., `some', `all'). Data from a large-scale production study seem to suggest that quantity words are understood via prototypes. But based on statistical and empirical model comparison, we show that a probabilistic pragmatic model that embeds a strict truth-conditional notion of meaning explains the data just as well as a model that encodes prototypes into the meaning of quantity words.
Natalia Levshina (MPI for Psycholinguistics Nijmegen)
Zu diesem Oberseminar gibt es einen Mitschnitt.
Zeit: 11.00 - 12.30 Uhr
Access link: https://zoom.us/j/92536167298
Meeting ID: 925 3616 7298
Title: From binary trade-offs to a causal network of Subject and Object cues: A typological corpus-based study based on Universal Dependencies
Abstract: In recent years, the notion of efficient trade-offs between different motivations and types of cues has become very popular in functional and typological approaches to language. The present paper is a case study of different cues that are used to express the core grammatical relationships of Subject and Object, which convey “who did what to whom”. These cues are case marking, rigid word order of Subject and Object, tight semantics and verb-medial order. They are inferred from online language corpora in thirty languages, annotated with the Universal Dependencies. Correlational and causal analyses show that the cues are not used very efficiently. Some of the cues are positively correlated, and the relationships between different cues are not bidirectional. This study suggests that the opportunities given to language users for exhibiting rational behaviour and using their language efficiently are very limited, and shaped by factors very different from rationality and communicative efficiency.
Bill Thompson (Princeton University)
Zeit: 16.00 - 17.30 Uhr
Access link: https://zoom.us/j/92895214463
Meeting ID: 928 9521 4463
Title: How Translatable are Common Words? Some Answers from Distributional Semantics
Abstract: We analysed the semantic networks of 1,016 concepts in 41 languages using distributional models of lexical semantics. We examined which semantic domains (e.g. animals, emotions, body parts and numbers) show the most and least alignment between different languages, and whether alignment is greater for more concrete terms (it is not). We examined how alignment varies for different parts of speech, and how it relates to human judgements of similarity and to lexical factors such as frequency and neighborhood density. The alignment between one language and another is statistically related to the cultural and historical relatedness of the languages, offering insights into the processes of cultural evolution that influence natural language semantics.
T. Mark Ellison (Universität zu Köln)
Zeit: 11.00 - 12.30 Uhr
Access link: https://zoom.us/j/97783628419
Meeting ID: 977 8362 8419
Title: A Bayesian Model of Prominence in Language
Abstract: This talk comes in two parts. In the first half, I will talk about modelling language in ways that focus on the probabilistic relationship between language and speakers. On the one hand, language arises when two speakers come together and interact. The properties of those speakers determine the distributions of language behaviours that happen between them. On the other hand, we can think of individual speakers as links between their conversations: for example, bilinguals link the speech communities to which they belong, becoming a conduit for influences between their languages. I will present some Bayesian studies of language aimed at showing the history of languages, or of their speaker groups.
The second half of the talk focusses on one feature of language in interaction that arises from the nature of the participants, namely, the use of code prominence. I introduce a simple model of communication in which the listener performs Bayesian inference to determine the intended meaning of a communicated form. The speaker internally simulates listener interpretation - in line with the forward modelling account of Pickering & Garrod (2013) - and varies their choice of linguistic form to ensure that the listener’s confidence in their interpretation remains above a certain threshold. If we assume that speakers ensure this communicative fidelity while nevertheless minimising their articulatory effort, we can explain a number of phenomena associated with prominence. The talk concludes with discussion of how such prominence affects language over time.
Oliver Bott (Universität Bielefeld)
Zeit: 14.15 - 15.45 Uhr
Title: Remention Biases Affect the Choice of Anaphoric Form (joint work with Torgrim Solstad)
Abstract: The choice of anaphoric forms (e.g. Mary vs. she) depends on a number of factors such as grammatical function, order of mention or topicality (Arnold 2008). For semantic/pragmatic re-mention biases however, which also impact referent salience, recent research has found conflicting results. Thus, Implicit Causality verbs of Stimulus-Experiencer (e.g., fascinate) and Experiencer-Stimulus type (e.g., admire) display strong preferences for subsequent explanations about the Stimulus argument (Ferstl et al. 2011). Yet, Fukumura & van Gompel (2010) and Rohde & Kehler (2014) found no effect of Implicit Causality on anaphoric form. Kehler & Rohde (2013) a.o. thus claim that the production of anaphoric form is dissociated from the likelihood of mention. On the other hand, Rosa & Arnold (2017) found that Transfer of Possession verbs, with a re-mention bias for goal arguments (e.g., the indirect object of give or the subject of get) do influence the choice of anaphoric form.
