Matthijs Westera

A free, thorough introduction to Python, aimed at ambitious beginners. 115 pages of focused, incremental exercises, organized into a coherent learning trajectory. Balances clear instructions with learning-by-doing (and learning by breaking things), for step-by-step, guided discovery. Teaches practical skills paired with thorough understanding. The exercises workbook is accompanied by a book of 'desserts', more applied Python projects that match the skills you develop through the main workbook. Have a look! I wrote these books teaching Python to undergraduate students. I found existing books and websites to be either too heavy on prose, or their exercises lacking the repetition required to build lasting skills and the cohesion required to develop understanding.

Manual transcription of intonation by experts remains an essential part of research on the structure and meaning of intonation across languages, as well as for developing computational methods for automatic intonation transcription. We present ToneSwiper, a Python program with a graphical user interface that facilitates manual intonation transcription in the ToDI framework (Transcription of Dutch Intonation; Gussenhoven, 2005), with possible adaptation to similar (e.g., ToBI-like) frameworks for other languages. For the trained annotator, it enables efficient ToDI transcription of speech by integrating an audio-player, a spectrogram and pitch contour plot, auto-scroll, dynamic audio stretching, and an intuitive hotkey interface that maps key sequences to ToDI elements, e.g., pressing up-down for a high-to-low accent (H*L). In this way, transcription is conducted by ‘swiping‘ over the arrow keys on the keyboard. We present the program and its motivation, as well as a small-scale pilot study on annotation efficiency and inter-rater agreement, using a highly challenging sample of task-oriented discourse from the Dutch Map Task Corpus (Lickley et al., 2005).

Discourses can contain stereotypes not just in the form of statements, but also in the form of questions (e.g., Beukeboom et al. 2023): for instance, when will you have kids? vs. when will you get a raise?. Whether a given question exhibits a stereotype can depend on attributes of the addressee (e.g., their gender), and, more generally, on what it was, in the prior discourse or broader context, that triggered or 'evoked' the question. The present work investigates to what extent we can predict the kinds of questions that a given discourse will evoke. To that end, we look at (reasonably) naturally occurring questions (Reddit data, including the 'Ask Me Anything' format), questions elicited (by TED-talks) through crowdsourcing (the TED-Q dataset; Westera et al. 2020), and questions elicited from large language models, in particular base models (which, unlike the more popular instruct models like Chat-GPT, have not been additionally finetuned to avoid stereotypes, thus better reflecting actual bias inherent in the raw training data). For the latter, we test models' alignment to human data, and categorize and quantify generated response to specifically stereotype-evoking prompts.

It has become fairly commonplace in linguistics to use instruct-tuned LLMs as a virtual assistant/annotator/rater, with prompts containing explicit task instructions. This taps into their accumulated statistics from the (increasingly complex and opaque) training regime about how to follow (similar) instructions. Language models used in this manner are capable of displaying behaviors that highly correlate with that of human annotators/raters. Alternatively, one can try to use LLMs, especially base (as opposed to instruction-tuned) models, more like naive participants in a psycholinguistic experiment, by passing in stimuli and observing a model’s response without (much) explicit instruction. This taps into their accumulated statistics about patterns in language in general, not specific to an ‘instruction following’ context. The aforementioned distinction somewhat resembles that between, on the human side, explicit knowledge and tacit knowledge. Some works have shown that the ‘tacit knowledge’ of LLMs, specifically their word probabilities, can provide a better fit to human data than LLM responses obtained from explicit instructions, for instance with regard to reading times and plausibility judgments. Unsurprisingly, instruct-tuning an LLM, while making it more suited for instruction following, can make it behave less like an ordinary human language user.

Discussions about the relation between education and Generative AI (mainly large language models) typically devolves into four problematic frames: 1. how to prevent AI cheating; 2. which kinds of AI to allow; 3. how best to teach students how to use AI; 4. how to develop personalized AI tutors. This presentation aims to refute each of them, presenting an alternative centered on trust, respect, safety, and the separation of formative (teaching) and summative (grading) aspects of education. The central argument is that we should make our (summative) assessment AI proof as soon as possible, so that we can focus on rebuilding trust in line with our didactic mission.

