Search

Typical arguments for results like Kleene's Second Recursion Theorem and the existence of self-writing computer programs bear the fingerprints of equational reasoning and combinatory logic. In fact, the connection of combinatory logic and computability theory is very old, and this paper extends this connection in new ways. In one direction, we counter the main trend in both computability theory and combinatory logic of heading straight to undecidability. Instead, this paper proposes using several very small equational logics to examine results in computability theory itself. These logics are decidable via term rewriting. We argue that they have something interesting to say about computability theory. They are closely related to fragments of combinatory logic which are decidable, and so this paper contributes to the study of such fragments. The paper has a few surprising results such as a classification of quine programs (programs which output themselves) in two decidable fragments. The classification goes via examination of normal forms in term rewriting systems, hence the title of the paper. The classification is an explanation of why all quine programs (in any language) are "pretty much the same, except for inessential details." In addition, we study the relational structure whose objects are the programs with the relation "p expresses q" meaning that if the program p is run on nothing, then it eventually outputs the program q., Comment: Presented to the 39th Conference on the Mathematical Foundations of Programming Semantics (MFPS XXXIX). MFPS took place together at Indiana University Bloomington on June 19-23, 2023

We extract mathematical concepts from mathematical text using generative large language models (LLMs) like ChatGPT, contributing to the field of automatic term extraction (ATE) and mathematical text processing, and also to the study of LLMs themselves. Our work builds on that of others in that we aim for automatic extraction of terms (keywords) in one mathematical field, category theory, using as a corpus the 755 abstracts from a snapshot of the online journal "Theory and Applications of Categories", circa 2020. Where our study diverges from previous work is in (1) providing a more thorough analysis of what makes mathematical term extraction a difficult problem to begin with; (2) paying close attention to inter-annotator disagreements; (3) providing a set of guidelines which both human and machine annotators could use to standardize the extraction process; (4) introducing a new annotation tool to help humans with ATE, applicable to any mathematical field and even beyond mathematics; (5) using prompts to ChatGPT as part of the extraction process, and proposing best practices for such prompts; and (6) raising the question of whether ChatGPT could be used as an annotator on the same level as human experts. Our overall findings are that the matter of mathematical ATE is an interesting field which can benefit from participation by LLMs, but LLMs themselves cannot at this time surpass human performance on it., Comment: 13 pages, 4 figures, presented to the 14th MathUI Workshop 2023

We forge connections between the theory of fractal sets obtained as attractors of iterated function systems and process calculi. To this end, we reinterpret Milner's expressions for processes as contraction operators on a complete metric space. When the space is, for example, the plane, the denotations of fixed point terms correspond to familiar fractal sets. We give a sound and complete axiomatization of fractal equivalence, the congruence on terms consisting of pairs that construct identical self-similar sets in all interpretations. We further make connections to labelled Markov chains and to invariant measures. In all of this work, we use important results from process calculi. For example, we use Rabinovich's completeness theorem for trace equivalence in our own completeness theorem. In addition to our results, we also raise many questions related to both fractals and process calculi., Comment: Expanded and edited into a journal version. Submitted to the CALCO 2023 special issue of LMCS

The Vietoris space of compact subsets of a given Hausdorff space yields an endofunctor $\mathscr V$ on the category of Hausdorff spaces. Vietoris polynomial endofunctors on that category are built from $\mathscr V$, the identity and constant functors by forming products, coproducts and compositions. These functors are known to have terminal coalgebras and we deduce that they also have initial algebras. We present an analogous class of endofunctors on the category of extended metric spaces, using in lieu of $\mathscr V$ the Hausdorff functor $\mathcal H$. We prove that the ensuing Hausdorff polynomial functors have terminal coalgebras and initial algebras. Whereas the canonical constructions of terminal coalgebras for Vietoris polynomial functors takes $\omega$ steps, one needs $\omega + \omega$ steps in general for Hausdorff ones. We also give a new proof that the closed set functor on metric spaces has no fixed points.

