Search

We propose a new model for forming and revising beliefs about unknown probabilities. To go beyond what is known with certainty and represent the agent’s beliefs about probability, we consider a plausibility map, associating to each possible distribution a plausibility ranking. Beliefs are defined as in Belief Revision Theory, in terms of truth in the most plausible worlds (or more generally, truth in all the worlds that are plausible enough). We consider two forms of conditioning or belief update, corresponding to the acquisition of two types of information: (1) learning observable evidence obtained by repeated sampling from the unknown distribution; and (2) learning higher-order information about the distribution. The first changes only the plausibility map (via a ‘plausibilistic’ version of Bayes’ Rule), but leaves the given set of possible distributions essentially unchanged; the second rules out some distributions, thus shrinking the set of possibilities, without changing their plausibility ordering.. We look at stability of beliefs under either of these types of learning, defining two related notions (safe belief and statistical knowledge), as well as a measure of the verisimilitude of a given plausibility model. We prove a number of convergence results, showing how our agent’s beliefs track the true probability after repeated sampling, and how she eventually gains in a sense (statistical) knowledge of that true probability. Finally, we sketch the contours of a dynamic doxastic logic for statistical learning.

In this article we introduce the concept of implied optical perspective in deep learning computer vision systems. Taking the BBC's experimental television programme “Made by Machine: When AI met the Archive” (2018) as a case study, we trace a conceptual and material link between the system used to automatically “watch” the television archive and a specific type of photographic practice. From a computational aesthetics perspective, we show how deep learning machine vision relies on photography, its technical regimes and epistemic advantages, and we propose a novel way to identify the latent camera through which the BBC archive was seen by machine.

This article shows how a machine can employ a network view to reason about complex social relations of news reliability. Such a network view promises a topic-agnostic perspective that can be a useful hint on reliability trends and their heterogeneous assumptions. In our analysis, we depart from the ever-growing numbers of papers trying to find machine learning algorithms to predict the reliability of news and focus instead on using machine reasoning to understand the structure of news networks by comparing it with our human judgements. Understanding and representing news networks is not easy, not only because they can be extremely vast but also because they are shaped by several overlapping network dynamics. We present a machine learning approach to analyse what constitutes reliable news from the view of a network. Our aim is to machine-read a network’s understanding of news reliability. To analyse real-life news sites, we used the Décodex dataset to train machine learning models from the structure of the underlying network. We then employ the models to draw conclusions how the Décodex evaluators came to assess the reliability of news.

Efficiently exploiting all sources of information such as labeled instances, classes’ representation, and relations of them has a high impact on the performance of Multi-Label Text Classification (MLTC) systems. Most of the current approaches use labeled documents as the primary source of information for MLTC. We investigate the effectiveness of different sources of information— such as the labeled training data, textual labels of classes, and taxonomy relations of classes— for MLTC. More specifically, first, for each document–class pair, different features are extracted using different sources of information. The features reflect the similarity of classes and documents. Then, MLTC is considered to be a ranking problem, and a learning to rank (LTR) approach is used for ranking classes regarding documents and selecting labels of documents. An important characteristic of many MLTC instances is that documents can belong to multiple classes and there are implicit relations between classes. We apply score propagation on top of LTR to incorporate co-occurrence patterns of classes in labeled documents. Our main findings are the following. First, using an LTR approach integrating all features, we observe significantly better performance than previous systems for MLTC. Specifically, we show that simple classification approaches fail when there is a high number of classes. Second, the analysis of feature weights reveals the relative importance of various sources of evidence, also giving insight into the underlying classification problem. Interestingly, the results indicate that the titles of documents are more informative than all other sources of information. Third, a lean-and-mean system using only four features is able to perform at 96% of the large LTR model that we propose in this paper. Fourth, using the co-occurrence information of classes helps in classifying documents more accurately. Our results show that the co-occurrence information is more helpful when the underlying classifier has a poor performance.

The felicity, or acceptability, of IS generics, i.e. generic sentences with indefinite singulars, is considerably more restricted compared to BP generics, generics with bare plurals. The goal of this paper is to account for the limited felicity of IS generics compared to BP generics, on the one hand, while preserving the close similarity between the two types of generics, on the other. We do so by proposing a causal analysis of IS generics, and show that this corresponds closely with a probabilistic analysis of BP generics.

Although studies on gay dating apps have shown the ways through which technical features intervene in sexual sociality, seldom do they shed light on the central feature that undergirds their use ‒ algorithms. This article examines Blued, China’s largest gay dating app, aiming to unfold the data structure and algorithms that navigate the use of its two prime functionalities: browsing and live streaming. Rather than seeing users as passive audiences that are subject to being measured and calculated based on algorithmically filtered information, this article argues that users can act on the data they provide to dating apps to influence and shape the algorithmic dating outcomes. For Blued browsing, gay men are datafied with numbers and tags for users to sort and filter. For Blued live, gay men are evaluated by the yanzhi algorithm, which literally means ‘value of a person’s face’, for users to pinpoint dating goals. By gaming the data and shaping their algorithmic outcomes, an algorithmic sociality is animated through the filter browsing and the yanzhi metric. In this way, algorithms have become ritual tools for gay sociality on Blued. This article ends by opening up critiques on the justification of social biases lurking inside this algorithmic sociality.