Rohde & Kehler (2014), improving on Fukumura & van Gompel's (2010) paradigm, point out that the choice of anaphoric form is especially important in contexts with two same-gender referents as a strategy to avoid ambiguity (cf. Levinson's (1987) m-implicatures). However, unlike Fukumura & van Gompel 2010 and Rosa & Arnold 2017, Rohde & Kehler did not use a forced-reference paradigm, in which participants are prompted to provide continuations for one particular referent. This is of particular importance when comparing bias congruent and bias-incongruent continuations, though. The present study presents a direct comparison of the effects of the two re-mention biases across a total of four experiments, applying a combination of the forced-reference paradigm with both same-gender and different-gender conditions. We ran the experiments in German, as opposed to the just reviewed ones, which all tested for influences of pragmatic biases on anaphor production in English. Whereas English has a rather restricted inventory of anaphoric forms available for coreference (Gundel et al. 1993), German has both personal (e.g., er/sie 'he/she') and demonstrative pronoun (e.g., dieser/diese 'this one' and jener/jene 'that one') paradigms and we hypothesized that this richness in forms could facilitate the elicitation of form-based effects
In this talk I will present data from four story continuation experiments employing a forced reference paradigm looking more closely into the underlying factors that drive form selection of referential expressions. In sum, the results of our first three experiments show that – modulated by well-known effects of audience design – referential biases affect reference form production across verb classes, including Implicit Causality verbs. This finding adds to the evidence in Rosa & Arnold 2017 and speaks against proposals assuming a general dissociation between likelihood of mention and choice of anaphoric form (Kehler & Rohde 2013). However, given the conflicting evidence from a fourth experiment with respect to the anaphoric category targeted by remention biases (demonstrative pronouns vs. proper names), our experiments call for a pragmatic explanation taking into account not only audience design but also the broader pragmatic context created by the experiment.
Ralf Vogel (Universität Bielefeld)
Title: Syntactic inventions
Abstract: It is a consequence of the widely assumed frequentist approach to grammaticalisation that some inventions that speakers make occur too rarely to induce language change. Still, these inventions are synchronic phenomena that at the same time are based on the language system in its current state, and do not follow from it (completely). I will show with case studies from several phenomena of German morphosyntax (verbal complexes, reflexivisation and case conflicts in German) that this slightly paradoxical idea may help to improve our understanding of those phenomena, and perhaps linguistic competence more broadly.
Of special interest for linguistic theory is the fact that such rare inventions are not arbitrary and do not pose any comprehension problems on the side of the addressee. They are based on the linguistic system in its current historical state, and created by speakers applying general mechanisms in a manner that is transparent to the addressee. These general mechanisms are part of linguistic competence and underlie the speakers’ ability to create new items, words or constructions.
I use the term ad hoc constructions for these special morphosyntactic phenomena.
What makes them particularly interesting, are the following properties: 1. they have unexpected features that do not follow from the linguistic units they are based on; 2. they can be shown to occur indeed too rarely to have a chance to grammaticalise; 3. they are nevertheless preferred systematically over alternative variants and have a sufficiently high acceptability rating. Each of these points will be shown to hold for the phenomena I am discussing on the basis of experimental and corpus studies.
The grammatical analysis will combine tools from the theory of generalised conversational implicature, construction grammar and optimality theory.
Elke Teich (Universität des Saarlandes)
Title: Conventionalization in diachronic linguistic change: the case of Scientific English
Abstract: The topic of this talk is conventionalization (i.e. the longer-term linguistic effects of repeated interaction) and its benefits for communication (i.e. message transmission). Widely acknowledged as a relevant process in language change, conventionalization provides a prerequisite for innovation (de Smet 2016) and may lead to grammaticalization as well as the formation of registers (Weinreich et al. 1968, Harris 1991). I will elaborate the idea that conventionalization is a cornerstone in in changing language use because it serves the maintenance of communication function by inducing significant surprisal and entropy-reducing effects. To show this, we pursue an exploratory, corpus-based approach, focusing on scientific writing (Degaetano-Ortlieb and Teich 2019), a well-studied and fairly controlled domain, and its evolution across 250+ years from the mid-17th century onwards. The data set we use is the Royal Society Corpus¹. To capture lexical and syntactic aspects of changing language use, we employ computational language models; and to evaluate the observed effects, we apply various measures of information content (surprisal, entropy, relative entropy). We find for instance that diachronically, within the scientific domain, relative entropy on n-gram models overall decreases pointing to converging language use over time but this is more pronounced for the grammatical level than for the lexical level. For qualitative interpretation, we inspect the linguistic items that significantly contribute to the observed trends looking at their (average) surprisal in syntagmatic context and the entropy of their paradigmatic context over time.