We consider dynamic versions of epistemic logic as formulated in Baltag and Moss "Logics for epistemic programs" (2004). That paper proposed a logical language (actually families of languages parameterized by action signatures) for dynamic epistemic logic. It had been shown that validity in the language is Pi-1-1-complete, so there are no recursively axiomatized complete logical systems for it. In contrast, this paper proves a weak completeness result for the fragment without action iteration, and a strong completeness result for the fragment without action iteration and common knowledge. Our work involves a detour into term rewriting theory. The argument uses modal filtration, and thus we obtain the finite model property and hence decidability. We also give a translation of our largest language into PDL, thereby obtaining a second proof of decidability. The paper closes with some results on expressive power. These are mostly concerned with comparing the action-iteration-free language with modal logic augmented by transitive closure operators. We answer a natural question about the languages we obtain by varying the action signature: we prove that a logical language with operators for private announcements is more expressive than one for public announcements., Comment: This paper was intended to be the "journal" version of our 1998 paper "The logic of common knowledge, public announcements,and private suspicions." It was mainly written in 2004, with a few bibliographic additions coming a few years later

This paper obtains a completeness result for inequational reasoning with applicative terms without variables in a setting where the intended semantic models are the full structures, the full type hierarchies over preorders for the base types. The syntax allows for the specification that a given symbol be interpreted as a monotone function, or an antitone function, or both. There is a natural set of five rules for inequational reasoning. One can add variables and also add a substitution rule, but we observe that this logic would be incomplete for full structures. This is why the completeness result in this paper pertains to terms without variables. Since the completeness is already known for the class of general (Henkin) structures, we are interested in full structures. We present a completeness theorem. Our result is not optimal because we restrict to base preorders which have a weak completeness property: every pair of elements has an upper bound and a lower bound. To compensate we add several rules to the logic. We also present extensions and variations of our completeness result., Comment: 23 pages

This paper studies presentations of the Sierpinski gasket as a final coalgebra for functors on several categories of metric spaces with additional designated points. The three categories which we study differ on their morphisms: one uses short (non-expanding) maps, a second uses Lipschitz maps, and a third uses continuous maps. The functor in all cases is very similar to what we find in the standard presentation of the gasket as an attractor. We prove that the Sierpinski gasket itself is the final coalgebra of a naturally-occurring functor in the continuous setting. In the short setting, the final coalgebra exists but it is better described as the completion of the initial algebra, and this is not isomorphic to the Sierpinski gasket. In the Lipschitz setting, the final coalgebra does not exist. We determine the initial algebras in all three settings as well., Comment: 21 pages

We advance the program of connections between final coalgebras as sources of circularity in mathematics and fractal sets of real numbers. In particular, we are interested in the Sierpinski carpet, taking it as a fractal subset of the unit square. We construct a category of square sets and an endofunctor on it which corresponds to the operation of gluing copies of a square set along segments. We show that the initial algebra and final coalgebra exist for our functor, and that the final coalgebra is bi-Lipschitz equivalent to the Sierpinski carpet. Along the way, we make connections to topics such as the iterative construction of initial algebras as colimits, corecursive algebras, and the classic treatment of fractal sets due to Hutchinson., Comment: In Proceedings ACT 2021, arXiv:2211.01102

Deep learning (DL) based language models achieve high performance on various benchmarks for Natural Language Inference (NLI). And at this time, symbolic approaches to NLI are receiving less attention. Both approaches (symbolic and DL) have their advantages and weaknesses. However, currently, no method combines them in a system to solve the task of NLI. To merge symbolic and deep learning methods, we propose an inference framework called NeuralLog, which utilizes both a monotonicity-based logical inference engine and a neural network language model for phrase alignment. Our framework models the NLI task as a classic search problem and uses the beam search algorithm to search for optimal inference paths. Experiments show that our joint logic and neural inference system improves accuracy on the NLI task and can achieve state-of-art accuracy on the SICK and MED datasets., Comment: 8 pages, 4 figures, The 10th Joint Conference on Lexical and Computational Semantics (*SEM2021) @ ACL2021

