This paper proposes a natural language translation method for machine-verifiable formal proofs that leverages the informalization (verbalization of formal language proof steps) and summarization capabilities of LLMs. For evaluation, it was applied to formal proof data created in accordance with natural language proofs taken from an undergraduate-level textbook, and the quality of the generated natural language proofs was analyzed in comparison with the original natural language proofs. Furthermore, we will demonstrate that this method can output highly readable and accurate natural language proofs by applying it to existing formal proof library of the Lean proof assistant.

Attention weight is a clue to interpret how a Transformer-based model makes an inference.In some attention heads, the attention focuses on the neighbors of each token.This allows the output vector of each token to depend on the surrounding tokens and contributes to make the inference context-dependent.We analyze the mechanism behind the concentration of attention on nearby tokens.We show that the phenomenon emerges as follows: (1) learned position embedding has sinusoid-like components, (2) such components are transmitted to the query and the key in the selfattention, (3) the attention head shifts the phases of the sinusoid-like components so that the attention concentrates on nearby tokens at specific relative positions.In other words, a certain type of Transformer-based model acquires the sinusoidal positional encoding to some extent on its own through Masked Language Modeling.

We have been developing an end-to-end math problem solving system that accepts natural language input.The current paper focuses on how we analyze the problem sentences to produce logical forms.We chose a hybrid approach combining a shallow syntactic analyzer and a manuallydeveloped lexicalized grammar.A feature of the grammar is that it is extensively typed on the basis of a formal ontology for pre-university math.These types are helpful in semantic disambiguation inside and across sentences.Experimental results show that the hybrid system produces a well-formed logical form with 88% precision and 56% recall.

This paper presents a hybrid approach to the verification of statements about historical facts. The test data was collected from the world history examinations in a standardized achievement test for high school students. The data includes various kinds of false statements that were carefully written so as to deceive the students while they can be disproven on the basis of the teaching materials. Our system predicts the truth or falsehood of a statement based on text search, word cooccurrence statistics, factoid-style question answering, and temporal relation recognition. These features contribute to the judgement complementarily and achieved the state-of-the-art accuracy.

A novel method to induce wide-coverage Combinatory Categorial Grammar (CCG) resources for Japanese is proposed in this article. For some languages including English, the availability of large annotated corpora and the development of data-based induction of lexicalized grammar have enabled deep parsing, i.e., parsing based on lexicalized grammars. However, deep parsing for Japanese has not been widely studied. This is mainly because most Japanese syntactic resources are represented in chunk-based dependency structures, while previous methods for inducing grammars are dependent on tree corpora. To translate syntactic information presented in chunk-based dependencies to phrase structures as accurately as possible, integration of annotation from multiple dependency-based corpora is proposed. Our method first integrates dependency structures and predicate-argument information and converts them into phrase structure trees. The trees are then transformed into CCG derivations in a similar way to previously proposed methods. The quality of the conversion is empirically evaluated in terms of the coverage of the obtained CCG lexicon and the accuracy of the parsing with the grammar. While the transforming process used in this study is specialized for Japanese, the framework of our method would be applicable to other languages for which dependency-based analysis has been regarded as more appropriate than phrase structure-based analysis due to morphosyntactic features.

We report on a project aiming at developing a system that solves a wide range of math problems written in natural language. In the system, formal analysis of natural language semantics is coupled with automated reasoning technologies including computer algebra, using logic as their common language. We have developed a prototype system that accepts as its input a linguistically annotated problem text. Using the prototype system as a reference point, we analyzed real university entrance examination problems from the viewpoint of end-to-end automated reasoning. Further, evaluation on entrance exam mock tests revealed that an optimistic estimate of the system’s performance already matches human averages on a few test sets.

Dependency-based Compositional Semantics (DCS) is a framework of natural language semantics with easy-to-process structures as well as strict semantics.In this paper, we equip the DCS framework with logical inference, by defining abstract denotations as an abstraction of the computing process of denotations in original DCS.An inference engine is built to achieve inference on abstract denotations.Furthermore, we propose a way to generate on-the-fly knowledge in logical inference, by combining our framework with the idea of tree transformation.Experiments on FraCaS and PASCAL RTE datasets show promising results.

