Natural logic offers a powerful relational conception of meaning that is a natural counterpart to distributed semantic representations, which have proven valuable in a wide range of sophisticated language tasks. However, it remains an open question whether it is possible to train distributed representations to support the rich, diverse logical reasoning captured by natural logic. We address this question using two neural network-based models for learning embeddings: plain neural networks and neural tensor networks. Our experiments evaluate the models' ability to learn the basic algebra of natural logic relations from simulated data and from the WordNet noun graph. The overall positive results are promising for the future of learned distributed representations in the applied modeling of logical semantics.

How can techniques drawn from machine learning be appliedto the learning of structured, compositional representations? In this work, we adopt functional programs as our representation, and cast the problem of learning symbolic representations as a symbolic analog of dimensionality reduction. By placing program synthesis within a probabilistic machinelearning framework, we are able to model the learning ofsome English inflectional morphology and solve a set of synthetic regression problems.

In relational databases, relations between objects, represented by binary matrices or tensors, may be arbitrarily complex. In practice however, there are recurring relational patterns such as transitive, permutation and sequential relationships, that seem to have a regular structure not captured by the classical notion of matrix rank or tensor rank. In this paper, we show that factorizing the relational tensor using a logistic or hinge loss instead of the more standard squared loss is more appropriate because it can accurately model many common relations with a fixed-size embedding that depends sub-linearly on the number of entities in the knowledge base. We illustrate this fact empirically by being able to efficiently predict missing links in several synthetic and real-world experiments. Further, we provide theoretical justification for logistic loss by studying its connection to a complexity measure from the field of information complexity called the sign rank. Sign rank is a more appropriate complexity measure as it has a low value for transitive, permutation, or sequential relationships, while being large for uniformly sampled binary matrices/tensors with a high probability.

This paper aims at developing the ambient intelligence sleep system that can derive a good sleep by providing a personally adapted sound. For this purpose, this paper explores the sounds that have a potential of deriving a good sleep and investigates their effect from the several viewpoints (e.g., the sleep latent time). To promote a good sleep, this paper focuses on heartbeat and respiration which are related to a sleep (i.e., its rate decreases as falling asleep) and proposes the ambient intelligent sleep system that provides the sound adjusted to the heartbeat and/or respiration rates, which are automatically measured by the piezoelectric-based mattress sensor without connecting any devices to human’s body. The human subjective experiments of the six subjects for a nap case and the seven subjects for a night sleep case have revealed the following implications: (1) the new wave sound adjusted to both the heartbeat rate (x 1.05) and respiration rate (x 1.05) can shorten the sleep latent time in a nap case in comparison with no sound or the other four types of the sounds; (2) the combination of the two sound sources (adjusted by the heartbeat and respiration rates) contributes to shortening the sleep latent time in comparison with one sound source; (3) the new wave sound can shorten not only the sleep latent time but also the Non-REM3 latent time in a night sleep case in comparison with no sound; and (4) the new wave sound can keep not only an appropriate sleep cycle but also the very similar sleep cycle from the Non-REM to the next one in a night sleep case in comparison with no sound.

As in-vehicle automation becomes increasingly prevalent and capable, there will be more opportunity for vehicle drivers to delegate control to automated systems. as well as increased ability for automated systems to intervene to increase road safety. With the decline in how much a driver must be engaged, two problems arise: driver disengagement and reduced ability to act when necessary; and also a likely decrease in active driving, which may reduce the engagement a driver can have for the purpose of enjoyment. As vehicles become more intelligent, they need to work collaboratively with human drivers, in the frame of a joint-cognitive system in order to both extend and backstop human capabilities to optimize safety, comfort, and engagement.

