This paper traces the development of theories of lexical acquisition in the past three decades to discusses what needs to be explained in them. In the last two decades of the 20^th century, the so-called constraints-based nativist theories raised and flourished. While it explained certain phenomena, the scope of these theories was limited, and have not addressed the most critical questions for lexical development. In this paper, I argue that theories of lexical development must explain not only how infants begin the endeavor of lexical learning but also how they proceed from there to possess the adult lexicon, which is extremely vast, complex, and largely language-specific system of abstract symbols. For this purpose, following issues should be addressed in theories of lexical acquisition: (1) how symbols can be grounded to the body, as proposed by Steven Harnad (1990); (2) what cognitive abilities or dispositions enable Symbol Grounding, and whether (or in what degree) they are human-specific or evolutionary shared with our ancestors; (3) the mechanism with which human children bootstrap themselves from the initial, rudimentary symbols to adult lexical representation; (4) what cognitive abilities enable such bootstrapping processes.

In recent years, artificial intelligence based on deep neural networks (DNNs) has made remarkable progress. Particularly in natural language processing (NLP), various DNN-based language models have emerged, using Transformer architectures that are pre-trained on large-scale text data. These pre-trained large language models (LLMs) have demonstrated high accuracy across a range of NLP tasks, leading to claims that they surpass human capabilities in language understanding. However, due to the black-box nature of LLMs, it is not clear whether LLMs realize human-like language understanding. The question of whether neural networks can acquire compositionality has been debated between two modeling of human cognition: connectionism and classical computationalism. In 1988, Fodor and Pylyshin proposed two types of systematicity regarding compositionality: systematicity of thought and systematicity of inference. This study revisits Fodor and Pylyshin’s discussion on systematicity and presents a method for evaluating whether LLMs demonstrate compositional language understanding, specifically focusing on inferential systematicity through natural language inference tasks. The findings of this paper indicate that there is still room for debate regarding whether current LLMs achieve the systematicity that underlies human language understanding and reasoning. Consequently, it emphasizes the need for further complementary research bridging the fields of NLP and cognitive science to delve deeper into this topic.

The goal of the present study is to approach the human language basis from the perspective of grammatical errors that humans make in their process of language learning. We adopt the generative biolinguistic perspective and analyze grammatical errors of Japanese learners of English as a foreign language (EFL) and learners of Japanese within the framework of the so-called Principles and Parameters Approach to Universal Grammar (Chomsky, 1965, 1981, 1995). It was found that grammatical errors of Japanese EFL learners are analyzable in terms of grammatical characteristics of other languages than their native language, Japanese, and target language, English and that those of learners of Japanese are also interpretable as grammatically acceptable with respect to grammatical characteristics of other languages than their native or target language. This finding supports generative error analysis (i.e., error analysis from the generative biolinguistic perspective) and deepens our further understanding of the notion of parameters that capture to what extent cross-linguistic variation is innately possible as the human language basis.

The difference between metaphors and similes tends to be regarded as the absence or presence of metaphorical indicators. However, we cannot explore all possible metaphorical indicators. Furthermore, we are able to define neither the types nor ranges of metaphorical indicators. There are possibilities for finding new metaphorical indicators and categorising their functions. Therefore, we explored clue phrases for the understanding of figurativity from large-scale simile examples. As a result, we found many clue phrases which have not been reported as metaphorical indicators in preceding studies. Most of the newly found clue phrases can be categorised into the expressions to differentiate between the topic and vehicle. The clue phrases are included with many simile examples. When the representative metaphorical indicators occurred in a simile expression, the clue phrases also co-occurred. The metaphorical indicators can be regarded as the degree to which the linguistically obtainable group of clue phrases in metaphorical expressions is grasped, while metaphors and similes should be a continuity of metaphorical expression with their gradations rather than two distinct sets.