While the experimental turn of the last decade has helped connect formal semantic/pragmatic theories to empirical reality, extending this empirical reach to naturally occurring data has lagged behind. One of the challenges is the lack of large corpora in which the relevant technical notions have been annotated. Nevertheless, some insight may be gained by relying on automatically computed approximations of these notions. We take such an approach to focus realization– the relation between focus, grammar and pitch accents, taking inspiration in particular from Unalternative Semantics.

The questions we entertain can affect how we process incoming information (Grossnickle, 2016). We are more likely to accept information that answers our previous questions, a bias known as need-for-closure (Webster & Kruglanski, 1994). Despite this, the effects of questions in social media on users' adoption of (dis)information have hardly been studied. Our work aims to detect the need-for-closure bias in a newly collected corpus of hundreds of thousands of Reddit posts. Do social media users more readily accept information that answers a question they were previously exposed to?

The start of 2023 witnessed a disruptive development in Conversational AI: ChatGPT. Large language model technology suddenly became available to millions of users. The underlying GPT-3.5 language model, with 175 billion parameters, trained on 300 billion words and finetuned with human feedback, displayed baffling fluidity, style transfer, and emergent behavior like chain-of-thought reasoning. Moreover, its context window of thousands of tokens enabled a form of conversational training: on-the-fly supervised (albeit volatile) training through prompting. From a conversational perspective, ChatGPT has session-spanning conversational memory, enabling it to pick up on previous interactions in a dialogue. In March 2023, GPT-3.5 was succeeded by GPT-4, with a larger context window, reportedly better accuracy in handling factual questions, and connecting image analysis to language model-based communicative interaction. Given these significant developments, one could be tempted to think that Conversational AI has come of age. Yet, a full slate of unresolved problems and research questions remains. Crucial debates surround the societal impact of large language models and the future of NLP, the environmental impact of training regimes as well as mass adoption, the impact and prevention of bias, and possible copyright infringement of training data. Central Research Topics in the field of Conversational AI are to a large extent orthogonal to the underlying technology, including large language models. This Research Topic of Frontiers addresses a number of such topics: the human perception of conversational agents and the effects of social cues exhibited by conversational agents on humans, the role of information presentation in hybrid conversational systems, the usage of carefully annotated data in addition to raw textual observational data, and the emergence of communicative patterns between humans and machines.

Hurford disjunctions are disjunctions in which one disjunct entails another. Some of these are perfectly natural while others seem infelicitous, at least out of context. The predominant approach to this phenomenon relies on Hurford’s Constraint, which states that such disjunctions are generally bad, together with the grammatical approach to exhaustivity, according to which, in the felicitous cases, local exhaustification of the disjuncts is available to break the inter-disjunct entailment. An alternative, inverse approach has not been seriously explored: to take the felicitous cases as basic, refrain from adopting HC, and try to rule out the infelicitous cases by other means. This paper develops this inverse approach, and offers a thorough theoretical-conceptual comparison of both approaches, touching on central topics such as the pragmatics-grammar debate surrounding exhaustivity, the status of Hurford’s Constraint as a derivative of considerations of redundancy, which semantics for disjunction to adopt, and constraints on sets of alternatives.

The notion of ’alternative’ is central to analyses of various semantic/pragmatic phenomena, such as disjunction, focus, discourse structure, questions, and implicature. However, basic questions concerning the various notions of alternatives have not received the attention they deserve, e.g. what exactly these notions signify, or how they are supposed to interact. This chapter reflects on such questions, centering on appeals to alternatives in characterizations of focus, disjunction, discourse goals (questions under discussion), and interrogatives. More precisely, this chapter criticizes the conflation of the set of focus alternatives with the meaning of an interrogative, discusses two conceptions of the alternatives introduced by disjunction (algebraic and attention-based), and departs from the predominant view of QUDs as, essentially, linguistic questions that represent discourse goals.