The Initial Algebra Theorem by Trnkov\'a et al.~states, under mild assumptions, that an endofunctor has an initial algebra provided it has a pre-fixed point. The proof crucially depends on transfinitely iterating the functor and in fact shows that, equivalently, the (transfinite) initial-algebra chain stops. We give a constructive proof of the Initial Algebra Theorem that avoids transfinite iteration of the functor. For a given pre-fixed point $A$ of the functor, it uses Pataraia's theorem to obtain the least fixed point of a monotone function on the partial order formed by all subobjects of $A$. Thanks to properties of recursive coalgebras, this least fixed point yields an initial algebra. We obtain new results on fixed points and initial algebras in categories enriched over directed-complete partial orders, again without iteration. Using transfinite iteration we equivalently obtain convergence of the initial-algebra chain as an equivalent condition, overall yielding a streamlined version of the original proof.

Despite the tremendous recent progress on natural language inference (NLI), driven largely by large-scale investment in new datasets (e.g., SNLI, MNLI) and advances in modeling, most progress has been limited to English due to a lack of reliable datasets for most of the world's languages. In this paper, we present the first large-scale NLI dataset (consisting of ~56,000 annotated sentence pairs) for Chinese called the Original Chinese Natural Language Inference dataset (OCNLI). Unlike recent attempts at extending NLI to other languages, our dataset does not rely on any automatic translation or non-expert annotation. Instead, we elicit annotations from native speakers specializing in linguistics. We follow closely the annotation protocol used for MNLI, but create new strategies for eliciting diverse hypotheses. We establish several baseline results on our dataset using state-of-the-art pre-trained models for Chinese, and find even the best performing models to be far outpaced by human performance (~12% absolute performance gap), making it a challenging new resource that we hope will help to accelerate progress in Chinese NLU. To the best of our knowledge, this is the first human-elicited MNLI-style corpus for a non-English language., Comment: Findings of EMNLP 2020

This paper studies fundamental questions concerning category-theoretic models of induction and recursion. We are concerned with the relationship between well-founded and recursive coalgebras for an endofunctor. For monomorphism preserving endofunctors on complete and well-powered categories every coalgebra has a well-founded part, and we provide a new, shorter proof that this is the coreflection in the category of all well-founded coalgebras. We present a new more general proof of Taylor's General Recursion Theorem that every well-founded coalgebra is recursive, and we study under which hypothesis the converse holds. In addition, we present a new equivalent characterization of well-foundedness: a coalgebra is well-founded iff it admits a coalgebra-to-algebra morphism to the initial algebra.

We present a new logic-based inference engine for natural language inference (NLI) called MonaLog, which is based on natural logic and the monotonicity calculus. In contrast to existing logic-based approaches, our system is intentionally designed to be as lightweight as possible, and operates using a small set of well-known (surface-level) monotonicity facts about quantifiers, lexical items and tokenlevel polarity information. Despite its simplicity, we find our approach to be competitive with other logic-based NLI models on the SICK benchmark. We also use MonaLog in combination with the current state-of-the-art model BERT in a variety of settings, including for compositional data augmentation. We show that MonaLog is capable of generating large amounts of high-quality training data for BERT, improving its accuracy on SICK., Comment: accepted to SCIL 2020

Do state-of-the-art models for language understanding already have, or can they easily learn, abilities such as boolean coordination, quantification, conditionals, comparatives, and monotonicity reasoning (i.e., reasoning about word substitutions in sentential contexts)? While such phenomena are involved in natural language inference (NLI) and go beyond basic linguistic understanding, it is unclear the extent to which they are captured in existing NLI benchmarks and effectively learned by models. To investigate this, we propose the use of semantic fragments---systematically generated datasets that each target a different semantic phenomenon---for probing, and efficiently improving, such capabilities of linguistic models. This approach to creating challenge datasets allows direct control over the semantic diversity and complexity of the targeted linguistic phenomena, and results in a more precise characterization of a model's linguistic behavior. Our experiments, using a library of 8 such semantic fragments, reveal two remarkable findings: (a) State-of-the-art models, including BERT, that are pre-trained on existing NLI benchmark datasets perform poorly on these new fragments, even though the phenomena probed here are central to the NLI task. (b) On the other hand, with only a few minutes of additional fine-tuning---with a carefully selected learning rate and a novel variation of "inoculation"---a BERT-based model can master all of these logic and monotonicity fragments while retaining its performance on established NLI benchmarks., Comment: AAAI camera-ready version

