National Institute for Japanese Language and Linguistics

The notion of the “perceptual center” or the “P-center” has been put forward to account for the repeated finding that acoustic and perceived syllable onsets do not necessarily coincide, at least in the perception of simple monosyllables or disyllables. The magnitude of the discrepancy between acoustics and perception—the location of the P-center in the speech signal— has proven difficult to estimate, though acoustic models of the effect do exist. The present study asks if the P-center effect can be documented in natural connected speech of English and Japanese and examines if an acoustic model that defines the P-center as the moment of the fastest energy change in a syllabic amplitude envelope adequately reflects the P-center in the two languages. A sensorimotor synchronization paradigm was deployed to address the research questions. The results provide evidence for the existence of the P-center effect in speech of both languages while the acoustic P-center model is found to be less applicable to Japanese. Sensorimotor synchronization patterns further suggest that the P-center may reflect perceptual anticipation of a vowel onset.

This study focuses on a method for differentiating between the stance of citizens and city councilors on political issues (i.e., in favor or against) and attempts to compare the arguments of both sides. We created a dataset by annotating citizen tweets and city council minutes with labels for four attributes: stance, usefulness, regional dependence, and relevance. We then fine-tuned pretrained large language model using this dataset to assign the attribute labels to a large quantity of unlabeled data automatically. We introduced multitask learning to train each attribute jointly with relevance to identify the clues by focusing on those sentences that were relevant to the political issues. Our prediction models are based on T5, a large language model suitable for multitask learning. We compared the results from our system with those that used BERT or RoBERTa. Our experimental results showed that the macro-F1-scores for stance were improved by 1.8% for citizen tweets and 1.7% for city council minutes with multitask learning. Using the fine-tuned model to analyze real opinion gaps, we found that although the vaccination regime was positively evaluated by city councilors in Fukuoka city, it was not rated very highly by citizens.

A method for superimposing the shape of the palate on three-dimensional (3D) electromagnetic articulography (EMA) data is proposed. A biteplate with a dental impression tray and EMA sensors is used to obtain the palatal shape and record the sensor positions. The biteplate is then 3D scanned, and the scanned palate is mapped to the EMA data by matching the sensor positions on the scanned image with those in the EMA readings. The average distance between the mapped palate and the EMA palate traces is roughly 1 mm for nine speakers and is comparable to the measurement error of the EMA.

Recent advances in non-invasive brain function measurement technologies, such as functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG), and the development of machine learning techniques, including deep learning, have led to increased research on the elucidation and quantitative understanding of information processing processes in the human brain. Since the emergence of word2vec, which represents the meaning of natural language words as vectors, features of language stimuli given to the human brain have been represented using large language models in natural language processing and used to estimate brain states. In this study, we used GPT-2, which is known to perform well as a feature for predicting brain states, and investigated the information processing processes in the human brain when reading Japanese short poems, i.e., tanka poetry. In particular, we investigated the hubness of the regions of interest in the brain by applying the PageRank algorithm. As a result, we have found that the cingulate cortex and the insula, which are said to be related to emotion, have hubness in brain regions, while occipital lobe, which are not said to be related to emotion, have also hubness.

Gathering citizen feedback, analyzing it, and comparing the results against other cities is essential for improving government policy and service quality. However, because different cities have different policies and services, the opinions of citizens in different cities also differ. This makes it difficult to analyze citizen opinions adapted for multiple cities using machine learning. In this study, we propose a method for extracting citizen opinions across cities. We evaluated our proposed method using a tweet dataset collected from citizens of Yokohama, Sapporo, and Sendai, confirming its effectiveness to fine-tune a model using the source city and re-fine-tune it with a few tweets from the target city. We clarified that training data in the target city can be effectively selected using the model trained with tweets from the source city, with high confidence in the prediction.

The task of detecting words with semantic differences across corpora is primarily addressed by word representations such as Word2Vec or BERT. However, there are no abundant computing resources available in the real world where linguists and sociologists apply these techniques. In this paper, we extend an existing CPU-trainable model which trains vectors of all time periods simultaneously. Experimental results demonstrate that the extended models achieved comparable or superior results to strong baselines in English corpora, SemEval-2020 Task 1, and Japanese. Furthermore, we compared the training time of each model and conducted a comprehensive analysis of Japanese corpora.

This paper examines curation methods utilizing Omeka S and IIIF in NINJAL Digital Archive “NINDA”, an online digital archive of language resources owned by the National Institute for Japanese Language and Linguistics（NINJAL）． Using specific examples, the paper will show that curation methods utilizing Omeka S and IIIF are effective in digitally archiving language resources. We also show that digital curation is effective in the digital archiving of language resources in terms of LOD and RDF. This paper examines curation methods utilizing Omeka S and IIIF in NINDA, an online digital archive of language resources owned by NINJAL.