Our goal in this paper is to achieve automatic extraction and classification of key phrases from software development documents, such as requirements, specifications, and so on. In software development projects, creating dictionaries is important for defining the terminologies used to enable accurate communication between customers and vendors, as well as among developers.However, each target domain, such as a medical, financial, transportational, or other field, has its own particular terminology; moreover, each customer employs its own terms and their respective meanings. Building a dictionary of a target domain requires experts' knowledge in the given domain and considerable effort. To assist in dictionary building, we are developing a software document terminology recognizer (SDTR) with the use of named entity recognition (NER) methods. A significant amount of research exists on NER; however, most of it is focused on general named entities, such as person names, or biological domain named entities, such as names of compounds. However, the problem of building effective entity recognizers in a new domain where you have very little supervised data available is very understudied. There are a lot of small domains each of them has different terminology because software is used in various domains and organizations. Also it is impractical to build taggers by traditional supervised NER methods for SDTR because the tuning cost in individual software development projects is limited.Building method of an SDTR should cover cross-domain terminologies using small size of corpus;nevertheless, an SDTR must cope with very specific terminologies for individual projects. In this paper, we propose a multi-layered SDTR system consisting of an identifier that uses general features based on the probability of phrases and spelling conventions, and an identifier that employs a temporary dictionary automatically built into the general feature identifier. Currently, our prototype achieves a greater than 0.8 F1-value on a small software development project corpus.

The increase of dementia patients is one of the problems caused by aged population not only in Japan but in many developed countries. As cognitive enhancement method for prevention of dementia, coimagination method is proposed: designed group conversation whose themes are selected from recent topics for training of recent episodic memory functions, since recent episodic memory functions decline before the onset of dementia. It is known that people who are disuse particular cognitive functions have higher risk of loosing the functions. However, the participants of the conversations supported by coimagination method sometimes refer to past topics rather than recent topics. The method is required for analyzing whether the topics deal with recent or past for effective intervention, which has not been established. Purpose of this study is to propose method for analyzing the temporal characteristics of the topics during conversation. Mental time travel, or chronesthesia, is the ability to be aware of one’s present, past or future which has been evolved in humans in particular. In order to estimate the mental time of the speaker from topics during conversation supported by coimagination method, we propose mental time estimation method by analyzing tenses and senses. We applied the method to the scripts of conversation supported by coimagination method. The result suggests that it’s possible to estimate mental time from analysis of the topics in conversation. The ratio of the reference of past, present and future of each speaker was enumerated. The individual differences of the tendencies were demonstrated as the ratios.

Unsupervised word embeddings provide rich linguistic and conceptual information about words. However, they may provide weak information about domain specific semantic relations for certain tasks such as semantic parsing of natural language queries, where such information about words can be valuable. To encode the prior knowledge about the semantic word relations, we present new method as follows: We extend the neural network based lexical word embedding objective function Mikolov, et.al. 2013 by incorporating the information about relationship between entities that we extract from knowledge bases. Our model can jointly learn lexical word representations from free text enriched by the relational word embeddings from relational data (e.g., Freebase) for each type of entity relations. We empirically show on the task of semantic tagging of natural language queries that our enriched embeddings can provide information about not only short-range syntactic dependencies but also long-range semantic dependencies between words. Using the enriched embeddings, we obtain an average of 2% improvement in F-score compared to the previous baselines.

Since the occurrence of the first `flash mob' organized by Bill Wasik (senior editor of the Harper's Magazine) in Manhattan in 2003, flash mob phenomenon has become widespread. Recent journalistic accounts have reported that this form of public engagement can pose significant threats to civil, political, social, and economic stability of a region. Gaps in the scientific understanding of such phenomenon and the imminent security risks posed by such acts call for a need to systematically study them. In this ongoing research, we shed light on the social dynamics of the flash mob phenomenon and build a conceptual model examining the necessary factors for the formation of flash mob and predicting its success or failure. Grounded in the sociological theories of collective action and collective identity formation, we evaluate the motivations of a flash mob practitioner and logically analyze the choices he/she would face with regards to acting or withdrawing from the flash mob. More broadly, this work is an attempt to bridge social and computational sciences that would help clarify and explain manifestations of emerging sociotechnical behaviors such as parkour, campaigns, and social movements that are widely observed.

This paper introduces a new qualitative spatial reasoning formalism, called   Interval Occlusion  Calculus (IOC), that takes into account multiple (distinct) viewpoints of a scene. This formalism extends  Allen's Interval Algebra by including an interval-based definition for spatial occlusion.