We present four types of neural language models trained on a large historical dataset of books in English, published between 1760-1900 and comprised of ~5.1 billion tokens. The language model architectures include static (word2vec and fastText) and contextualized models (BERT and Flair). For each architecture, we trained a model instance using the whole dataset. Additionally, we trained separate instances on text published before 1850 for the two static models, and four instances considering different time slices for BERT. Our models have already been used in various downstream tasks where they consistently improved performance. In this paper, we describe how the models have been created and outline their reuse potential., Comment: 5 pages, 1 figure

The COVID-19 pandemic, in addition to the obvious implications for public health and the economy at a global scale, has also had significant social and cultural effects: the partial or complete lockdown of public life breaks social and cultural networks and habits, and necessitates the invention of new forms of communication, contact, and inspiration. Over the course of the pandemic, various initiatives have emerged that have tried to capture this unique moment in history by collecting documentation on the social and cultural impact of the pandemic on people’s daily lives. Most of these collections were created bottom-up, either by actively approaching individuals or communities to contribute or be interviewed, or by opening platforms where citizens could directly contribute documentation. In this paper, we analyse two such citizen-generated collections of experiences with the COVID-19 pandemic: the ‘Corona in the City’ collection, assembled by the Amsterdam Museum for an online exhibition with the same title, and the ‘Dagboek Corona’ (Corona Diary) collection of diary contributions, assembled by the Dutch public historian and journalist José Boon. The paper discusses the background of these citizen-generated COVID-19 collections, assessing their scope and accessibility, and presenting a first, qualitative analysis of their content. We conclude with a reflection on the implications of the findings for people’s social life and identity, indicating how the datasets can be explored in further research and how the findings may benefit effective policy making.

Reference annotation datasets containing harmony annotations are at the core of a wide range of studies in music information retrieval (MIR) and related fields. The majority of these datasets contain single reference annotations describing the harmony of each piece. Nevertheless, studies showing differences among annotators in many other MIR tasks make the notion of a single ‘ground-truth’ reference annotation a tenuous one. In this paper, we introduce and analyse the Chordify Annotator Subjectivity Dataset (CASD) containing chord labels for 50 songs from 4 expert annotators in order to gain a better understanding of the differences between annotators in their chord label choice. Our analysis reveals that annotators use distinct chord-label vocabularies, with low chord-label overlap across all annotators. Between annotators, we find only 73 percent overlap on average for the traditional major–minor vocabulary and 54 percent overlap for the most complex chord labels. A factor analysis reveals the relative importance of triads, sevenths, inversions and other musical factors for each annotator on their choice of chord labels and reported difficulty of the songs. Our results further substantiate the existence of a harmonic ‘subjectivity ceiling’: an upper bound for evaluations in computational harmony research. Current state-of-the-art chord-estimation systems perform beyond this subjectivity ceiling by about 10 percent. This suggests that current ACE algorithms are powerful enough to tune themselves to particular annotators' idiosyncrasies. Overall, our results show that annotator subjectivity is an important factor in harmonic transcriptions, which should inform future studies into harmony perception and computational models of harmony.

Many generic sentences express stable inductive generalizations. Stable inductive generalizations are typically true for a causal reason. In this paper we investigate to what extent this is also the case for the generalizations expressed by generic sentences. More in particular, we discuss the possibility that many generic sentences of the form ‘ks have feature e’ are true because (members of) kind k have the causal power to ‘produce’ feature e. We will argue that such an analysis is quite close to a probabilistic based analysis of generic sentences according to which ‘relatively many’ ks have feature e, and that, in fact, this latter type of analysis can be ‘grounded’ in terms of causal powers. We will argue, moreover, that the causal power analysis is sometimes preferred to a correlation-based analysis, because it takes the causal structure that gives rise to the probabilistic data into account.

We investigate the semantic knowledge of language models (LMs), focusing on (1) whether these LMs create categories of linguistic environments based on their semantic monotonicity properties, and (2) whether these categories play a similar role in LMs as in human language understanding, using negative polarity item licensing as a case study. We introduce a series of experiments consisting of probing with diagnostic classifiers (DCs), linguistic acceptability tasks, as well as a novel DC ranking method that tightly connects the probing results to the inner workings of the LM. By applying our experimental pipeline to LMs trained on various filtered corpora, we are able to gain stronger insights into the semantic generalizations that are acquired by these models., Comment: Published in ACL Findings 2021

In this paper we argue that for the (probabilistic) interpretation of generic sentences of the form “Gs are f,” three types of alternatives play a role: (i) alternative features of f, (ii) alternative groups, or kinds, of G, and (iii) alternative causal background factors. In the first part of this paper we argue for the relevance of these alternatives. In the second part, we describe the results of some experiments that empirically tested in particular the second use of alternatives.

