Abstract: This chapter outlines the main challenges a theory for the cultural evolution of language should address and proposes a particular theory which is worked out and explored in greater detail in the remaining chapters of this book. The theory rests on two biologically inspired mechanisms, namely selection and self-organization, mapped onto the cultural, more specifically, linguistic domain. Selectionism is an alternative to rational top-down design. It introduces a distinction between processes that generate possible linguistic variants in a population (for example, different ways to express tense and aspect) and processes that select some variants to survive and become dominant in a language, based on criteria that translate into increased communicative success, such as expressive adequacy, minimal cognitive effort, learnability and social conformity. Self-organization occurs when speakers and hearers align their communication systems based on the outcome of each interaction. It explains how convergence may arise without central coordination or direct telepathic meaning transfer. This chapter explains these basic hypotheses in more detail and introduces a methodology for exploring them based on the notion of a language game.

Abstract: There is often more than one way to represent syntactic structures, even within a given formalism. Selecting one representation over another may affect parsing performance. Therefore, selecting between alternative syntactic representations (hencefor th, syntactic selection) is an essential step in designing an annotation scheme. We present a methodology for syntactic selection and apply it to six central dependency structures. Our methodology compares pairs of annotation schemes that differ in the annotation of a single structure. It selects th e more learnable scheme, namely the one that can be better learned using statistical parsers. We find that in three of the structures, one annotation is unequivocally better than the alternatives. Our r esults are consistent over various settings involving five parsers and two definitions of learnability. F urthermore, we show that the learnability gains incurred by our selections are both considerable ( error reductions of up to 19.8%) and additive. The contribution of this work is in demonstrating that syntactic selection has a substantial and predictable effect on parsing performance, and showing that this effect can be effectively used in designing syntactic annotation schemes.

Abstract: In this paper we study the approximate learnability of valuations commonly used throughout economics and game theory for the quantitative encoding of agent preferences. We provide upper and lower bounds regarding the learnability of important subclasses of valuation functions that express no-complementarities. Our main results concern their approximate learnability in the distributional learning (PAC-style) setting. We provide nearly tight lower and upper bounds of $\tilde{\Theta}(n^{1/2})$ on the approximation factor for learning XOS and subadditive valuations, both widely studied superclasses of submodular valuations. Interestingly, we show that the $\tilde{\Omega}(n^{1/2})$ lower bound can be circumvented for XOS functions of polynomial complexity; we provide an algorithm for learning the class of XOS valuations with a representation of polynomial size achieving an $O(n^{\eps})$ approximation factor in time $O(n^{1/\eps})$ for any $\eps > 0$. This highlights the importance of considering the complexity of the target function for polynomial time learning. We also provide new learning results for interesting subclasses of submodular functions. Our upper bounds for distributional learning leverage novel structural results for all these valuation classes. We show that many of these results provide new learnability results in the Goemans et al. model (SODA 2009) of approximate learning everywhere via value queries. We also introduce a new model that is more realistic in economic settings, in which the learner can set prices and observe purchase decisions at these prices rather than observing the valuation function directly. In this model, most of our upper bounds continue to hold despite the fact that the learner receives less information (both for learning in the distributional setting and with value queries), while our lower bounds naturally extend.

Abstract: This chapter outlines the main challenges a theory for the cultural evolution of language should address and proposes a particular theory which is worked out and explored in greater detail in the remaining chapters of this book. The theory rests on two biologically inspired mechanisms, namely selection and self-organization, mapped onto the cultural, more specifically, linguistic domain. Selectionism is an alternative to rational top-down design. It introduces a distinction between processes that generate possible linguistic variants in a population (for example, different ways to express tense and aspect) and processes that select some variants to survive and become dominant in a language, based on criteria that translate into increased communicative success, such as expressive adequacy, minimal cognitive effort, learnability and social conformity. Self-organization occurs when speakers and hearers align their communication systems based on the outcome of each interaction. It explains how convergence may arise without central coordination or direct telepathic meaning transfer. This chapter explains these basic hypotheses in more detail and introduces a methodology for exploring them based on the notion of a language game.