The neo-Gricean approach to exhaustivity is based on the idea that exhaustivity arises when relevant propositions are not asserted. This paper presents a new pragmatic approach based on the idea that exhaustivity arises when relevant propositions are not mentioned, or more precisely, when the speaker did not intend to draw attention to them. This seemingly subtle shift from information to attention results in different predictions on a range of challenges for the neo-Gricean approach, some of which have been brought up in support of the grammatical approach to exhaustivity. This paper discusses three such challenges: exhaustivity on the hints of a quizmaster, exhaustivity on questions, and exhaustivity without an opinionatedness assumption. The two pragmatic approaches are compared on these puzzles along with the grammatical approach.

Cognitive scientists have long used distributional semantic representations of categories. The predominant approach uses distributional representations of category-denoting nouns, such as “city” for the category city. We propose a novel scheme that represents categories as prototypes over representations of names of its members, such as “Barcelona,” “Mumbai,” and “Wuhan” for the category city. This name-based representation empirically outperforms the noun-based representation on two experiments (modeling human judgments of category relatedness and predicting category membership) with particular improvements for ambiguous nouns. We discuss the model complexity of both classes of models and argue that the name-based model has superior explanatory potential with regard to concept acquisition.

Theories of intonational meaning can be organized into two broad categories. Specialist theories aim to capture the meaning of a particular type of intonation contour, or even just a particular usage of that contour, typically using tools from formal semantics. By contrast, generalist theories aim to capture the meanings of a broader range of contours, typically by assigning more basic, underspecified meanings to a larger set of prosodic morphemes. Both strands have yielded important insights, but neither is entirely satisfactory: specialist theories have limited empirical scope and explanatory potential, and generalist theories have not readily yielded concrete, testable predictions from their basic meanings. In recent years, following developments in formal pragmatics, partial but promising attempts have been made to combine the strengths of both. With this goal as a focal point, the current chapter provides an overview of theoretical and empirical work on intonational meaning.

We take a closer look at van Tiel et al.’s (2016) experimental results on diversity in scalar inference rates. In contrast to their finding that semantic similarity had no significant effect on scalar inference rates, we show that a sufficiently fine-grained notion of semantic similarity does have an effect: the more similar the two terms on a scale, the lower the scalar inference rate. Moreover, we show that a context-sensitive notion of semantic similarity (in particular ELMo; Peters et al., 2018) can explain more of the variance in the data, but only modestly, only for stimuli that contain informative context words, and only when the scalar terms themselves are sufficiently context-sensitive.

Both disjunctive assertions and disjunctive questions can imply “not both”, i.e., that only one of the disjuncts is true. For assertions this is known to be part of what the speaker means (e.g., an implicature), whereas for questions this is instead a presupposition. This puzzle is challenging for predominant pragmatic and grammatical accounts of exhaustivity in the literature. This paper outlines a solution based on Attentional Pragmatics combined with (other) general pragmatic principles.

Hurford disjunctions are disjunctions where one disjunct entails another. Some of these are perfectly natural while others seem infelicitous, at least out of context. The predominant approach to this phenomenon relies on Hurford’s Constraint, which states that such disjunctions are generally bad, together with grammatical exhaustification, which can rescue some of them by exhaustifying the weaker disjunct to break the entailment. An alternative, pragmatic approach to Hurford disjunctions relies on neither Hurford’s Constraint nor grammatical exhaustification, but it has received much less attention. This paper offers a comprehensive overview and comparison of both approaches. It touches on central topics such as the granularity of one’s semantics, the status of Hurford’s Constraint as a derivative of considerations of redundancy, constraints on relevance and questions under discussion, levels of categorization, and ways in which pragmatic principles can operate in embedded contexts.

We present a new dataset of TED-talks annotated with the questions they evoke and, where available, the answers to these questions. Evoked questions represent a hitherto mostly unexplored type of linguistic data, which promises to open up important new lines of research, especially related to the Question Under Discussion (QUD)-based approach to discourse structure. In this paper we introduce the method and open the first installment of our data to the public. We summarize and explore the current dataset, illustrate its potential by providing new evidence for the relation between predictability and implicitness – capitalizing on the already existing PDTB-style annotations for the texts we use – and outline its potential for future research. The dataset should be of interest, at its current scale, to researchers on formal and experimental pragmatics, discourse coherence, information structure, discourse expectations and processing. Our data-gathering procedure is designed to scale up, relying on crowdsourcing by non-expert annotators, with its utility for Natural Language Processing in mind (e.g., dialogue systems, conversational question answering).