The “Todai Robot Project (Can a robot get into The University of Tokyo?)” was initiated by the National Institute of Informatics in 2011 as an AI grand challenge. The goal of the project was to create an AI system that answered real questions on university entrance examinations consisting of two parts, i.e., multiple-choice style national standardized tests and a written test that included short essays. The tasks naturally required the development of ground-breaking underlying technologies in research areas including natural language processing, image processing, speech recognition, automated theorem proving, computer algebra, and computer simulations. It simultaneously required interdisciplinary research synthesis.Our software took a (digitalized and annotated version of) mock National Center Test for University Admissions (NCTUA), provided by a prep. school, with more than five thousand students. The results revealed that its abilities were still far below the average scores of entrants to The University of Tokyo. However, it was competent enough to pass the entrance exams of 404 out of 744 private universities in Japan. The rapid rise of new AI technologies may negatively affect the labor market in the short term, and we will need to re-design our education systems that were intended to be optimal for the modern industrial society.

When David Hilbert started so called "Hilbert's program" (formalization of mathematics) in the early 20th century to give a solid foundation to mathematics, he unintentionally introduced the possibility of automatization of mathematics. Theoretically, the possibility was denied by Gödel's incompleteness theorem. However, an interesting issue remains: is "mundane mathematics" automatizable? We are developing a system that solves a wide range of math problems written in natural language, as a part of the Todai Robot Project, an AI challenge to pass the university entrance examination. We give an overview and report on the progress of our project, and the theoretical and methodological difficulties to be overcome.

Many real-world data mining problems contain hidden information (e.g., unobservable latent dependencies). We propose a perceptron-style method, latent structured perceptron, for fast discriminative learning of structured classification with hidden information. We also give theoretical analysis and demonstrate good convergence properties of the proposed method. Our method extends the perceptron algorithm for the learning task with hidden information, which can be hardly captured by traditional models. It relies on Viterbi decoding over latent variables, combined with simple additive updates. We perform experiments on one synthetic data set and two real-world structured classification tasks. Compared to conventional nonlatent models (e.g., conditional random fields, structured perceptrons), our method is more accurate on real-world tasks. Compared to existing heavy probabilistic models of latent variables (e.g., latent conditional random fields), our method lowers the training cost significantly (almost one order magnitude faster) yet with comparable or even superior classification accuracy. In addition, experiments demonstrate that the proposed method has good scalability on large-scale problems.

BACKGROUND: Current research has shown that major difficulties in event extraction for the biomedical domain are traceable to coreference. Therefore, coreference resolution is believed to be useful for improving event extraction. To address coreference resolution in molecular biology literature, the Protein Coreference (COREF) task was arranged in the BioNLP Shared Task (BioNLP-ST, hereafter) 2011, as a supporting task. However, the shared task results indicated that transferring coreference resolution methods developed for other domains to the biological domain was not a straight-forward task, due to the domain differences in the coreference phenomena. RESULTS: We analyzed the contribution of domain-specific information, including the information that indicates the protein type, in a rule-based protein coreference resolution system. In particular, the domain-specific information is encoded into semantic classification modules for which the output is used in different components of the coreference resolution. We compared our system with the top four systems in the BioNLP-ST 2011; surprisingly, we found that the minimal configuration had outperformed the best system in the BioNLP-ST 2011. Analysis of the experimental results revealed that semantic classification, using protein information, has contributed to an increase in performance by 2.3% on the test data, and 4.0% on the development data, in F-score. CONCLUSIONS: The use of domain-specific information in semantic classification is important for effective coreference resolution. Since it is difficult to transfer domain-specific information across different domains, we need to continue seek for methods to utilize such information in coreference resolution.

Abstract As described in this paper, we specifically examine the structural learning problem of a supertagging task. Supertagging is a task to assign the most probable lexical entry to each word in a sentence. A supertagger is extremely important for a lexicalized grammar parser because an accurate supertagger can greatly reduce lexical ambiguity in downstream parser. Supertagging is more challenging than conventional sequence labeling tasks (e.g., part-of-speech tagging). First, the supertags are numerous. Supertags are the lexical entries defined in a lexicalized grammar, which consists of rich syntactic/semantic information. Second, the inter-supertag relation is more complex. A proper supertag assignment is expected to be compatible with other supertag assignments in a sentence to construct a parse tree. Commonly used adjacent label features (e.g., first-order edge feature) in a sequence labeling model are too rough for the supertagging task. Long-range information is extremely important for the supertagging task. Two approaches to consider long-range information in a supertagger's training stage are proposed. Specifically, we propose a dependency-informed supertagger to use word-to-word dependency derived from a dependency parser and generate long-range features as soft constraints in the training. In the forest-guided supertagger, we constrain the classifier to learn in a grammar-satisfying space and use a CFG filter to impose grammar constraints for the update of model parameters. The experiments show that the proposed structure-guided supertaggers perform significantly better than the baseline supertaggers. Based on the improved supertaggers, theF-score of the final parser is also improved. Using the forest-guided supertagger in a shift-reduce HPSG parser, we achieved a competitive parsing performance of 89.31%F-score with higher parsing speed than that of a state-of-the-art HPSG parser.