Abstract: This article presents one of the first systematic usability investigations for a Volunteered Geographic Information (VGI) editor. This research is motivated by the fact that although VGI is now widely consumed, contribution rates are lagging considerably. Compared to traditional GIS interfaces, with complex interfaces resulting in high cognitive loads and barriers to participation, VGI tools and interfaces need to be easy to use and learn to encourage and facilitate contributions. This article develops a case study of OpenStreetMap, one of the most successful VGI projects, and its default editor, Potlatch2. Ten participants with no prior experience of VGI contribution, were instructed to contribute data to OSM in a structured exercise, while being monitored using an eye tracker and audio/video screen capture software. Each participant was asked to Think Aloud, i.e. describe what they were thinking and doing as they completed the tasks. The results highlight significant usability issues impacting learnability, especially from the perspective of a new contributor: hidden functionality, lack of user feedback between interactions and the inefficient and inconsistent placement of functionality and map controls. The facilitation of VGI contributions clearly depends on designing targeted interfaces, optimized to the needs of specific levels of contributors with defined goals and expectations.

Abstract: In this dissertation we study statistical and online learning problems from an optimization viewpoint.The dissertation is divided into two parts : I. We first consider the question of learnability for statistical learning problems in the general learning setting. The question of learnability is well studied and fully characterized for binary classification and for real valued supervised learning problems using the theory of uniform convergence. However we show that for the general learning setting uniform convergence theory fails to characterize learnability. To fill this void we use stability of learning algorithms to fully characterize statistical learnability in the general setting. Next we consider the problem of online learning. Unlike the statistical learning framework there is a dearth of generic tools that can be used to establish learnability and rates for online learning problems in general. We provide online analogs to classical tools from statistical learning theory like Rademacher complexity, covering numbers, etc. We further use these tools to fully characterize learnability for online supervised learning problems. II. In the second part, for general classes of convex learning problems, we provide appropriate mirror descent (MD) updates for online and statistical learning of these problems. Further, we show that the the MD is near optimal for online convex learning and for most cases, is also near optimal for statistical convex learning. We next consider the problem of convex optimization and show that oracle complexity can be lower bounded by the so called fat-shattering dimension of the associated linear class. Thus we establish a strong connection between offline convex optimization problems and statistical learning problems. We also show that for a large class of high dimensional optimization problems, MD is in fact near optimal even for convex optimization.

Abstract: To autonomously learn behaviors in complex environments, vision-based agents need to develop useful sensory abstractions from high-dimensional video. We propose a modular, curiosity-driven learning system that autonomously learns multiple abstract representations. The policy to build the library of abstractions is adapted through reinforcement learning, and the corresponding abstractions are learned through incremental slow-feature analysis (IncSFA). IncSFA learns each abstraction based on how the inputs change over time, directly from unprocessed visual data. Modularity is induced via a gating system, which also prevents abstraction duplication. The system is driven by a curiosity signal that is based on the learnability of the inputs by the current adaptive module. After the learning completes, the result is multiple slow-feature modules serving as distinct behavior-specific abstractions. Experiments with a simulated iCub humanoid robot show how the proposed method effectively learns a set of abstractions from raw un-preprocessed video, to our knowledge the first curious learning agent to demonstrate this ability.

Abstract: Computational learning theory explores the limits of learnability. Studying language acquisition from this perspective involves identifying classes of languages that are learnable from the available data, within the limits of time and computational resources available to the learner. Different models of learning can yield radically different learnability results, where these depend on the assumptions of the model about the nature of the learning process, and the data, time, and resources that learners have access to. To the extent that such assumptions accurately reflect human language learning, a model that invokes them can offer important insights into the formal properties of natural languages, and the way in which their representations might be efficiently acquired. In this chapter we consider several computational learning models that have been applied to the language learning task. Some of these have yielded results that suggest that the class of natural languages cannot be efficiently learned from the primary linguistic data (PLD) available to children, through

Abstract: Relative clauses as complex syntactic structures in human languages have attracted the attention of a lot of second language acquisition researchers. They are difficult for learners to produce, comprehend and imitate. The present study was an attempt to investigate the learnability of the English relative clauses by Persian learners. After discussing the importance of relative clauses, the role it plays in language acquisition, the hypotheses were proposed and investigated. The findings show that the acquisition or frequency rank order of four types of relative clauses was OS > OO > SS > SO, and that OS and OO relative clause types would be easier to acquire than SS and SO types.