We present a novel, domain expert-controlled, replicable procedure for the construction of concept-modeling ground truths with the aim of evaluating the application of word embeddings. In particular, our method is designed to evaluate the application of word and paragraph embeddings in concept-focused textual domains, where a generic ontology does not provide enough information. We illustrate the procedure, and validate it by describing the construction of an expert ground truth, QuiNE-GT. QuiNE-GT is built to answer research questions concerning the concept of naturalized epistemology in QUINE, a 2-million-token, single-author, 20th-century English philosophy corpus of outstanding quality, cleaned up and enriched for the purpose. To the best of our ken, expert concept-modeling ground truths are extremely rare in current literature, nor has the theoretical methodology behind their construction ever been explicitly conceptualised and properly systematised. Expert-controlled concept-modeling ground truths are however essential to allow proper evaluation of word embeddings techniques, and increase their trustworthiness in specialised domains in which the detection of concepts through their expression in texts is important. We highlight challenges, requirements, and prospects for future work.

This paper is part of an on-going programme on the study of the logical aspects of social network formation. It recalls the so-called social network model, discussing the properties of a notion of closeness between agents (in terms of the number of traits they have in common); then introduces an extended social network model in which different agents might assign different values to different traits, discussing the properties of the notion of weighted closeness that arises. These notions are used to define social network creation operations by means of a threshold strategy. The paper studies the properties of the social networks the updates create, providing sound and complete axiom systems for formal languages describing these updates’ effects.

We present chant21, a Python package to support the plainchant formats gabc and Volpiano in music21, and two large corpora of plainchant. The CantusCorpus contains over 60,000 medieval melodies collected from the Cantus database, encoded in the Volpiano typeface. The GregoBaseCorpus contains over 9,000 transcriptions from more recent chant books in the gabc format. Chant21 converts both formats to music21, while retaining the textual structure of the chant: its division in sections, words, syllables and neumes. We present two case studies. First, we report evidence for the melodic arch hypothesis from the GregoBaseCorpus. Second, we analyze connections between differentiæ and antiphon openings in the CantusCorpus, and show that the systematicity of the connection can be quantified using an entropy-based measure.

A growing volume of heritage data is being digitized and made available as text via optical character recognition (OCR). Scholars and libraries are increasingly using OCR-generated text for retrieval and analysis. However, the process of creating text through OCR introduces varying degrees of error to the text. The impact of these errors on natural language processing (NLP) tasks has only been partially studied. We perform a series of extrinsic assessment tasks — sentence segmentation, named entity recognition, dependency parsing, information retrieval, topic modelling and neural language model fine-tuning — using popular, out-of-the-box tools in order to quantify the impact of OCR quality on these tasks. We find a consistent impact resulting from OCR errors on our downstream tasks with some tasks more irredeemably harmed by OCR errors. Based on these results, we offer some preliminary guidelines for working with text produced through OCR.

This paper is part of an on-going programme in which we provide a logical study of social network formations. In the proposed setting, agent a will consider agent b as part of her network if the number of features (properties) on which they differ is small enough, given the constraints on the size of agent a’s ‘social space’. We import this idea about a limit on one’s social space from the cognitive science literature. In this context we study the creation of new networks and use the tools of Dynamic Epistemic Logic to model the updates of the networks. By providing a set of reduction axioms we are able to provide sound and complete axiomatizations for the logics studied in this paper.

A growing field of research has made use of a semiartificial language paradigm to investigate the role of awareness in L2 acquisition. A central and empirically still unresolved issue in this field concerns the possibility of learning implicitly, that is, without intention to learn and without awareness of what has been learned. Up until now, studies on implicit learning have mainly been conducted in laboratory settings under highly controlled conditions with university students as participants. The present study investigated whether and to what extent the results obtained in such settings can be extrapolated to the general population. Building on Williams (2005), we designed two crowdsourcing experiments that examined the learning of novel form-meaning mappings under incidental conditions in 163 participants. Our design allowed us to disentangle the effects of awareness at the level of noticing and understanding. The results of the two experiments demonstrated the implicit learning effect outside the lab in a more varied sample of participants and indicated that awareness at both levels appears to have a facilitative effect on learning outcomes.