We release ManyNames v2 (MN v2), a verified version of an object naming dataset that contains dozens of valid names per object for 25K images. We analyze issues in the data collection method originally employed, standard in Language & Vision (L&V), and find that the main source of noise in the data comes from simulating a naming context solely from an image with a target object marked with a bounding box, which causes subjects to sometimes disagree regarding which object is the target. We also find that both the degree of this uncertainty in the original data and the amount of true naming variation in MN v2 differs substantially across object domains. We use MN v2 to analyze a popular L&V model and demonstrate its effectiveness on the task of object naming. However, our fine-grained analysis reveals that what appears to be human-like model behavior is not stable across domains, e.g., the model confuses people and clothing objects much more frequently than humans do. We also find that standard evaluations underestimate the actual effectiveness of the naming model: on the single-label names of the original dataset (Visual Genome), it obtains −27% accuracy points than on MN v2, that includes all valid object names.

We take a close look at a recent dataset of TED-talks annotated with the questions they implicitly evoke, TED-Q (Westera et al., 2020). We test to what extent the relation between a discourse and the questions it evokes is merely one of similarity or association, as opposed to deeper semantic/pragmatic interpretation. We do so by turning the TED-Q dataset into a binary classification task, constructing an analogous task from explicit questions we extract from the BookCorpus (Zhu et al., 2015), and fitting a BERT-based classifier alongside models based on different notions of similarity. The BERT-based classifier, achieving close to human performance, outperforms all similarity-based models, suggesting that there is more to identifying true evoked questions than plain similarity.

This work applies the well-known BERT model to a selection of part-of-speech tagged, dependency-parsed and coreference-annotated text, extracting gradients and attention weights for inspection. This reveals that, in BERT, more information flows from a noun to a pronoun if they corefer; open-class words are generally more informative than closed-class words; and there is a slightly underwhelming correlation between BERT’s gradients and dependency parses. It also highlights that attention weights and gradients are of course correlated, but they do not always reveal exactly the same patterns.

Humans use language to refer to entities in the external world. Motivated by this, in recent years several models that incorporate a bias towards learning entity representations have been proposed. Such entity-centric models have shown empirical success, but we still know little about why. In this paper we analyze the behavior of two recently proposed entity-centric models in a referential task, Entity Linking in Multi-party Dialogue (SemEval 2018 Task 4). We show that these models outperform the state of the art on this task, and that they do better on lower frequency entities than a counterpart model that is not entity-centric, with the same model size. We argue that making models entity-centric naturally fosters good architectural decisions. However, we also show that these models do not really build entity representations and that they make poor use of linguistic context. These negative results underscore the need for model analysis, to test whether the motivations for particular architectures are borne out in how models behave when deployed.

In the literature, English rise-fall-rise (RFR) intonation is known both as a marker of secondary information and as a marker of topics. This paper aims to make plausible that these two uses can be derived from a common core, which in turn can be derived from a recent theory of intonational meaning more generally, according to which rises and falls indicate (non-)compliance with the maxims (Westera 2013, 2014, 2017). The core meaning of RFR, I propose, is that the main question under discussion (Qud) is not compliantly addressed, while some secondary Qud is. Several more concrete predictions are derived from this core meaning, pertaining to secondary information, topic marking, exhaustivity, and discourse strategies. The resulting account is shown to generate certain ingredients of existing accounts, while also doing some things differently in ways that may be empirically accurate. If the proposed account is on the right track, it provides an important new intonational window on QUDs.