Bilingual dictionaries can be automatically extended by new translations using comparable corpora. The general idea is based on the assumption that similar words have similar contexts across languages. However, previous studies have mainly focused on Indo-European languages, or use only a bag-of-words model to describe the context. Furthermore, we argue that it is helpful to extract only the statistically significant context, instead of using all context. The present approach addresses these issues in the following manner. First, based on the context of a word with an unknown translation (query word), we extract salient pivot words. Pivot words are words for which a translation is already available in a bilingual dictionary. For the extraction of salient pivot words, we use a Bayesian estimation of the point-wise mutual information to measure statistical significance. In the second step, we match these pivot words across languages to identify translation candidates for the query word. We therefore calculate a similarity score between the query word and a translation candidate using the probability that the same pivots will be extracted for both the query word and the translation candidate. The proposed method uses several context positions, namely, a bag-of-words of one sentence, and the successors, predecessors, and siblings with respect to the dependency parse tree of the sentence. In order to make these context positions comparable across Japanese and English, which are unrelated languages, we use several heuristics to adjust the dependency trees appropriately. We demonstrate that the proposed method significantly increases the accuracy of word translations, as compared to previous methods.

Abstract Many parsing techniques assume the use of a packed parse forest to enable efficient and accurate parsing. However, they suffer from an inherent problem that derives from the restriction of locality in the packed parse forest. Deterministic parsing is one solution that can achieve simple and fast parsing without the mechanisms of the packed parse forest by accurately choosing search paths. We propose new deterministic shift-reduce parsing and its variants for unification-based grammars. Deterministic parsing cannot simply be applied to unification-based grammar parsing, which often fails because of its hard constraints. Therefore, this is developed by using default unification, which almost always succeeds in unification by overwriting inconsistent constraints in grammars.

This paper describes the rational behind designing workflow systems based on the Unix make by showing a number of idioms useful for workflows comprising many tasks. It also demonstrates a specific design and implementation of such a workflow system called GXP make. GXP make supports all the features of GNU make and extends its platforms from single node systems to clusters, clouds, supercomputers, and distributed systems. Interestingly, it is achieved by a very small code base that does not modify GNU make implementation at all. While being not ideal for performance, it achieved a useful performance and scalability of dispatching one million tasks in approximately 16,000 seconds (60 tasks per second, including dependence analysis) on an 8 core Intel Nehalem node. For real applications, recognition and classification of protein-protein interactions from biomedical texts on a supercomputer with more than 8,000 cores are described.

On large datasets, the popular training approach has been stochastic gradient descent (SGD). This paper proposes a modification of SGD, called averaged SGD with feedback (ASF), that significantly improves the performance (robustness, accuracy, and training speed) over the traditional SGD. The proposal is based on three simple ideas: averaging the weight vectors across SGD iterations, feeding the averaged weights back into the SGD update process, and deciding when to perform the feedback (linearly slowing down feedback). Theoretically, we demonstrate the reasonable convergence properties of the ASF. Empirically, the ASF outperforms several strong baselines in terms of accuracy, robustness over the noise, and the training speed. To our knowledge, this is the first study of ``feedback'' in stochastic gradient learning. Although we choose latent conditional models for verifying the ASF in this paper, the ASF is a general purpose technique just like SGD, and can be directly applied to other models.