In this paper, we propose a unified logical framework for representing and analyzing various forms of correlated information change. Our main thesis is that "logical dynamics," in the sense of van Benthem (Exploring logical dynamics. CSLI Publications, Stanford, 1996; Logical dynamics of information and interaction. Cambridge University Press, Cambridge, 2011), and in particular dynamic epistemic notions of conditional, as developed in Baltag and Smets (Electron Notes Theor Comput Sci 165:5---21, 2006a; Stud Log 89:185---209, 2008a; Texts in logic and games. Amsterdam University Press, Amsterdam, pp 9---58, 2008b), play a central role in understanding and modeling a wide range of apparently very different information-gathering phenomena which do have one specific feature in common, namely the very act of learning new information may directly change the reality that is being learned. On the one hand, we focus on the way in which an introspective agent changes her beliefs when learning new higher-order information, i.e., information that may refer to her own beliefs. On the other hand, we analyze situations in which an observer learns about a phenomenon by performing observations that may perturb the very phenomenon under study, as in the case of quantum measurements, or observations in social sciences, psychology and medicine. Our formal techniques are based on ideas from dynamic logic and on the modeling of "dynamic conditionals." We offer a semantics based on "test frames," i.e., Kripke frames labeled by propositional formulae which yields a unified setting for the two types of correlated information change under study. We show how this framework can be used to analyze the ontic and epistemic---informational aspects of quantum measurements and to compare them with other types of observation, testing, belief revision, counterfactual conditionals, etc.

Health ismorethan the absence of disease. It is also more than a biological phenomenon. It is inherently social, psychological, cultural and historical. While this has been recognised by major health actors for decades, open questions remain as to how to build systems that reflect the complexity of health, disease and sickness, and in a context that is increasingly technologised. We argue that an urgent change of approach is necessary. Methods and concepts from the humanities and social science must be embedded in the concepts and methods of the health sciences if we are to promote sustainable interventions capable of engaging with the recognised complexity of health, disease and sickness. Our vision is one of radical interdisciplinarity, integrating aspects of biological, psychological, social and humanities approaches across areas of urgent health need. Radical interdisciplinarity, we argue, entails the practical, methodological and conceptual integration of these approaches to health.

This paper is part of an on-going research programme on the logical aspects of social network creation. In our proposed setting we make use of a notion of distance between agents, defined as the number of features distinguishing them. Following the ideas of the cognitive science literature, we build the social network by adopting a group-size approach: each agent's new social space will contain the λ agents that are closer to her. The paper studies the properties of the resulting networks, providing a sound and complete axiom system for a logical language which describes the changes that the social network creation operation brings about. This is done not only for the basic case in which the distance is relative to all basic features and the more general one in which the distance is relative to a given subset of them, but also for the more interesting scenario in which the distance is relative to a set of more complex social features.

Visual Cultural Heritage has extensively been explored using multimedia methods, but has so far been limited to still images. In particular, Early Cinema has hardly been explored. We analyze the Desmet collection, a recently digitized collection of early cinema (1907-1916), in the context of intertitles. Intertitles played an important role in silent movies in order to convey the main narratives, and split the film into semantically meaningful segments. We first build several classifiers to detect these intertitles, and evaluate it on a gold standard collection annotated by an expert. We illustrate the usefulness of using Deep Learning methods to extract semantic features to analyze the role of intertitles in early cinema. Furthermore, we attempt to structure and map the narrative progression of a film with respect to the locations at which shots were filmed.

INEX ran as an independent evaluation forum for 10 years before it teamed up with CLEF in 2012. Even before 2012 there was considerable collaboration between INEX and CLEF, and these collaborations increased in intensity when CLEF moved beyond its traditional cross-lingual focus in 2009/2010 shifting to include all experimental IR. This led to the merger of CLEF and INEX, and effectively to the inclusion of INEX as a large track or lab into CLEF in 2012. This chapter details the efforts of the INEX lab in CLEF (2012–2014), as well as the ongoing activities as separate labs, under the labels Social Book Search (2015–2016), and Microblog Contextualization (2016–2018).

A substantial body of research has demonstrated that both native and non-native speakers are sensitive to the statistics of multiword sequences (MWS). However, this research has predominantly focused on demonstrating that a given sample of participants shows evidence of learning the statistical properties of MWS. Recent theoretical approaches to language learning and processing emphasize the importance of moving away from group-level analyses towards analyses that account for individual differences (IDs). Here, through a within subject design embedded within an IDs framework, we investigate whether and to what extent individual variability in the online processing of MWS are associated with the statistical learning (SL) ability of an individual. Second language learners were administered a battery of SL tasks in the visual and auditory modalities, using verbal and non-verbal stimuli, with adjacent and non-adjacent contingencies along with two online processing tasks of MWS designed to assess sensitivity to the statistics of spoken and written language. We found a number of significant associations between the SL ability and the two processing tasks: Individuals who performed better on an auditory verbal adjacent SL task demonstrated greater sensitivity to the statistics of MWS in the spoken language, whereas individuals with better performance on a visual, non-verbal sequence learning task demonstrated greater sensitivity to the statistics of MWS in the written language. We discuss the implications of these findings for the study of IDs in the processing of MWS.