Distributional semantics has had enormous empirical success in Computational Linguistics and Cognitive Science in modeling various semantic phenomena, such as semantic similarity, and distributional models are widely used in state-of-the-art Natural Language Processing systems. However, the theoretical status of distributional semantics within a broader theory of language and cognition is still unclear: What does distributional semantics model? Can it be, on its own, a fully adequate model of the meanings of linguistic expressions? The standard answer is that distributional semantics is not fully adequate in this regard, because it falls short on some of the central aspects of formal semantic approaches: truth conditions, entailment, reference, and certain aspects of compositionality. We argue that this standard answer rests on a misconception: These aspects do not belong in a theory of expression meaning, they are instead aspects of speaker meaning, i.e., communicative intentions in a particular context. In a slogan: words do not refer, speakers do. Clearing this up enables us to argue that distributional semantics on its own is an adequate model of expression meaning. Our proposal sheds light on the role of distributional semantics in a broader theory of language and cognition, its relationship to formal semantics, and its place in computational models.

We introduce a novel, scalable method aimed at annotating potential and actual Questions Under Discussion (QUDs) in naturalistic discourse. It consists of asking naive participants first what questions a certain portion of the discourse evokes for them and subsequently which of those end up being answered as the discourse proceeds. This paper outlines the method and design decisions that went into it and on characterizing high-level properties of the resulting data. We highlight ways in which the data gathered via our method could inform our understanding of QUD-driven phenomena and QUD models themselves. We also provide access to a visualization tool for viewing the evoked questions we gathered using this method (N=4765 from 111 crowdsourced annotators).

Central to explaining many linguistic phenomena is an understanding of what the goals of the given discourse are. This is made difficult however by the fact that goals are often left implicit in discourse. Much theoretical work in semantics and pragmatics assumes that discourse goals can be identified with implicit or explicit questions, or Questions Under Discussion (QUD; e.g., Ginzburg 1996; Roberts 1996). Semantic/pragmatic theories typically yield strong, falsifiable predictions given a certain QUD, but no comprehensive theory exists of what that QUD should be for any given piece of discourse. This limits the testability of these theories in practice, and it stands in the way of a proper understanding of results from experimental linguistics, where participants’ judgments are due in part to their understandings of the implicit goals underlying the linguistic stimuli (e.g., Schwarz 1996; Westera and Brasoveanu 2014). I propose to employ language models to help overcome this challenge, by using them to generate (or compute the probability of) a plausible QUD based on a discourse. To my awareness no quantitative, data-driven model of QUDs like this has been attempted. This is work in progress, and besides hoping to demonstrate the promise of this kind of approach and obtaining feedback, I foremost wish to draw attention to this important open issue for QUD-based theories, and the need for a tighter integration with computational modeling.

The theory of Intonational Compliance Marking (ICM) maintains that speakers of English use final rising intonation to indicate a suspension (potential violation) of a conversational maxim (Westera 2013; 2014). This paper aims to show that a certain kind of rising declarative, one which has been prominent in the literature (e.g., Gunlogson 2008), can be adequately understood in ICM’s terms as involving a suspension of the maxim of Quality. By explicating certain minimal assumptions about pragmatics, this understanding accounts for three core features of such rising declaratives: their question-likeness, the speaker bias they express and their badness out of the blue. In a nutshell, their question-likeness is derived from principles of general cooperative discourse, their bias from the relative importance of the maxim of Quality, and their badness out of the blue from a competition between rising declaratives and interrogatives. The account is compared in detail to various existing accounts of rising declaratives of the relevant sort, highlighting explanatory and empirical differences.

Exhaustivity is typically explained in terms of the exclusion of unmentioned alternatives. For this to work, the set of alternatives must be asymmetrical, lest both a proposition and its negation get excluded, yielding a contradiction (the Symmetry Problem). Since exhaustivity is regularly observed, these alternative sets must tend to be asymmetrical, and this requires an explanation. Existing explanations are based on considerations of brevity, but these run into certain problems. A new solution is proposed, explaining the asymmetry of alternatives in terms of the fact that discourse strategies with asymmetrical questions under discussion (Quds) are favored because they allow part of the answer to be communicated implicitly, namely as an exhaustivity implicature.