Ostensive inferential communication is a form of interaction wherein a speaker’s communicative intention is effectively conveyed through various cues directed towards the listener, such as eye gaze, pointing gestures, and utterances. This review examines studies exploring the effects and roles of ostensive inferential communication in children’s word learning. The argument is made that this mode of communication serves as a foundational element of language development. Infants exhibit sensitivity to adults’ ostensive cues from an early age, utilizing pointing gestures for communicative purposes from infancy. Their acquisition of word meanings is rooted in inferencing the speaker’s referential intention across different situations, even when words and objects lack direct associations. Furthermore, infants acquire both object labels and part labels, with a particular emphasis on the importance of subtle ostensive pointing. The article concludes by addressing unresolved issues and suggesting future research directions.

In the process of compound truncation for loanwords in Japanese, the most common form is to take the first two moras of each component word. However, when the second mora of the first component contains a long sound, it may be shortened. It has been argued that two principles operate in the formation of shortened words: energy conservation and recoverability. Word truncation is the process of attempting to shorten a word, but if the word is shortened too much, the original word cannot be recovered. This idea leads to the hypothesis that when recoverability is highly demanded, people would attempt to shorten a long sound while preserving the information in the next mora. We conducted an experiment using nonce words and analyzed a database of real words, which revealed that shortening is rather reduced under conditions where recoverability is highly demanded. Shortening tended to be more frequent for meaningful words that should be recoverable compared to nonwords. Additionally, shortening was observed in cases where shortened forms resulted in more phonologically competing items. We reanalyzed the results from the perspective that phonologically similar words are more accessible as they gain more activation in the mental lexicon. We found that words with more phonological neighbors were more likely to be shortened when they were longer than average. This pattern was also supported by an experiment with nonce words. The results were interpreted in the view of preference for accessibility over recoverability in the process of compound word truncation.

Storytelling is a culturally universal phenomenon deeply intertwined with human language communication and social cognition. This paper explores the cultural evolution of stories from two perspectives: (1) their adaptive function for humans and (2) the cognitive and environmental constraints for humans transmitting and consuming stories. Drawing on empirical studies, the paper discusses how stories encapsulate valuable knowledge that aids adaptation to social and natural environments. Then, the interplay between the cognitive basis and the adaptive function of stories is discussed, especially focusing on their role in facilitating language communication. Finally, we present a hypothesis that stories have made it possible to transmit information more efficiently, and that the existence of these stories may have influenced the way humans are. We emphasize the importance of interdisciplinary research to test this hypothesis.

Relative embodiment (RE) measures the degree to which a given verb’s meaning is related to a human’s body or action. Verbs with a high RE are recognized faster than verbs with a low RE. However, the actual robustness of this phenomenon remains unclear because word combinations can change the meaning of the verbs. Accordingly, we conducted a two-factor within-participant experiment (word combination × stimulus onset asynchrony [SOA]) involving 42 native Japanese speakers. Regarding the word combination factor, we compared expressions denoting literal actions (literal condition) and metaphorical actions (metaphorical condition) that involved high RE verbs. The experiment also included expressions that used low RE verbs to denote abstract actions (abstract condition). SOA factors were compared between two conditions where the words were processed consecutively (SOA 250 ms condition) and separately (SOA 800 ms condition). Participants had to determine whether the target word was a verb or a noun. The results revealed significantly shorter reaction times in the literal condition compared to the metaphorical condition. Moreover, reaction times were significantly shorter in the literal condition compared to the abstract condition in the SOA 800 ms condition. Metaphorical expressions received fewer benefits from RE when low-imageability nouns were combined with high RE verbs. Unlike previous studies, this study did not observe any embodiment effects when the words were read consecutively (e.g., in normal reading). Thus, even if RE affects verbs in single-word processing, its effect may be small or negligible in the case of everyday sentences and phrases.