One of the great successes of the application of generalized quantifiers to natural language has been the ability to formulate robust semantic universals. When such a universal is attested, the question arises as to the source of the universal. In this paper, we explore the hypothesis that many semantic universals arise because expressions satisfying the universal are easier to learn than those that do not. While the idea that learnability explains universals is not new, explicit accounts of learning that can make good on this hypothesis are few and far between. We propose a model of learning — back-propagation through a recurrent neural network — which can make good on this promise. In particular, we discuss the universals of monotonicity, quantity, and conservativity and perform computational experiments of training such a network to learn to verify quantifiers. Our results are able to explain monotonicity and quantity quite well. We suggest that conservativity may have a different source than the other universals. EARLY ACCESS

In this work, we address the evaluation of distributional semantic models trained on smaller, domain-specific texts, particularly philosophical text. Specifically, we inspect the behaviour of models using a pretrained background space in learning. We propose a measure of consistency which can be used as an evaluation metric when no in-domain gold-standard data is available. This measure simply computes the ability of a model to learn similar embeddings from different parts of some homogeneous data. We show that in spite of being a simple evaluation, consistency actually depends on various combinations of factors, including the nature of the data itself, the model used to train the semantic space, and the frequency of the learned terms, both in the background space and in the in-domain data of interest.

Sitting too much is unhealthy, but a widespread habit in many societies. Realizing behavioral change in this area is hard. Our societies promote being seated via the way its places are structured: they are filled with chairs for example. How can we make healthier environments that invite people to move around more? This article shows how philosophical research in the area of embodied/enactive cognitive science let to a built vision for the office of the future, of 2025. Multidisciplinary studio RAAAF [Rietveld Architecture-Art-Affordances] and visual artist Barbara Visser built this world without chairs, titled The End of Sitting. This large rock-like landscape integrates many affordances for standing. Affordances are the possibilities for action provided by the environment. This landscape of standing affordances allows people to work standing while being supported by the material structure of the environment. This unorthodox working landscape is both an enactive art installation and the materialization of a philosophical worldview that understands people as embodied minds situated in a landscape of affordances. It stimulates reflection on the way built environments can naturally invite more active and healthy behavior.

This reply to Oaksford and Chater’s (O&C)’s critical discussion of our use of logic programming (LP) to model and predict patterns of conditional reasoning will frame the dispute in terms of the semantics of the conditional. We begin by outlining some common features of LP and probabilistic conditionals in knowledge-rich reasoning over long-term memory knowledge bases. For both, context determines causal strength; there are inferences from the absence of certain evidence; and both have analogues of the Ramsey test. Some current work shows how a combination of counting defeaters and statistics from network monitoring can provide the information for graded responses from LP reasoning. With this much introduction, we then respond to O&C’s specific criticisms and misunderstandings.

Semantic universals are properties of meaning shared by the languages of the world. We offer an explanation of the presence of such universals by measuring simplicity in terms of ease of learning, showing that expressions satisfying universals are simpler than those that do not according to this criterion. We measure ease of learning using tools from machine learning and analyze universals in a domain of function words (quantifiers) and content words (color terms). Our results provide strong evidence that semantic universals across both function and content words reflect simplicity as measured by ease of learning.

The availability of massive data and computing power allowing for effective data driven neural approaches is having a major impact on machine learning and information retrieval research, but these models have a basic problem with efficiency. Current neural ranking models are implemented as multistage rankers: for efficiency reasons, the neural model only re-ranks the top ranked documents retrieved by a first-stage efficient ranker in response to a given query. Neural ranking models learn dense representations causing essentially every query term to match every document term, making it highly inefficient or intractable to rank the whole collection. The reliance on a first stage ranker creates a dual problem: First, the interaction and combination effects are not well understood. Second, the first stage ranker serves as a "gate-keeper" or filter, effectively blocking the potential of neural models to uncover new relevant documents. In this work, we propose a standalone neural ranking model (SNRM) by introducing a sparsity property to learn a latent sparse representation for each query and document. This representation captures the semantic relationship between the query and documents, but is also sparse enough to enable constructing an inverted index for the whole collection. We parameterize the sparsity of the model to yield a retrieval model as efficient as conventional term based models. Our model gains in efficiency without loss of effectiveness: it not only outperforms the existing term matching baselines, but also performs similarly to the recent re-ranking based neural models with dense representations. Our model can also take advantage of pseudo-relevance feedback for further improvements. More generally, our results demonstrate the importance of sparsity in neural IR models and show that dense representations can be pruned effectively, giving new insights about essential semantic features and their distributions.

In this paper we design a new logical system to explicitly model the different deductive reasoning steps of a boundedly rational agent. We present an adequate system in line with experimental findings about an agent’s reasoning limitations and the cognitive effort that is involved. Inspired by Dynamic Epistemic Logic, we work with dynamic operators denoting explicit applications of inference rules in our logical language. Our models are supplemented by (a) impossible worlds (not closed under logical consequence), suitably structured according to the effect of inference rules, and (b) quantitative components capturing the agent’s cognitive capacity and the cognitive costs of rules with respect to certain resources (e.g. memory, time). These ingredients allow us to avoid problematic logical closure principles, while at the same time deductive reasoning is reflected in our dynamic truth clauses. We finally show that our models can be reduced to awareness-like plausibility structures that validate the same formulas and a sound and complete axiomatization is given with respect to them.