This dissertation presents a precise, unified and explanatory theory of human conversation, centered on two broad phenomena: exhaustivity implications and intonational meaning. In a nutshell: (i) speakers have two types of communicative intentions, namely information sharing and attention sharing, (ii) these types of intentions ideally comply with a certain set of rationality criteria, or maxims, (iii) speakers of English and related languages use intonation, in particular socalled trailing tones and boundary tones, to indicate whether such compliance is achieved, and (iv) exhaustivity implications arise when this holds, at least, for the attention-sharing intention. The research presented here goes against a number of widespread assumptions in the field. The result is a perspective on conversation that enables new solutions to a broad range of well-known puzzles surrounding exhaustivity and intonation. Among these are the “symmetry problem”, the “epistemic step” without a competence assumption, the role of informationally redundant disjuncts, the bias expressed by rising declaratives, the range of uses of rise-fall-rise intonation, the effects of different intonation contours in lists, and differences between questions with rising and falling intonation.

It is commonly assumed that final rises/falls (H%/L%) in English indicate whether the speaker thinks that the utterance is pragmatically complete (Pierrehumbert and Hirschberg 1990; Westera 2013; Malamud and Stephenson 2015, among many). We follow Hobbs 1990 in treating rising/falling accents (L*H/H*L) analogously. Since Rise-Fall-Rise (RFR) is a falling accent (H*L) plus a rise (H%), this predicts that an utterance with RFR must be pragmatically complete in one way and incomplete in another. (The falling accent of RFR is in fact delayed, but we remain agnostic about the effect of delay; for a compatible proposal see Gussenhoven 1983.) We make this prediction more precise in terms of questions under discussion (quds) and show that it gives us a unifying understanding of RFR, and in particular of the relation between non-at-issue meaning and (utterance-initial) topics.

We argue for a purely pragmatic account of the ignorance inferences associated with superlative but not comparative modifiers (at least vs. more than). Ignorance inferences for both modifiers are triggered when the question under discussion (QUD) requires an exact answer, but when these modifiers are used out of the blue the QUD is implicitly reconstructed based on the way these modifiers are typically used, and on the fact that at least n, but not more than n, mentions and does not exclude the lower bound exactly n. The paper presents new experimental evidence for the context-sensitivity of ignorance inferences, and also for the hypothesis that the higher processing cost reported in the literature for superlative modifiers is context-dependent in the exact same way.

A compositional theory of English intonational meaning is presented that is derived from Gussenhoven’s biological codes. The resulting theory is compared to the more top-down approaches to focus and contrastive topic in the literature, suggesting how such stipulated semantic/pragmatic notions as ‘alternative’ and ‘strategy’ can be grounded – and revealing how they might be amended.

Thirty years after Groenendijk and Stokhof’s (1984) dissertation, the exhaustive interpretation of answers is still one of the central topics in semantics and pragmatics. Groenendijk and Stokhof identified three main problems for a pragmatic account of exhaustivity, which to this date remain largely open. In the present paper I show how these can be resolved by adopting a richer notion of meaning, and taking into account its pragmatic thrust. The resulting theory may be the only one to this date that explains exhaustivity, from start to end, as a genuine case of Gricean conversational implicature.

I show that the exhaustive interpretation of answers can be explained as a conversational implicature through the Maxim of Relation, dealing with the problematic epistemic step (Sauerland, 2004). I assume a fairly standard Maxim of Relation, that captures the same intuition as Roberts’ (1996) contextual entailment. I show that if a richer notion of meaning is adopted, in particular that of attentive semantics (Roelofsen, 2011), this Maxim of Relation automatically becomes strong enough to enable exhaustivity implicatures. The results suggest that pragmatic reasoning is sensitive not only to the information an utterance provides, but also to the possibilities it draws attention to. Foremost, it shows that exhaustivity implicatures can be genuine conversational implicatures.

Declarative sentences that end with a rising pitch in English (among other languages) have many uses. I single out several prominent uses that the literature so far has treated mostly independently. I present a compositional, unifying analysis, where the final rising pitch marks the violation of a conversational maxim, and its steepness indicates the speaker’s emotional activation. Existing theories are reproduced from these basic assumptions. I believe it contributes to a solid theoretical foundation for future work on the semantics and pragmatics of intonation.