Abstract: The acquisition of Bantu noun class prefixes has long been an issue of theoretical interest, due in part to the large number of gender classes. In contrast, the acquisition of Bantu nominal agreement has received little attention. Given findings from other languages, one might expect the phonologically transparent system of Bantu agreement to be mastered early and easily. However, the recent discovery that Sotho languages permit null prefixes under certain grammatical conditions raises the possibility that learning nominal agreement might be more challenging than originally thought. The goal of this study was therefore to assess Sesotho-speaking 2–3-year-olds’ acquisition of nominal agreement as a function of full versus reduced noun class prefixes. Although the children exhibited early phonological underspecification, they otherwise represented nominal agreement with little problem, whether the noun class prefix was produced or not. The implications for learnability, and the development of lexical representations more generally, are discussed.

Abstract: In this paper it is studied for which oracles A and which types of A-r.e. matroids the class of all A-r.e. closed sets in the matroid is learnable by an unrelativised learner. The learning criteria considered comprise in particular criteria more general than behaviourally correct learning, namely behaviourally correct learning from recursive texts, partial learning and reliably partial learning. For various natural classes of matroids and learning criteria, characterisations of learnability are obtained.

Abstract: A major theoretical debate in language acquisition research regards the learnability of hierarchical structures. The artificial grammar learning methodology is increasingly influential in approaching this question. Studies using an artificial centre-embedded AnBn grammar without semantics draw conflicting conclusions. This study investigates the facilitating effect of distributional biases in simple AB adjacencies in the input sample—caused in natural languages, among others, by semantic biases—on learning a centre-embedded structure. A mathematical simulation of the linguistic input and the learning, comparing various distributional biases in AB pairs, suggests that strong distributional biases might help us to grasp the complex AnBn hierarchical structure in a later stage. This theoretical investigation might contribute to our understanding of how distributional features of the input—including those caused by semantic variation—help learning complex structures in natural languages.

Abstract: Markovian models with hidden state are widely-used formalisms for modeling sequential phenomena. Learnability of these models has been well studied when the sample is given in batch mode, and algorithms with PAC-like learning guarantees exist for specic classes of models such as Probabilistic Deterministic Finite Automata (PDFA). Here we focus on PDFA and give an algorithm for infering models in this class under the stringent data stream scenario: unlike existing methods, our algorithm works incrementally and in one pass, uses memory sublinear in the stream length, and processes input items in amortized constant time. We provide rigorous PAC-like bounds for all of the above, as well as an evaluation on synthetic data showing that the algorithm performs well in practice. Our algorithm makes a key usage of several old and new sketching techniques. In particular, we develop a new sketch for implementing bootstrapping in a streaming setting which may be of independent interest. In experiments we have observed that this sketch yields important reductions in the examples required for performing some crucial statistical tests in our algorithm.

Abstract: Multi-instance multi-label learning (MIML) is a new machine learning framework where one data object is described by multiple instances and associated with multiple class labels. During the past few years, many MIML algorithms have been developed and many applications have been described. However, there lacks theoretical exploration to the learnability of MIML. In this paper, through proving a generalization bound for multi-instance single-label learner and viewing MIML as a number of multi-instance single-label learning subtasks with the correlation among the labels, we show that the MIML hypothesis class constructed from a multi-instance single-label hypothesis class is PAC-learnable.

Abstract: Automatic classes are classes of languages for which a finite automaton can decide the membership problem for the languages in the class, in a uniform way, given an index for the language. For alphabet size of at least 4, every automatic class of erasing pattern languages is contained, for some constant n, in the class of all languages generated by patterns which contain (1) every variable only once and (2) at most n symbols after the first occurrence of a variable. It is shown that such a class is automatically learnable using a learner with the length of the long-term memory being bounded by the length of the first example seen. The study is extended to show the learnability of related classes such as the class of unions of two pattern languages of the above type.