Research in the health sciences has been highly successful in revealing the aetiologies of many morbidities, particularly those involving the microbiology of communicable disease. This success has helped form a narrative to be found in numerous public health documents, about interventions to reduce the burden of non‐communicable diseases (e.g., obesity or alcohol related pathologies). These focus on tackling the purported pathogenic factors causing the diseases as a means of prevention. In this paper, we argue that this approach has been sub‐optimal. The mechanisms of aetiology and of prevention are sometimes significantly different and failure to make this distinction has hindered efforts at preventing non‐communicable diseases linked to diet, exercise and alcohol consumption. We propose a sociological approach as an alternative based on social practice theory. (A virtual abstract for this paper can be found at: https://www.youtube.com/channel/UC_979cmCmR9rLrKuD7z0ycA).

Most high-level auditory functions require one to detect the onset and offset of sound sequences as well as registering the rate at which sounds are presented within the sound trains. By recording event-related brain potentials to onsets and offsets of tone trains as well as to changes in the presentation rate, we tested whether these fundamental auditory capabilities are functional at birth. Each of these events elicited significant event-related potential components in sleeping healthy neonates. The data thus demonstrate that the newborn brain is sensitive to these acoustic features suggesting that infants are geared towards the temporal aspects of segregating sound sources, speech and music perception already at birth.

Despite the impressive improvements achieved by unsupervised deep neural networks in computer vision and NLP tasks, such improvements have not yet been observed in ranking for information retrieval. The reason may be the complexity of the ranking problem, as it is not obvious how to learn from queries and documents when no supervised signal is available. Hence, in this paper, we propose to train a neural ranking model using weak supervision, where labels are obtained automatically without human annotators or any external resources (e.g., click data). To this aim, we use the output of an unsupervised ranking model, such as BM25, as a weak supervision signal. We further train a set of simple yet effective ranking models based on feed-forward neural networks. We study their effectiveness under various learning scenarios (point-wise and pair-wise models) and using different input representations (i.e., from encoding query-document pairs into dense/sparse vectors to using word embedding representation). We train our networks using tens of millions of training instances and evaluate it on two standard collections: a homogeneous news collection(Robust) and a heterogeneous large-scale web collection (ClueWeb). Our experiments indicate that employing proper objective functions and letting the networks to learn the input representation based on weakly supervised data leads to impressive performance, with over 13% and 35% MAP improvements over the BM25 model on the Robust and the ClueWeb collections. Our findings also suggest that supervised neural ranking models can greatly benefit from pre-training on large amounts of weakly labeled data that can be easily obtained from unsupervised IR models., Comment: In proceedings of The 40th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR2017)

Enculturation is known to shape the perception of meter in music but this is not explicitly accounted for by current cognitive models of meter perception. We hypothesize that the induction of meter is a result of predictive coding: interpreting onsets in a rhythm relative to a periodic meter facilitates prediction of future onsets. Such prediction, we hypothesize, is based on previous exposure to rhythms. As such, predictive coding provides a possible explanation for the way meter perception is shaped by the cultural environment. Based on this hypothesis, we present a probabilistic model of meter perception that uses statistical properties of the relation between rhythm and meter to infer meter from quantized rhythms. We show that our model can successfully predict annotated time signatures from quantized rhythmic patterns derived from folk melodies. Furthermore, we show that by inferring meter, our model improves prediction of the onsets of future events compared to a similar probabilistic model that does not infer meter. Finally, as a proof of concept, we demonstrate how our model can be used in a simulation of enculturation. From the results of this simulation, we derive a class of rhythms that are likely to be interpreted differently by enculturated listeners with different histories of exposure to rhythms.

There is broad consensus in the field of IR that search is complex in many use cases and applications, both on the Web and in domain specific collections, and both professionally and in our daily life. Yet our understanding of complex search tasks, in comparison to simple look up tasks, is fragmented at best. The workshop addresses many open research questions: What are the obvious use cases and applications of complex search? What are essential features of work tasks and search tasks to take into account? And how do these evolve over time? With a multitude of information, varying from introductory to specialized, and from authoritative to speculative or opinionated, when to show what sources of information? How does the information seeking process evolve and what are relevant differences between different stages? With complex task and search process management, blending searching, browsing, and recommendations, and supporting exploratory search to sensemaking and analytics, UI and UX design pose an overconstrained challenge. How do we evaluate and compare approaches? Which measures should be taken into account? Supporting complex search tasks requires new collaborations across the fields of CHI and IR, and the proposed workshop will bring together a diverse group of researchers to work together on one of the greatest challenges of our field.