Degaetano-Ortlieb S. and E. Teich, 2019. Towards an optimal code for communication: the case of scientific English. Corpus Linguistics and Linguistic Theory (open access), DOI: doi.org/10.1515/cllt-2018-0088
De Smet, H. 2016. How gradual change progresses: The interaction between convention and innovation. Language Variation and Change, 28:83–102.
Harris, Z., 1991. A theory of language and information: A mathematical approach. Clarendon Press, Oxford. Weinreich U., W. Labov and M. I. Herzog, 1968.
Empirical foundations for a theory of language change. In W.P. Lehmann and Y. Malkiel (eds.), Directions for Historical Linguistics. University of Texas Press, Austin, Texas, pp. 95-195.
Natasha Korotkova (Universität Konstanz)
Title: Find, must, and conflicting evidence (joint work with Pranav Anand)
Abstract: Recent years have seen a lot of interest in the so-called subjective attitudes: English "find" and its counterparts in, e.g., French, German, or Norwegian. Unlike vanilla doxastics (i.e. "think"), find-verbs have been argued to only allow matters of opinion, rather than fact, in their complements. In this talk, we consider one underanalyzed class of expressions in find-complements: epistemic modals. Those modals have been often analyzed as subjective expressions, which makes them prime candidates for embedding under "find". However, epistemics are prohibited in subjective attitude complements. The "find+must" ban has been attributed to "must" not being subjective in the right way. We argue instead that the real culprit is a matter of evidence: find-verbs require their subject to have direct evidence for the complement, while "must" and its counterparts in other languages require a lack of direct evidence. Therefore, the "find+must" combination yields an evidential clash. We support our claim by novel cross-linguistic data on find-verbs and a range of indirect expressions, including bona fide evidentials, and analyze the "find+must" ban as a semantic contradiction.
Chundra Cathcart (Universität Zürich)
Title: Horizontal and vertical pressures in language change: fleshing out admixture models
Abstract: This talk presents preliminary results from a handful of studies using Bayesian admixture models to analyze cross-linguistic patterns with an eye to understanding shared history and historical contact between languages. Admixture models (such as the Structure algorithm from population genetics) have enjoyed less use in linguistics than phylogenetic methods; they have the advantage of directly modeling historical language contact, but do not explicitly model diachronic transitions between linguistic states. I focus on two extensions to this basic model which work towards bridging this gap: a neural model of sound change across Indo-Aryan dialects which accounts for both historical dialect group-level trends as well as individual language-level idiosyncrasies, and a model of typological distributions which models both spread and match factors by enhancing the admixture model with Markovian dynamics.
Johanna Nichols (University of California, Berkeley)
Title: Better characters/variables for historical linguistics
Abstract: Both wordlist-based and typology-based comparisons have become widely used in the last decade or so, and there has been much discussion of methods and interpretation. Here I discuss three problems I consider more fundamental: (1) The quality and usefulness of the characters or variables themselves; I propose several that promise much more as comparanda. (2) The coding strategies, especially the treatment of "no" answers on yes/no variables, can have major impact on distance measures and NeighborNet trees. (3) Handling synonymy in typologically-defined wordlist studies, where synonymy is common. Preliminary findings show that a combination of etymological and typological variables can yield good results for both typological and historical analysis.
Title: Error-driven Learning in Modeling Spoken Word Recognition
Abstract: Effective linguistic communication relies on the recognition of words (McQueen, 2007). Although spoken word recognition (SWR) is a vital task in speech comprehension, psycholinguists are still debating some fundamental assumptions decades after the first cognitive theories of SWR (Marslen-Wilson and Welsh, 1978) and initial computational modelings (McClelland and Elman, 1986; Norris, 1994). I present the current state of two projects in which we investigated the theory of error-driven learning, outlined by Rescorla and Wagner (1972) for animal and human learning, as a theory of SWR. Computational modelings of a excised word recognition task were carried out using the naive discriminative learning (NDL) and the linear discriminative learning (LDL) frameworks. First, Arnold et al. (2017) and Shafaei-Bajestan and Baayen (2018) applied NDL-based models that iteratively learn to classify German and English words, respectively, from their acoustic representations and reported model performance comparable to human performance. Second, Baayen et al. (2019) estimated the linear mappings between words’ feature vectors and the words’ semantic vectors directly and achieved superior accuracy in recognition of words compared to NDL. Assumptions in the models, issues in model implementation, initial results, and plans for future work are discussed.