In this paper, we examine the effectiveness of intonational and gestural cues in understanding syntactically ambiguous Japanese sentences. In Tokyo Japanese, it is well known that syntactic ambiguous sentences are distinguished by the intonation. Moreover, in our previous study, it is reported that the intonational distinction decreases when gestures are accompanied by the utterance but the speakers are not aware of the fact. In this study, we examine whether gestures are effective cues in compensating for the loss of intonation for the hearers’ understanding. In the experiment, we created three types of stimuli: (i) audio of utterances whose speakers concentrated only on reading aloud ambiguous sentences, (ii) audio + video of utterances whose speakers gave gestures accompanied with ambiguous utterances, and (iii) audio extracted from (ii). The results show that the hearers were able to understand the meaning of ambiguous sentences when they listened to the audio of utterances without gesture more correctly than when they watched/listened to other stimuli. In addition, the hearers’ confidence in understanding the intended meaning was influenced by the type of stimuli, while the speakers’ confidence in their communication was not. The results indicate that the hearers understood the intended meaning more accurately when they listened to the utterances whose intonation patterns correspond to the syntactic structure than when they watched the videos with gestures. Moreover, the hearers were aware of this fact. We conclude that different means can be used effectively for speakers and hearers to communicate and understand intentions, respectively.

The objective of the study was to ascertain whether the use of symbols by young children and adults changes under different conditions in which the ease of establishing common ground was contrasted. In the experiment, participants were randomly assigned to either a condition where they had to communicate about objects that were hidden from the addressee or a condition where the objects were visible to the addressee. The stimuli consisted of containers, half of which had typical shapes that were linguistically easy to describe, and the other half had atypical shapes that were linguistically harder to describe. The analyses focused on the production of adjectival expressions, ideophones, iconic gestures, and metaphorical expressions under each respective condition by participants. The findings demonstrated that both adults and young children utilized adjectival expressions when referencing stimuli of a typical shape and metaphorical expressions for atypically shaped items. Conversely, young children employed iconic gestures while referencing stimuli obscured from their partner’s sight. Moreover, five-year-olds used a greater number of ideophones when referring to stimuli invisible to the addressee. The findings underscore that metaphorical expressions serve a depictive function in communication and that the usage of iconic symbols by young children is influenced by the presence of a shared visual common ground with the addressee.

The foundation of language underlies the process in which language emerges through interaction and dynamics. Based on this idea, this paper focuses on constructive approaches to the emergence of language and communication, including multi-agent reinforcement learning (MARL), emergent communication (EC), symbol emergence in robotics, experimental semiotics, iterated learning model (ILM), and co-evolution. Although they have similar motivations, their efforts have been fragmented from community to community, and their philosophies and methodologies have not been shared well with each other. We introduce them from a birds-eye viewpoint and aim to suggest the direction that they should take in the future.

Language, as a representative symbol system, undergoes developmental learning in everyday life and cultural evolution throughout history. This paper explores the cognitive and social foundations that contribute to the emergence of symbol systems, including language. It presents a novel hypothesis, the Collective Predictive Coding (CPC) hypothesis, which emphasizes the interdependence between the formation of internal representations through physical interaction with the environment and the sharing and use of meaning through social semiotic interaction with others within a symbol emergence system. The paper reviews previous approaches to symbol emergence systems, such as emergent communication, multi-agent reinforcement learning, iterated learning models, and symbol emergence in robotics. After defining CPC and the CPC hypothesis, the paper explores research opportunities and challenges in the next generation of symbol emergence in robotics. It also explores potential cross-disciplinary research directions for the future, including the connection with computational linguistics.

Japanese sentence-final particles such as “yo” and “ne” reflect our cognitive mechanism of manipulating and integrating linguistic and pragmatic/sociolinguistic information. However, despite the importance of this mechanism in cognitive science, few constructive approaches have been presented so far. Therefore, we developed the Subjective-BERT model for acquiring the meanings of the two sentence-final particles, “yo” and “ne” and those of content words. The proposed model was pre-trained on the dataset with not only linguistic information but also sensory and mental information. In a previous study (Mandokoro et al., 2022a), we demonstrated that the model acquires the meanings of the particles by learning the relations between not only words but also words and sensory/mental information. This paper further explores how the meaning of each word is represented and processed in the model by analyzing the internal representations. Among the main results: (i) the analysis of the attention chain between three tokens (a particle, a content word, and its referent), revealed that the information of the particle is first conveyed to the content word, which in turn controls the subsequent information flow from the content word to its referent; and (ii) the contrastive analysis of the transformation processes of embeddings when each sentence-final particle is used, enables the identification of the mechanism used to generate the attention chains, suggesting that the residual connections and feedforward neural network layers, as well as self-attention, contribute to the understanding of utterance meaning.