This article aims to achieve two goals: to show that probability is not the only way of dealing with uncertainty (and even more, that there are kinds of uncertainty which are for principled reasons not addressable with probabilistic means); and to provide evidence that logic-based methods can well support reasoning with uncertainty. For the latter claim, two paradigmatic examples are presented: Logic Programming with Kleene semantics for modelling reasoning from information in a discourse, to an interpretation of the state of affairs of the intended model, and a neural-symbolic implementation of Input/Output logic for dealing with uncertainty in dynamic normative contexts., Forthcoming with DOI 10.1007/s11023-017-9428-3 in the Special Issue "Reasoning with Imperfect Information and Knowledge" of Minds and Machines (2017). The final publication will be available at http://link.springer.com. --- Changes to previous version: Fixed some typos and a broken reference

For philosophers to be able to take a computational turn in their field, especially if that field relies heavily on historical material, it is crucial to be able to build high-quality, easily and freely accessible corpora in a sustainable format composed from multi-language, multi-script books from different historical periods. At the moment, corpora matching these needs are virtually non-existent. Within the CLARIN-NL project @PhilosTEI, we have addressed the problem of building this kind of corpora by developing an open-source, web-based, user-friendly workflow from textual images to TEI, based on state-of-the-art open-source OCR software Tesseract, and a multi-language version of TICCL, a powerful OCR post-correction tool. We have demonstrated the utility of the @PhilosTEI tool by applying it to a multilingual, multi-script corpus of important 18th to 20th century European philosophical texts.

In this paper, we propose that a sentence like John T ate broccoli F should pragmatically be interpreted as follows: (a) Focus should be interpreted exhaustively; John ate only broccoli; (b) Topic must be interpreted exhaustively: Only John ate (only) broccoli; and (c) The speaker takes it to be possible (or even knows, if he is competent) that at least one alternative of the form x ate y not entailed by the sentence is true. It will be shown that in terms of this analysis we can also account for all the scope-inversion data of Büring (Linguist Philos 20: 175–194, 1997), without giving rise to some of the problems of the latter analysis.

In this paper we provide a formal model for occasion-sensitive semantics motivated by so called ‘Travis cases’ (Travis 1978, 2000, 2008, 2009). We suggest that understanding of an utterance of φ (knowing its truth conditions) can be modelled as a twofold partitioning of worlds in logical space, where the initial partition is induced by the context-invariant meaning of the sentence uttered and the latter on the basis of context-dependent goals. Our model uses only a single parameter to capture occasion-sensitivity of sentences: practical goals.

Dependency Grammar has been taken as a formalism for syntactic representation, comparable to close competitors such as phrase structure grammar or categorial grammar. This paper argues that in fact the dependency graphs (DGs) should—like semantic frames—be seen as a semantic formalism like e.g. FOL, Montague’s IL or Discourse Representation Structures. For this, arrows must have semantically interpretable labels and two additional kinds of arrows need to be added: scope arrows and anaphoric arrows.

This paper studies the logical features of social group creation. We focus on the mechanisms which indicate when agents can form a team based on the correspondence in their set of features (behavior, opinions, etc.). Our basic approach uses a semi-metric on the set of agents, which is used to construct a network topology. Then it is extended with epistemic features to represent the agents’ epistemic states, allowing us to explore group-creation alternatives where what matters is not only the agent’s differences but also what they know about them. We use tools of dynamic epistemic logic to study the properties of different strategies to network formations.

There is a growing interests in integration of Internet of Things (IoT) in smart environments, which creates an opportunity to understand users’ information needs using onsite physical sensor logs. However, the physical context creates numerous external factors that play a role in users’ information interactions, thus creating new external biases in the collected information interaction logs. In order to provide an effective personalized experiences for users in smart environment, we need to take care of these external biases in the behavioral user models. Our general aim is to understand users’ onsite physical behaviors for providing online and onsite personalized services like personalized tour guides. We focus on the cultural heritage domain and collect onsite users’ physical information interaction logs of visits in a museum. This prompts the question: How to understand users’ behavior in the existence of external biases? Our main finding is that users behave differently in their solitude in comparison to a busy museum situation. Specifically, visitors’ crowd bias has a considerable effect on users’ following position rank bias based check-in behavior. Our study investigates on understanding users’ onsite physical behavior accurately, whichcan improve the state-of-the-art onsite behavioral user models.

Mobile devices and the internet of things blend our virtual online behavior with our actions in the real-world. The physical context creates numerous external factors that play a role in the user's online interactions, thus creating new external biases in the collected information interaction logs. Our general aim is to improve our understanding of onsite users' behavior, which allow us to create better online and onsite contextual suggestion systems. We focus on the cultural heritage domain and have collected onsite users' information interaction logs of visits in a museum. This prompts the question: How to understand users' behavior in order to be able to predict their onsite behaviors? Our main findings are the following: First, users behave differently in different onsite contextual situations. Second, there is a significant dependency between users' onsite behaviors and other users, who are interacting with next point of interests (POIs). Third, we have proposed a contextual Skip-Or-Stay behavior classifier based on four different pairwise contextual features that significantly improves a defined baseline based on all the considered evaluation metrics. Fourth, we have analyzed the importance of different contexts in the Skip-or-Stay behavior predictions.