MOTIVATION: Metabolic and signaling pathways are an increasingly important part of organizing knowledge in systems biology. They serve to integrate collective interpretations of facts scattered throughout literature. Biologists construct a pathway by reading a large number of articles and interpreting them as a consistent network, but most of the models constructed currently lack direct links to those articles. Biologists who want to check the original articles have to spend substantial amounts of time to collect relevant articles and identify the sections relevant to the pathway. Furthermore, with the scientific literature expanding by several thousand papers per week, keeping a model relevant requires a continuous curation effort. In this article, we present a system designed to integrate a pathway visualizer, text mining systems and annotation tools into a seamless environment. This will enable biologists to freely move between parts of a pathway and relevant sections of articles, as well as identify relevant papers from large text bases. The system, PathText, is developed by Systems Biology Institute, Okinawa Institute of Science and Technology, National Centre for Text Mining (University of Manchester) and the University of Tokyo, and is being used by groups of biologists from these locations.

In the recent decades, high-throughput screening methods were established, bringing forth major breakthroughs in the fields of molecular biology and biomedicine. Since researchers in these fields need to interpret an enormous quantity of data and the publication rates of scientific articles are exploding, demands on text mining technology are growing with each passing year.

Currently, relation extraction (RE) and event extraction (EE) are the two main streams of biological information extraction. In 2009, the majority of these RE and EE research efforts were centered around the BioCreative II.5 Protein-Protein Interaction (PPI) challenge and the "BioNLP event extraction shared task." Although these challenges took somewhat different approaches, they share the same ultimate goal of extracting bio-knowledge from the literature. This paper compares the two challenge task definitions, and presents a unified system that was successfully applied in both these and several other PPI extraction task settings. The AkaneRE system has three parts: A core engine for RE, a pool of modules for specific solutions, and a configuration language to adapt the system to different tasks. The core engine is based on machine learning, using either Support Vector Machines or Statistical Classifiers and features extracted from given training data. The specific modules solve tasks like sentence boundary detection, tokenization, stemming, part-of-speech tagging, parsing, named entity recognition, generation of potential relations, generation of machine learning features for each relation, and finally, assignment of confidence scores and ranking of candidate relations. With these components, the AkaneRE system produces state-of-the-art results, and the system is freely available for academic purposes at http://www-tsujii.is.s.u-tokyo.ac.jp/satre/akane/.

MOTIVATION: While text mining technologies for biomedical research have gained popularity as a way to take advantage of the explosive growth of information in text form in biomedical papers, selecting appropriate natural language processing (NLP) tools is still difficult for researchers who are not familiar with recent advances in NLP. This article provides a comparative evaluation of several state-of-the-art natural language parsers, focusing on the task of extracting protein-protein interaction (PPI) from biomedical papers. We measure how each parser, and its output representation, contributes to accuracy improvement when the parser is used as a component in a PPI system. RESULTS: All the parsers attained improvements in accuracy of PPI extraction. The levels of accuracy obtained with these different parsers vary slightly, while differences in parsing speed are larger. The best accuracy in this work was obtained when we combined Miyao and Tsujii's Enju parser and Charniak and Johnson's reranking parser, and the accuracy is better than the state-of-the-art results on the same data. AVAILABILITY: The PPI extraction system used in this work (AkanePPI) is available online at http://www-tsujii.is.s.u-tokyo.ac.jp/downloads/downloads.cgi. The evaluated parsers are also available online from each developer's site.

BACKGROUND: Associating literature with pathways poses new challenges to the Text Mining (TM) community. There are three main challenges to this task: (1) the identification of the mapping position of a specific entity or reaction in a given pathway, (2) the recognition of the causal relationships among multiple reactions, and (3) the formulation and implementation of required inferences based on biological domain knowledge. RESULTS: To address these challenges, we constructed new resources to link the text with a model pathway; they are: the GENIA pathway corpus with event annotation and NF-kB pathway. Through their detailed analysis, we address the untapped resource, 'bio-inference,' as well as the differences between text and pathway representation. Here, we show the precise comparisons of their representations and the nine classes of 'bio-inference' schemes observed in the pathway corpus. CONCLUSIONS: We believe that the creation of such rich resources and their detailed analysis is the significant first step for accelerating the research of the automatic construction of pathway from text.

This paper defines a generative probabilistic model of parse trees, which we call PCFG-LA. This model is an extension of PCFG in which non-terminal symbols are augmented with latent variables. Fine-grained CFG rules are automatically induced from a parsed corpus by training a PCFG-LA model using an EM-algorithm. Because exact parsing with a PCFG-LA is NP-hard, several approximations are described and empirically compared. In experiments using the Penn WSJ corpus, our automatically trained model gave a performance of 86.6% (F1, sentences ≤ 40 words), which is comparable to that of an unlexicalized PCFG parser created using extensive manual feature selection.