Abstract: This paper presents an interactive multimodal retrieval system featuring multiple search strategies. In contrast to the system-centric perspective often found in multimedia retrieval, we follow a more user-centered approach considering the search as an interactive process. To assist in this process, the discussed system supports directed and exploratory search as well as faceted navigation and a transition between these information seeking strategies.In order to integrate these strategies, a consistent retrieval and interaction model based on the principle of polyrepresentation is developed. To complete the functionality, a preference-based mechanism for graded relevance feedback is presented that overcomes limitations of binary as well as total order-based approaches. To improve the learnability of the system and to give users back the feeling of control over the search process, various visualizations are offered that open paths of communication between the system and the user in order to bridge the gap between the system's notion of the information need and the one of the actual user.

Abstract: We prove that any concept in any description logic that extends $\mathcal{ALC}$ with some features amongst I (inverse), Qk (quantified number restrictions with numbers bounded by a constant k), Self (local reflexivity of a role) can be learnt if the training information system is good enough. That is, there exists a learning algorithm such that, for every concept C of those logics, there exists a training information system consistent with C such that applying the learning algorithm to the system results in a concept equivalent to C.

Abstract: The concept of state is central to dynamical systems. In any timeseries problem—such as filtering, planning or forecasting—models and algorithms summarize important information from the past into some sort of state variable. In this chapter, we start with a broad examination of the concept of state, with emphasis on the fact that there are many possible representations of state for a given dynamical system, each with different theoretical and computational properties. We then focus on models with predictively defined representations of state that represent state as a set of statistics about the short-term future, as opposed to the classic approach of treating state as a latent, unobservable quantity. In other words, the past is summarized into predictions about the actions and observations in the short-term future, which can be used to make further predictions about the infinite future.While this representational idea applies to any dynamical system problem, it is particularly useful in a model-based RL context, when an agent must learn a representation of state and a model of system dynamics online: because the representation (and hence all of the model’s parameters) are defined using only statistics of observable quantities, their learning algorithms are often straightforward and have attractive theoretical properties. Here, we survey the basic concepts of predictively defined representations of state, important auxiliary constructs (such as the systems dynamics matrix), and theoretical results on their representational power and learnability.

Abstract: We describe theoretical bounds and a practical algorithm for teaching a model by demonstration in a sequential decision making environment. Unlike previous efforts that have optimized learners that watch a teacher demonstrate a static policy, we focus on the teacher as a decision maker who can dynamically choose different policies to teach different parts of the environment. We develop several teaching frameworks based on previously defined supervised protocols, such as Teaching Dimension, extending them to handle noise and sequences of inputs encountered in an MDP. We provide theoretical bounds on the learnability of several important model classes in this setting and suggest a practical algorithm for dynamic teaching.

Abstract: We describe theoretical bounds and a practical algorithm for teaching a model by demonstration in a sequential decision making environment. Unlike previous efforts that have optimized learners that watch a teacher demonstrate a static policy, we focus on the teacher as a decision maker who can dynamically choose different policies to teach different parts of the environment. We develop several teaching frameworks based on previously defined supervised protocols, such as Teaching Dimension, extending them to handle noise and sequences of inputs encountered in an MDP.We provide theoretical bounds on the learnability of several important model classes in this setting and suggest a practical algorithm for dynamic teaching.

Abstract: This paper describes research within the ALADIN project, which aims to develop an adaptive, assistive vocal interface for people with a physical impairment. One of the components in this interface is a self-learning grammar module, which maps a user’s utterance to its intended meaning. This paper describes a case study of the learnability of this task on the basis of a corpus of commands for the card game patience. The collection, transcription and annotation of this corpus is outlined in this paper, followed by results of preliminary experiments using a shallow concept-tagging approach. Encouraging results are observed during learning curve experiments, that gauge the minimal amount of training data needed to trigger accurate concept tagging of previously unseen utterances.

Abstract: In this paper we investigate learning of probabilistic subsequential transducers in an active learning environment. In our learning algorithm the learner interacts with an oracle by asking probabilistic queries on the observed data. We prove our algorithm in an identification in the limit model. We also provide experimental evidence to show the correctness and to analyze the learnability of the proposed algorithm.