Recently, researchers started to pay attention to the detection of temporal shifts in the meaning of words. However, most (if not all) of these approaches restricted their efforts to uncovering change over time, thus neglecting other valuable dimensions such as social or political variability. We propose an approach for detecting semantic shifts between different viewpoints--broadly defined as a set of texts that share a specific metadata feature, which can be a time-period, but also a social entity such as a political party. For each viewpoint, we learn a semantic space in which each word is represented as a low dimensional neural embedded vector. The challenge is to compare the meaning of a word in one space to its meaning in another space and measure the size of the semantic shifts. We compare the effectiveness of a measure based on optimal transformations between the two spaces with a measure based on the similarity of the neighbors of the word in the respective spaces. Our experiments demonstrate that the combination of these two performs best. We show that the semantic shifts not only occur over time, but also along different viewpoints in a short period of time. For evaluation, we demonstrate how this approach captures meaningful semantic shifts and can help improve other tasks such as the contrastive viewpoint summarization and ideology detection (measured as classification accuracy) in political texts. We also show that the two laws of semantic change which were empirically shown to hold for temporal shifts also hold for shifts across viewpoints. These laws state that frequent words are less likely to shift meaning while words with many senses are more likely to do so., Comment: In Proceedings of the 26th ACM International on Conference on Information and Knowledge Management (CIKM2017)

This is a publisher ranking study based on a citation data grant from Elsevier, specifically, book titles cited in Scopus history journals (20072011) and matching metadata from WorldCat® (i.e., OCLC numbers, ISBN codes, publisher records, and library holding counts). Using both resources, we have created a unique relational database designed to compare citation counts to books with international library holdings or libcitations for scholarly book publishers. First, we construct a ranking of the top 500 publishers and explore descriptive statistics at the level of publisher type (university, commercial, other) and country of origin. We then identify the top 50 university presses and commercial houses based on total citations and mean citations per book (CPB). In a third analysis, we present a map of directed citation links between journals and book publishers. American and British presses/publishing houses tend to dominate the work of library collection managers and citing scholars; however, a number of specialist publishers from Europe are included. Distinct clusters from the directed citation map indicate a certain degree of regionalism and subject specialization, where some journals produced in languages other than English tend to cite books published by the same parent press. Bibliometric rankings convey only a small part of how the actual structure of the publishing field has evolved; hence, challenges lie ahead for developers of new citation indices for books and bibliometricians interested in measuring book and publisher impacts.

The field of web archiving is at a turning point. In the early years of web archiving, the single URL has been the dominant unit for preservation and access. Access tools such as the Internet Archive's Wayback Machine reflect this notion as they allowed consultation, or browsing, of one URL at a time. In recent years, however, the single URL approach to accessing web archives is being gradually replaced by search interfaces. This paper addresses the theoretical and methodological implications of the transition to search on web archive research. It introduces ‘search as research’ methods, practices already applied in studies of the live web, which can be repurposed and implemented for critically studying archived web data. Such methods open up a variety of analytical practices that were so far precluded by the single URL entry point to the web archive, such as the re-assemblage of existing collections around a theme or an event, the study of archival artefacts and scaling the unit of analysis from the single URL to the full archive, by generating aggregate views and summaries. The paper introduces examples to ‘search as research’ scenarios, which have been developed by the WebART project at the University of Amsterdam and the Centrum Wiskunde & Informatica, in collaboration with the National Library of the Netherlands. The paper concludes with a discussion of current and potential limitations of ‘search as research’ methods for studying web archives, and the ways with which they can be overcome in the near future.

Collecting ground truth data for music research requires large amounts of time and money. To avoid these costs, researchers are now trying to collect information through online multiplayer games with the underlying purpose of collecting scientific data. In this paper we present two case studies of such games created for data collection in music information retrieval (MIR): Emotify, for emotional annotation of music, and Hooked, for studying musical catchiness. In addition to the basic requirement of scientific validity, both applications address essential development and design issues, for example, acquiring licensed music or employing popular social frameworks. As such, we hope that they may serve as blueprints for the development of future serious games, not only for music but also for other humanistic domains. The pilot launch of these two games showed that their models are capable of engaging participants and supporting large-scale empirical research.

This article aims to propagate Logic Programming as a formal tool to deal with non-monotonic reasoning. In philosophy and linguistics non-monotonic reasoning is modelled using Minimal Models as standard, i.e., by imposing an order (or selection function) on the class of all models and then by defining entailment as only caring about the minimal models of the premises with respect to the order. In this article we investigate the question whether instead of minimal models we should use logic programming to model non-monotonic reasoning. Logic programming is an attractive alternative to a minimal models approach in that it makes concrete predictions in an efficient and transparent way. We study this question by focusing on one particular phenomenon that gives rise to non-monotonic inferences: conditional sentences.