This is the proceedings of the Seventeenth conference on Theoretical Aspects of Rationality and Knowledge, 17-19 July 2019, Institut de Recherche en Informatique de Toulouse (IRIT), Toulouse University Toulouse, France. The mission of the TARK conferences is to bring together researchers from a wide variety of fields, including Artificial Intelligence, Cryptography, Distributed Computing, Economics and Game Theory, Linguistics, Philosophy, and Psychology, in order to further our understanding of interdisciplinary issues involving reasoning about rationality and knowledge.

This paper explores relational syllogistic logics, a family of logical systems related to reasoning about relations in extensions of the classical syllogistic. These are all decidable logical systems. We prove completeness theorems and complexity results for a natural subfamily of relational syllogistic logics, parametrized by constructors for terms and for sentences., Comment: Final version

This paper enlarges classical syllogistic logic with assertions having to do with comparisons between the sizes of sets. So it concerns a logical system whose sentences are of the following forms: {\sf All $x$ are $y$} and {\sf Some $x$ are $y$}, {\sf There are at least as many $x$ as $y$}, and {\sf There are more $x$ than $y$}. Here $x$ and $y$ range over subsets (not elements) of a given \emph{infinite} set. Moreover, $x$ and $y$ may appear complemented (i.e., as $\overset{-}{x}$ and $\overset{-}{y}$), with the natural meaning. We formulate a logic for our language that is based on the classical syllogistic. The main result is a soundness/completeness theorem. There are efficient algorithms for proof search and model construction., Comment: 28 pages, under review in The Review of Symbolic Logic

A majority digraph is a finite simple digraph $G=(V,\to)$ such that there exist finite sets $A_v$ for the vertices $v\in V$ with the following property: $u\to v$ if and only if "more than half of the $A_u$ are $A_v$". That is, $u\to v$ if and only if $ |A_u \cap A_v | > \frac{1}{2} \cdot |A_u|$. We characterize the majority digraphs as the digraphs with the property that every directed cycle has a reversal. If we change $\frac{1}{2}$ to any real number $\alpha\in (0,1)$, we obtain the same class of digraphs. We apply the characterization result to obtain a result on the logic of assertions "most $X$ are $Y$" and the standard connectives of propositional logic., Comment: 15 pages, 2 figures

We study finite state transduction of automatic and morphic sequences. Dekking proved that morphic sequences are closed under transduction and in particular morphic images. We present a simple proof of this fact, and use the construction in the proof to show that non-erasing transductions preserve a condition called alpha-substitutivity. Roughly, a sequence is alpha-substitutive if the sequence can be obtained as the limit of iterating a substitution with dominant eigenvalue alpha. Our results culminate in the following fact: for multiplicatively independent real numbers alpha and beta, if v is an alpha-substitutive sequence and w is a beta-substitutive sequence, then v and w have no common non-erasing transducts except for the ultimately periodic sequences. We rely on Cobham's theorem for substitutions, a recent result of Durand.

Terminal coalgebras for a functor serve as semantic domains for state-based systems of various types. For example, behaviors of CCS processes, streams, infinite trees, formal languages and non-well-founded sets form terminal coalgebras. We present a uniform account of the semantics of recursive definitions in terminal coalgebras by combining two ideas: (1) abstract GSOS rules l specify additional algebraic operations on a terminal coalgebra; (2) terminal coalgebras are also initial completely iterative algebras (cias). We also show that an abstract GSOS rule leads to new extended cia structures on the terminal coalgebra. Then we formalize recursive function definitions involving given operations specified by l as recursive program schemes for l, and we prove that unique solutions exist in the extended cias. From our results it follows that the solutions of recursive (function) definitions in terminal coalgebras may be used in subsequent recursive definitions which still have unique solutions. We call this principle modularity. We illustrate our results by the five concrete terminal coalgebras mentioned above, e.\,g., a finite stream circuit defines a unique stream function.

