This paper presents a novel, data-driven language model that produces entire lyrics for a given input melody. Previously proposed models for lyrics generation suffer from the inability of capturing the relationship between lyrics and melody partly due to the unavailability of lyrics-melody aligned data. In this study, we first propose a new practical method for creating a large collection of lyrics-melody aligned data and then create a collection of 1,000 lyrics-melody pairs augmented with precise syllable-note alignments and word/sentence/paragraph boundaries. We then provide a quantitative analysis of the correlation between word/sentence/paragraph boundaries in lyrics and melodies. We then propose an RNN-based lyrics language model conditioned on a featurized melody. Experimental results show that the proposed model generates fluent lyrics while maintaining the compatibility between boundaries of lyrics and melody structures.

One critical issue of zero anaphora resolution (ZAR) is the scarcity of labeled data. This study explores how effectively this problem can be alleviated by data augmentation. We adopt a state-ofthe-art data augmentation method, called the contextual data augmentation (CDA), that generates labeled training instances using a pretrained language model. The CDA has been reported to work well for several other natural language processing tasks, including text classification and machine translation (Kobayashi, 2018;Gao et al., 2019). This study addresses two underexplored issues on CDA, that is, how to reduce the computational cost of data augmentation and how to ensure the quality of the generated data. We also propose two methods to adapt CDA to ZAR: [MASK]-based augmentation and linguistically-controlled masking. Consequently, the experimental results on Japanese ZAR show that our methods contribute to both the accuracy gain and the computation cost reduction. Our closer analysis reveals that the proposed method can improve the quality of the augmented training data when compared to the conventional CDA.

No abstract

Abstract-We found a way to use mathematical search to provide better navigation for reading papers on computers. Since the superficial information of mathematical expressions is ambiguous, considering not only mathematical expressions but also the texts around them is necessary. We present how to extract a natural language description, such as variable names or function definitions that refer to mathematical expressions with various experimental results. We first define an extraction task and constructed a reference dataset of 100 Japanese scientific papers by hand. We then propose the use of two methods, pattern matching and machine learning based ones for the extraction task. The effectiveness of the proposed methods is shown through experiments by using the reference set.

A number of studies have presented machine-learning approaches to semantic role labeling with availability of corpora such as FrameNet and PropBank. These corpora define the semantic roles of predicates for each frame independently. Thus, it is crucial for the machine-learning approach to generalize semantic roles across different frames, and to increase the size of training instances. This paper explores several criteria for generalizing semantic roles in FrameNet: role hierarchy, human-understandable descriptors of roles, semantic types of filler phrases, and mappings from FrameNet roles to thematic roles of VerbNet. We also propose feature functions that naturally combine and weight these criteria, based on the training data. The experimental result of the role classification shows 19.16% and 7.42% improvements in error reduction rate and macro-averaged F1 score, respectively. We also provide in-depth analyses of the proposed criteria.

Recently, several text mining programs have reached a near-practical level of performance. Some systems are already being used by biologists and database curators. However, it has also been recognized that current Natural Language Processing (NLP) and Text Mining (TM) technology is not easy to deploy, since research groups tend to develop systems that cater specifically to their own requirements. One of the major reasons for the difficulty of deployment of NLP/TM technology is that re-usability and interoperability of software tools are typically not considered during development. While some effort has been invested in making interoperable NLP/TM toolkits, the developers of end-to-end systems still often struggle to reuse NLP/TM tools, and often opt to develop similar programs from scratch instead. This is particularly the case in BioNLP, since the requirements of biologists are so diverse that NLP tools have to be adapted and reorganized in a much more extensive manner than was originally expected. Although generic frameworks like UIMA (Unstructured Information Management Architecture) provide promising ways to solve this problem, the solution that they provide is only partial. In order for truly interoperable toolkits to become a reality, we also need sharable type systems and a developer-friendly environment for software integration that includes functionality for systematic comparisons of available tools, a simple I/O interface, and visualization tools. In this paper, we describe such an environment that was developed based on UIMA, and we show its feasibility through our experience in developing a protein-protein interaction (PPI) extraction system.

Many recent Short Answer Scoring (SAS) systems have employed Quadratic Weighted Kappa (QWK) as the evaluation measure of their systems. However, we hypothesize that QWK is unsatisfactory for the evaluation of the SAS systems when we consider measuring their effectiveness in actual usage. We introduce a new task formulation of SAS that matches the actual usage. In our formulation, the SAS systems should extract as many scoring predictions that are not critical scoring errors (CSEs). We conduct the experiments in our new task formulation and demonstrate that a typical SAS system can predict scores with zero CSE for approximately 50% of test data at maximum by filtering out low-reliablility predictions on the basis of a certain confidence estimation. This result directly indicates the possibility of reducing half the scoring cost of human raters, which is more preferable for the evaluation of SAS systems.

The method for evaluating the colour rendering of light-emitting diode sources is controversial, especially for the appearance of human complexions. A psychophysical experiment was conducted using Chinese models to examine the effect of various illumination settings, characterized by two levels of preferred skin colour index and four levels of correlated colour temperature, on preference for the appearance of the Chinese complexion. Results showed that the preferred skin colour index was an effective indicator. Taking incomplete colour adaptation into account, the preferred skin colour index was recalculated and shown to be the best colour rendering index for the appearance of the Chinese complexion. This research examined the concept of preferred skin colour index for Chinese women. This can supplement other colour quality evaluation methods, and shows promise for commercial lighting applications.