Title: Linguistic politeness as strategic behavior: (some) costs and benefits of polite language use
Abstract: Behavioral ecology explains the behavior of animals by treating them as rational agents driven by a maximisation of their benefit-cost differentials. On the uncontroversial assumptions that humans are animals, and that speaking is a type of behavior, this overall approach should - at least in principle - generalise to language use: it should be possible to understand speakers as “rational decision-makers who make tradeoffs between costs and benefits”.
Despite the obvious difficulties involved in determining both the relevant types behavior (language use) and the relevant costs and benefits, there has been some success in applying this reasoning to “pockets” of linguistic behavior such as indirect speech (Pinker, Nowak & Lee 2006) or politeness (Quinley 2012). Most recently, the Responsibility Exchange Theory (RET; Chaudhry and Loewenstein 2019) effectively provides a proof of concept of this general approach for a narrow class of dyadic interactions (assignment of responsibility for a positive/negative outcome): it establishes functional classes of linguistic behavior (apologising, thanking) and works out a compelling theory of the associated social costs/benefits. In my talk, I build on the conceptual foundation proposed by Chaudhry and Loewenstein (2019) and look into ways of extending their approach.
Title: Factive presuppositions? An empirical challenge
Abstract: A long-standing and widely-held assumption is that the content of the complement of factive predicates like “know” is presupposed whereas that of non-factive predicates like “think” is not. There is, however, disagreement in the literature about which properties define factive predicates and whether the contents of the complements of particular predicates exhibit the properties attributed to factive predicates. The resulting disagreement about which predicates are factive is troublesome because the distinction between factive and non-factive predicates has played a central role in the study of presuppositions. This talk, which is based on joint work with Judith Degen (Stanford University), investigates properties of the contents of the complements of clause-embedding predicates with the goal of understanding how such predicates can be classified. We argue that predicates presumed factive are more heterogeneous than previously assumed and that there is little empirical support for the assumed categorical distinction between factive and non-factive predicates. We conclude by discussion the implications for future research on and analyses of presuppositions and other projective content.
Title: Why do we have to say certain things? On the obligatorification of dependents
Abstract: A common feature of the grammaticalization of function words is that they develop the requirement for obligatory dependents. For instance, English the does not occur except when followed by a nominal construction. This talk offers an account of the historical development of obligatorification - how dependents develop from optional extras to required accompaniments. I will show how the process leading to obligatorification is driven by universal communicative requirements. In this sense, the development in question sets in before grammaticalization “proper”, rather than being a result of grammaticalization as has been widely (if only implicitly) assumed. In fact, specific semantic structures create the need for overt hosts at every synchronic stage of every language. Only rarely does this requirement for an overt dependent develop into a syntactic requirement as a result of grammaticalization. I illustrate this with both diachronic and synchronic examples from diverse parts of speech that stem from several languages.
Title: Pronouns, Descriptions, Bridging
Abstract: Recent work on coreference and non-coreference anaphora makes no clear distinctions between different kinds of non-coreference anaphora and also seems to imply that coreference anaphora is a special kind of non-coreference anaphora. In this talk I take a closer look at the interpretation processes for anaphoric pronouns and definite descriptions from an interpretation-theoretic perspective. I identify three strategies for the interpretation of pronouns and descriptions. The first strategy always leads to coreference, the second can produce coreference effects and the third normally does not, although it may involve coreference in certain special cases. None of these three strategies can be reduced to either of the two others.
The formal framework in which the investigation is conducted is a version of DRT in which definite noun phrases (definite descriptions and pronouns among them) are treated as triggers of ‘identification presuppositions’ – presuppositions whose resolution identifies the referents of their triggers. The framework makes it possible to describe the strategies in precise and unambiguous terms and to ask precise questions about the ways they are related. A good part of the talk will be devoted to discussing this approach to the semantics and pragmatics of definite noun phrases and to showing how it works for anaphoric pronouns and descriptions.