Abstract: The present work determines the exact nature of linear time computable notions which characterise automatic functions (those whose graphs are recognised by a finite automaton). The paper also determines which type of linear time notions permit full learnability for learning in the limit of automatic classes (families of languages which are uniformly recognised by a finite automaton). In particular it is shown that a function is automatic iff there is a one-tape Turing machine with a left end which computes the function in linear time where the input before the computation and the output after the computation both start at the left end. It is known that learners realised as automatic update functions are restrictive for learning. In the present work it is shown that one can overcome the problem by providing work-tapes additional to a resource-bounded base tape while keeping the update-time to be linear in the length of the largest datum seen so far. In this model, one additional such worktape provides additional learning power over the automatic learner model and the two-work-tape model gives full learning power.

Abstract: We present an interactive tool for browsing course requisites as a case study of dependency visualization. This tool uses multiple interactive visualizations to allow the user to explore the dependencies between courses. A usability study revealed that the proposed browser provides significant advantages over traditional methods, in terms of learnability, efficiency and user confidence. The results are discussed within a general framework for interactive visualization of dependencies.

Abstract: A apreensibilidade e legibilidade de revistas cientificas sao problemas centrais das comissoes editoriais, dos revisores e em especial dos leitores. O objetivo deste trabalho foi avaliar a apreensibilidade e legibilidade de artigos cientificos d a Revista Bragantia, com o intuito de melhorar a qualidade dos artigos cientificos e incentivar o uso dos indices metricos. Para avaliacao da apreensibilidade, foram determinados os valores de Facilidade de Leitura Flesch , Flesch Kincaid e o Indice de Coleman Liau. Para obter um conceito pratico de qualidade na legibilidade dos artigos cientificos , utilizou-se um questionario estruturado, direcionado a 100 academicos do curso de Ciencias Biologicas, abordando a categoria de acordo com a f ormatacao (espacamento, colunas, numeracao e recuo) e tamanho do texto e o tipo e cor das letras . Os maiores valores medios encontrados foram nos anos de 1996-00 para FLF (26,7 ±10,7), 1991-95 para FK (16,74 ±1,8) e 1980-85 para CL (12,64 ±0,85). De acordo com 90% dos entrevistados, o espacamento do texto em todos os anos foi considerado otimo, assim como o tamanho, tipo e cor das letras. Os artigos obtiveram alta qualidade na legibilidade e baixa apreensibilidade, revelando dificil acesso ao publico estudantil em geral, sendo viavel apenas para o publico academico. Palavra chave: Frases. Informacao. Metricas. Qualidade textual. SUMMARY Learnability and legibility of scientific journals are core problems for editorial committees, for reviewers and notably for readers. The objective of this paper is to evaluate learnability and readability of scientific articles in Bragantia Journal aiming at improving the quality of scientific articles and stimulating the use of metric indexes. For learnability evaluation, Flesch reading facilitating values, Kincaid Flesch and Coleman Liau Index were determined. A structured questionnaire was used focusing on 100 Biological Sciences undergraduate students, covering each category according to formatting (spacing, columns, numbering and text board) and text size and letter types and colors. Obtained highest mean values were during 1996-00 for FLF ( 26,7 ±10,7), 1991-95 for FK ( 16,74 ±1,8) and during 1980-85 for CL (12,64±0,85). According to 90% of interviewers, text spacing during all years was considered great as well as letter size, types and colors. Articles were highly evaluated for legibility but were low for learnability showing difficult access to students in general and viable only for those involved within the academic context. Key words: Phrase, information, metric text quality

Abstract: Difficulties with plaintext passwords are well-documented. Many alternative authentication schemes have been proposed, but a key evaluation metric has typically been ignored: learnability. For wide deployment, some form of tutorial is the most assistance users would be provided. This paper presents the results of two user studies (one local and one on Amazon Mechanical Turk) of 134 total participants. Our studies compared four methods of teaching a new authentication scheme: a single page of instructions, a hypertext tutorial with images, an interactive demo, and a video tutorial. As one may expect, demo and video users invested more time in their tutorials than text and hypertext users. We found few differences in the learnability and security between the conditions, but to our surprise, the hypertext and video tutorials resulted in greater password memorability than the text and demo tutorials.