For endofunctors of varieties preserving intersections, a new description of the final coalgebra and the initial algebra is presented: the former consists of all well-pointed coalgebras. These are the pointed coalgebras having no proper subobject and no proper quotient. The initial algebra consists of all well-pointed coalgebras that are well-founded in the sense of Osius and Taylor. And initial algebras are precisely the final well-founded coalgebras. Finally, the initial iterative algebra consists of all finite well-pointed coalgebras. Numerous examples are discussed e.g. automata, graphs, and labeled transition systems.

We begin by discussing the history of quantum logic, dividing it into three eras or lives. The first life has to do with Birkhoff and von Neumann's algebraic approach in the 1930's. The second life has to do with attempt to understand quantum logic as logic that began in the late 1950's and blossomed in the 1970's. And the third life has to do with recent developments in quantum logic coming from its connections to quantum computation. We discuss our own work connecting quantum logic to quantum computation (viewing quantum logic as the logic of quantum registers storing qubits), and make some speculations about mathematics based on quantum principles., Comment: To appear in a special issue of Journal of Philosophical Logic

We consider infinite sequences of symbols, also known as streams, and the decidability question for equality of streams defined in a restricted format. This restricted format consists of prefixing a symbol at the head of a stream, of the stream function `zip', and recursion variables. Here `zip' interleaves the elements of two streams in alternating order, starting with the first stream. For example, the Thue-Morse sequence is obtained by the `zip-specification' {M = 0 : X, X = 1 : zip(X,Y), Y = 0 : zip(Y,X)}. Our analysis of such systems employs both term rewriting and coalgebraic techniques. We establish decidability for these zip-specifications, employing bisimilarity of observation graphs based on a suitably chosen cobasis. The importance of zip-specifications resides in their intimate connection with automatic sequences. We establish a new and simple characterization of automatic sequences. Thus we obtain for the binary zip that a stream is 2-automatic iff its observation graph using the cobasis (hd,even,odd) is finite. The generalization to zip-k specifications and their relation to k-automaticity is straightforward. In fact, zip-specifications can be perceived as a term rewriting syntax for automatic sequences. Our study of zip-specifications is placed in an even wider perspective by employing the observation graphs in a dynamic logic setting, leading to an alternative characterization of automatic sequences. We further obtain a natural extension of the class of automatic sequences, obtained by `zip-mix' specifications that use zips of different arities in one specification. We also show that equivalence is undecidable for a simple extension of the zip-mix format with projections like even and odd. However, it remains open whether zip-mix specifications have a decidable equivalence problem.

This paper provides a general account of the notion of recursive program schemes, studying both uninterpreted and interpreted solutions. It can be regarded as the category-theoretic version of the classical area of algebraic semantics. The overall assumptions needed are small indeed: working only in categories with "enough final coalgebras" we show how to formulate, solve, and study recursive program schemes. Our general theory is algebraic and so avoids using ordered, or metric structures. Our work generalizes the previous approaches which do use this extra structure by isolating the key concepts needed to study substitution in infinite trees, including second-order substitution. As special cases of our interpreted solutions we obtain the usual denotational semantics using complete partial orders, and the one using complete metric spaces. Our theory also encompasses implicitly defined objects which are not usually taken to be related to recursive program schemes. For example, the classical Cantor two-thirds set falls out as an interpreted solution (in our sense) of a recursive program scheme., Comment: this version includes the corrections from the corrigendum in Theoret. Comput. Sci. 403 (2008), 409-415

The Aristotelian syllogistic cannot account for the validity of many inferences involving relational facts. In this paper, we investigate the prospects for providing a relational syllogistic. We identify several fragments based on (a) whether negation is permitted on all nouns, including those in the subject of a sentence; and (b) whether the subject noun phrase may contain a relative clause. The logics we present are extensions of the classical syllogistic, and we pay special attention to the question of whether reductio ad absurdum is needed. Thus our main goal is to derive results on the existence (or non-existence) of syllogistic proof systems for relational fragments. We also determine the computational complexity of all our fragments.

We explore a connection between quantum logic and quantum computing., Comment: To appear in JSL