Time permitting, we will have a look at a potential problem for the analysis: English descriptions with head nouns that denote ‘inalienable relations’, like the mouth, the father, the weight. Given the definition of bridging I will be using these descriptions ought to be perfect bridging descriptions. But in fact they are not. For instance, in normal contexts the sentence ‘No one mentioned the weight’ can’t be understood as meaning that no one mentioned his or her own weight. Nor can in most contexts: ‘Susan grew up like an orphan. She never even met the father’ be used felicitously to say that Susan never met her own father). The solution of this puzzle has to do with the different processes that are available for reinterpreting relational nouns as non-relational and non-relational nouns as relational.
Title: Linguistic diversity within Chibchan
Abstract: Chibchan languages are spoken at the very heart of the Americas, on the isthmus connecting both continents and in adjacent regions (Panama, Costa Rica, Colombia). Among the language families of Central and South America, the Chibchan family is particularly diverse in typological terms (Adelaar 2007). For instance, the Rama language of Nicaragua has only three phonemic vowels, /a/, /i/, /u/ (Craig 1989: 37), whereas Bribri (Costa Rica) has fourteen (Chevrier 2017: 56). In the domain of verbal person marking, some Chibchan languages use unbound elements, whereas others use prefixes, suffixes, or both (Pache 2015). This talk aims to discuss the following questions: (1) which are the domains of particular variability/relative uniformity within Chibchan? (2) What could have been factors triggering family-internal variability?
Title: Presuppositions, scales, and adjective order
Abstract: Many accounts of language assume that communication is inherently Gricean, and thus that contextually enriched meanings depend in part on a sensitivity to speaker states. However, current models of how core semantic phenomena interact with context often ignore speaker-specific information. For example, in the case of gradable adjectives, research on this topic has mainly focused on the role of extra-linguistic context, such as the distribution of a feature across a domain, or informativeness.
Here, we ask whether, in addition to statistical distributions, listener’s standards of comparison for adjectives are also sensitive to thresholds communicated by (a) existential presuppositions, to investigate whether listeners accommodate individual differences, and (b) different adjective orderings, to test whether the compositional operations involved in understanding AAN-sequences can help comprehenders decide whether scalar thresholds are affected by the speaker’s statements. The results are jointly informative for recent discussions of scalar vs. absolute adjectives, the question of how scalar thresholds are computed, and the compositional semantics of multi-adjective sequences.
Hizniye Isabella Boga
Title: The languages of Italy - Measuring the similarity between close and distant varieties
Abstract: One of the oldest questions in dialectology is how to define a “language” as opposed to a “dialect” (Gooskens 2018). The theoretical definition of a language as the standardised form and dialects as sub-categorical varieties of “inferior” character have been assumed for a very long time. Only with J. K. Chambers and Peter Trudgill’s introduction of the definition “language as a collection of mutually intelligible dialects”, an equality of varieties was emphasised.
The task of my thesis research revolved around measuring distances and similarities of 58 Romance varieties with a focus on the Italian varieties. The goal is to determine which varieties are closer to each other and can hence be seen as dialects of the same language or whether they are distant enough to be considered independent languages.
The methods at hand are the Levenshtein Distance Normalized Divided (LDND) and the Needleman-Wunsch algorithm Normalized Divided (NWND) with a built-in scorer system of PMI distances. With the resulting distances and similarities determined by the LDND and the NWND method, I used a model-based clustering method to allocate similar varieties into one cluster, dissimilar varieties into another cluster and varieties of mixed and unclear affiliation into a further one. Within those clusters, it is visible which varieties are close enough to be varieties of the same language and which varieties are distant enough to be independent languages.
Jakub Szymanik (University of Amsterdam)
Title: Ease of learning explains semantic universals
Abstract: Despite extraordinary differences between natural languages, linguists have identified many semantic universals – shared properties of meaning – that are yet to receive a unified explanation. We analyze universals in a domain of content words (color terms) and a domain of function words (quantifiers). Using tools from machine learning, we show that meanings satisfying attested universals are easier to learn than those that are not. Thus, ease of learning can explain the presence of semantic universals in many different linguistic domains.
Ramon Ferrer i Cancho (Universitat Politècnica de Catalunya)
Title: An emerging theory of word order
Word order is a fascinating phenomenon. During decades, researchers have been collecting many word order regularities that have fed theory. Some of these regularities are the Greenbergian universals of word order, consistent branching or the low number of dependency crossings in the syntactic dependency structures of sentences. Here we will argue these regularities can be regarded as adaptations to the limited resources of the human brain with the help of an emergent theory of word order that provides a unified explanation to word variation and word order change. We will discuss the negative consequences of denying or neglecting the role of functional pressures for the construction of a parsimonious theory of language.