I present a unifying solution to two well-known empirical puzzles: (i) how to account for the exhaustive interpretation of answers, and (ii) how to account for the semantics of the final rise in American English. It relies on the hypotheses that pragmatic reasoning is sensitive to the possibilities that a sentence draws attention to and that the final rise conveys the speaker’s uncertain cooperativity. The take-home message is that the Gricean approach to exhaustivity is viable, provided that we enrich the underlying semantics with attentive content.

A disjunction may pragmatically imply that only one of the disjuncts is true. The traditional Gricean account of this exhaustivity implicature is not without problems. Nevertheless, we think that not the Gricean picture itself, but the underlying conception of meanings as chunks of information may be unfit. Starting instead from a conception of meanings as proposals, within the framework of inquisitive semantics, we develop, algebraically characterise and conceptually motivate a formal semantics and pragmatics, the latter still Gricean in spirit. Among the difficulties we discuss and resolve are the problem of characterising relevant alternatives, the problem of embedded implicatures and the unwanted negation problem. The analysis is extended to a pragmatic account of mention-some questions.

Within the framework of inquisitive semantics, we investigate the semantic prerequisites of an account of discourse coherence. In inquisitive semantics two views on meaning exist. Basic inquisitive semantics, InqB, follows from the view that to utter a sentence is to provide and request information (Roelofsen, 2011). Unrestricted inquisitive semantics, InqU, follows from the view that to utter a sentence is to propose to update the common ground in any of several ways (Ciardelli, Groenendijk, & Roelofsen, 2009). We illustrate with a simple example that InqU, but not InqB, can be a semantic foundation for an account of discourse coherence. However, the clauses of InqU have not been motivated conceptually with as much rigour as those of InqB, and they are technically not as well understood. In this paper we precise its conceptual motivation and, based on this conception of meaning, define a semantics driven by general algebraic concerns. We show that the algebraic backbone of InqU is a commutative, idempotent semiring, which will facilitate an integration of inquisitive semantics with other formalisms. The algebraic structure gives rise to a compliance order on meanings that we put forward as a core notion for an account of discourse coherence.

This paper addresses an important open question in cognitive science: what is it that makes us human? Two uniquely human capacities will be discussed. The first is tool-use, which we will discuss initially from the perspective of Clark and Chalmers’ Extended Mind Hypothesis (EMH). Some problematic aspects of the EMH are highlighted and a view on tool-use based on Dennett’s homuncular functionalism is introduced as an alternative. The second capacity is language, which we will discuss from a memetic viewpoint. We formulate the Aggregate Mind Hypothesis, stating that the human mind is not identical to the software implemented on the brain, but rather to an aggregate mind of the brain and the vehicles of language.

In de literatuur van Van Gunsteren1,2 en Van Ruyven1 wordt ‘De Ongekende Samenleving’ naar voren gebracht als oorzaak van veel problemen voor bestuurders. Ik beargumenteer in dit essay dat het concept DOS gezien kan worden als een instantie van het algemenere concept dat ik ‘De Ongekende Werkelijkheid’ noem. Ik introduceer eerst het concept DOS, waarna ik inga op mogelijke oplossingen voor de problemen gerelateerd aan DOS zoals voorgesteld door Groenewegen3 en Lavis4 (raamwerken voor kennisuitwisseling), Van Gunsteren en Van Ruyven (selectieve besturingssystemen) en Bakker5 (scenariostudies). Daarna introduceer ik DOW als een tot inkeer komen van het gedachtegoed van de Verlichting en als een veel algemener verschijnsel waarvan DOS een instantie is. Ik laat zien dat DOS veel overeenkomsten vertoont met het ‘frameprobleem’, een instantie van DOW binnen de artificiële intelligentie, en relateer de oplossingen voor beide problemen aan elkaar. Daarna zal ik een conceptuele opdeling van DOW voorstellen op grond van veranderlijkheid en mate van onkenbaarheid.