In this paper, we focus on wokototen markings, which are a system of kunten annotations used to facilitate the reading of classical Chinese documents by Japanese readers. Using digitized data, we performed basic measurements of wokototen by using a chart that summarizes the wokototen markings of actual kunten materials described by Hiroshi Tsukishima, and we quantitatively clarified their characteristics. Kunten materials are classical Chinese books with annotations, called kunten , on the Chinese text. The wokototen is a type of kunten . In ancient East Asian countries, kunten systems were developed as a way of directly annotating Chinese documents so that they could be read and understood by non-native readers. For this reason, kunten materials and kunten are treated as historical sources for linguistic and historical research. The shape and position of a wokototen marking determines what kind of reading it indicates. The results of our basic survey quantitatively show that almost all the wokototen charts in actual kunten materials contain particles represented by “te”, “ni”, and “wo”, the most common shapes of wokototen are dots and shapes that can be written with a single stroke, such as ｜, ─, and ＼, and that the most common places to find these markings are to the right of characters in the horizontal direction and below characters in the vertical direction.

Universal dependencies (UD) are part of an international project that aims to construct cross-lingual dependency treebanks. The consistent annotation standards of grammar (parts of speech, morphological features, and syntactic dependencies) are defined across different languages and compiled as treebanks of more than 100 languages. The languages written without word delimitation must define the word units of their syntactic words on the UD guideline. The preceding UD Japanese resources are based on the short-unit words by NINJAL, which is defined by their lexicon-based morphology. This study introduces UD Japanese resources UD_Japanese_BCCWJ-GSDLUW, UD_Japanese_PUDLUW, and UD_Japanese_BCCWJLUW based on the long-unit words by NINJAL, which are more suitable as syntactic words than NINJAL’s short-unit words in Japanese.

The present work attempts to examine the relationship between grammar and discourse. (i) First, it compares Warrongo (an ergative language that has antipassives and an S/O pivot) and English (an accusative language that has passives and an S/A pivot). Despite these polar opposite morphosyntactic characteristics, Warrongo and English behave almost in the same way in discourse – in terms of new mentions, lexical mentions and topic continuity. There are, however, two differences in discourse. First, Warrongo antipassives and S/O pivot have much higher functional loads than English passives and S/A pivot. Second, Warrongo antipassives have a use that English passives do not have. (ii) Then, the present work shows that grammar and discourse are not independent of each other and that they share one principle. The hierarchy of “O > S > A” is attested in grammar and discourse crosslinguistically and irrespective of the morphosyntactic types of the languages concerned.

The goal of keyword extraction is to extract from a text, words, or phrases indicative of what it is talking about. In this work, we look at keyword extraction from a number of different perspectives: Statistics, Automatic Term Indexing, Information Retrieval (IR), Natural Language Processing (NLP), and the emerging Neural paradigm. The 1990s have seen some early attempts to tackle the issue primarily based on text statistics [13, 17]. Meanwhile, in IR, efforts were largely led by DARPA’s Topic Detection and Tracking (TDT) project [2]. In this contribution, we discuss how past innovations paved a way for more recent developments, such as LDA, PageRank, and Neural Networks. We walk through the history of keyword extraction over the last 50 years, noting differences and similarities among methods that emerged during the time. We conduct a large meta-analysis of the past literature using datasets from news media, science, and medicine to business and bureaucracy, to draw a general picture of what a successful approach would look like.

This study analyzes the little known Manchukuo Young Girl Envoys – Manchukuo's first government-appointed diplomats – and their official visit to Japan between June 22 and July 12, 1932. Existing studies on the Envoys tend to interpret them from an angle of contemporary Japanese people's national sentiment and deem their visit to Japan a show that the Japanese authorities in Manchukuo and Japan orchestrated together, to render Japan a strong world power. This study problematizes that view and considers the Envoys more of a product of intense power struggles inside Manchukuo's highest Japanese ruling strata, suggesting that Manchukuo's decision-making circle in 1932 was far from being a unified entity. Examining the Envoys' interactions in Japan and comparing relevant Japanese- and Chinese-language news coverage on them, this study argues for the possibility of tracing the intertwined national ideals of Manchukuo's Chinese and Japanese government leaders based on the inspiring example of the Envoys.

We investigated pupillary responses to the world's shortest fixed verses, Japanese haiku as aesthetic poetry (AP) and senryu as comic poetry (CP), in comparison with non-poetry control stimuli (NP) comprised of slogans that had the same rhythm patterns. Native Japanese speakers without literary training listened to these stimuli while we recorded their pupil diameters. We found that participants' pupils were significantly dilated for CP compared to NP in an early time window. While AP also evoked larger dilations than NP, the latency for AP-related pupil dilation was relatively long. Thus, lay people experience quick and intense arousal in response to funny and humorous words, while aesthetic properties of words may also elicit intense but slower changes in listeners' arousal levels, presumably because they evoke more implicit and subtle emotional effects. This study is the first to provide evidence that poetic language elicits human pupillary dilation. A better understanding of the cognitive and neural substrates for the sensitive awareness of pleasures expressed via poetic language will provide insights for improving mental and physical health. Hence, pupillometry can act as a useful convenient measurement to delineate the sympathetic activation of emotional contexts via language.