An apetizer: here
Torgrim Solstad (ZAS Berlin)
Title: Predictive Language Processing: The View from Implicit Causality
Prediction in language (Kamide 2008; DeLong et al. 2014), whereby we understand the incorporation of possible (and likely) future information states into processing, still hasn't attracted much attention in theoretical linguistics despite the central status of prediction in human cognition (Bubic et al. 2010; Clark 2013). Bringing together insights from experimental and theoretical research for one particular phenomenon, Implicit Causality, I want to argue that much could be gained by attempting to bridge this gap.
Implicit Causality verbs (e.g. Garvey/Caramazza 1974; Brown/Fish 1983) have been at the core of psycholinguistic research on predictive processing. Selecting for two animate arguments, such verbs display a strong preference for an explanation focusing on one argument, as shown in numerous sentence continuation experiments:
(1) JOHN annoyed Mary because... HE was rude.
(2) John admired MARY because... SHE was clever.
Although there is good evidence as to the processing profile of Implicit Causality, its predictive nature is still insufficiently understood. Some important questions include:
- What is predicted: Is it a particular word (e.g. HE/SHE in (1)/(2)), a referent, or a type of explanation (e.g. a property of John's in (1), and one of Mary's in (2))?
- What triggers the prediction: Is it encoded in lexical semantics (annoy/admire in (1)/(2)) or part of world knowledge?
Based on a formal-semantic theory of Implicit Causality (Bott/Solstad 2014), results from previous experimental research (e.g. Koornneef/van Berkum 2006; Pykkönen/Järvikivi 2010; Featherstone/Sturt 2010) and recent insights into the nature of predictive processing in general (e.g. Kuperberg/Jaeger 2016; Yan et al. 2017), I will propose a framework for predictive processing of Implicit Causality. By bringing together insights from theoretical and experimental research, we can delineate more precisely the top-down and bottom-up processes generating and validating predictions: Which linguistic levels are involved and how do they interact?
Although limited to one particular phenomenon, I contend that approaching predictions in this manner has the potential to mutually benefit both psycholinguistics and theoretical linguistics. On the one hand, a number of aspects concerning prediction processes may be better understood if they are based on more elaborate theoretical linguistic analysis, if only for constraining the possible hypothesis space, thus allowing for more precise experimental predictions and better control of experimental design and materials. On the other hand, research on prediction extends an invitation to reconsider or expand theoretical linguistic assumptions to accommodate the results obtained in experimental research, potentially offering a broader empirical base for linguistic studies, connecting phenomena previously assumed to be unrelated.
Susanne Dietrich (Tübingen)
Title: Processing of presuppositions during speech perception: a functional magnetic resonance imaging (fMRI) study
Abstract: Discourse structure enables us to generate expectations based upon linguistic materials that has already been introduced. The present functional magnetic resonance imaging (fMRI) study addresses auditory perception of test-sentences in which discourse coherence was manipulated by using presuppositions (PSP) that either correspond or fail to correspond to items in preceding context-sentences. Thereby, in- and definite determiners referring to either (non-) uniqueness or (not) existence of an item were used as PSP triggers. Discourse violation within the (non-) uniqueness subset yielded hemodynamic activation within the pre-supplementary motor area (pre-SMA) and bilateral inferior frontal gyrus (IFG). Considering the existence subset, these regions occurred only, if subjects accommodated the discourse. These findings indicate involvement of (i) the working memory (IFG) referring the PSP to contextual information and (ii) a regulator (pre-SMA) managing the process of comprehension by signaling detected errors to the system. This enables the system to continue the process of comprehension, for example, by up-dating the context or tolerating slight errors.
Shirley-Ann Rueschemeyer (York)
Title: Perspective taking during language comprehension
Abstract: Humans are constantly engaged in social interactions, and many of these interactions are supported by language. In this talk I will be presenting a series of studies investigating how language and social cognitive mechanisms interact in order to facilitate communication. I will start by showing that embodied lexical-semantic representations are activated by words in a flexible manner that reflects both linguistic and pragmatic constraints. Secondly, I will show the results of studies that suggest that when pragmatic constraints affect semantic processing, this is supported by interactions between neural language and mentalizing systems. Lastly, I will suggest that language comprehension is affected by assumptions we hold about other co-listeners as well as speakers. One key mechanism supporting perspective taking between co-listeners may be simulation. Together the studies presented in this talk provide insight into how high level language and social cognitive processes work in concert during successful communicative acts.