Instruction prompt tuning (IPT) uses crafted prompts and in-context demonstrations (ICD) to guide large language models (LLMs) in performing previously unseen tasks, transferring knowledge while keeping the LLM mostly frozen. Social media multi-party conversation (MPC) analysis, on the other hand, has achieved remarkable performance on in-domain tasks including exact speaker identification. However, due to several challenges such as the lack of contextual generalization, absence of adequate ICD, and biases in LLMs, significant work remains to be done for quantifying mental health on social media MPC data using out-of-domain (OOD) knowledge transfer. In this article we propose DoRA , a novel dual-encoder demonstration retriever architecture designed to transfer an LLM’s knowledge about MPC modeling for previously unseen depression screening tasks. Our method pairs soft embeddings of MPC prompts with top-ranked ICD claims for depression screening, leveraging IPT-based OOD cross-task transfer using DoRA for the first time. Experiments conducted in zero-shot and few-shot settings across benchmark datasets using multiple LLMs demonstrate significant downstream performance with DoRA . Specifically, there is a 21.54% increase in recall for depressed utterance classification and a 21.11% increase in F 1 score for depressed speaker identification. These results highlight DoRA ’s potential as a screening tool for digital mental health applications.

Due to the complexity of traditional fine-tuning methods, prompt-based engineering methods, such as prompt tuning (PT), have recently received increasing attention for casting various downstream tasks into a large language model (LLM) format by prepending soft tunable embeddings to input sequences while keeping the majority of the LLM frozen [1]. On the other hand, the discourse analysis of text-based multi-party conversations (MPCs) has been used to obtain valuable insights like speaker emotion recognition [2]. Although most prior work considers indomain knowledge transfer using PT, much work remains to be done for PT-based out-of-domain (OOD) knowledge transfer such as adapting a LLM's knowledge about MPC modelling to a semantically dissimilar task such as depression detection. This study aims to detect human depression in MPCs by leveraging an LLM's knowledge about MPC modelling.

Parameter-efficient, soft, and prompt-based tuning methods have received increasing attention in various downstream tasks due to the high cost of traditional fine-tuning methods in pre-trained language models (PLM). Prompt tuning (PT) is one such effective mechanism which has achieved remarkable performance in transferring the acquired knowledge of a PLM to perform an unseen task within the same domain using task-specific prompts and informative instructions. Even though most prior work considers in-domain knowledge transfer using PT, much work remains to be done for PT-based out-of-domain knowledge transfer. In this study we propose ProDepDet, a novel framework specifically designed to use a PLM's knowledge about structure and semantic modelling in multi-party conversations to perform the unseen, out-of-domain task of depression detection. To our knowledge, this study is the first attempt to adapt the acquired knowledge of a PLM for out-of-domain task modelling using PT-based cross-task transferability. Experiments on few-shot and full data settings across multiple benchmark datasets demonstrate the superiority of our PT framework in two downstream tasks including depressed utterance classification and depressed speaker identification.

In response to the globalization of education, an increasing number of Japanese universities are adopting English-medium instruction in computing classes. However, instructors often encounter significant barriers, ranging from the students’ insufficient English language competency to the passive learner mentality, and research on EMI-based computing education remains limited. In this short paper, we share the strategies that we implemented both at the course level and at the department level to support students’ learning in EMI-based computing courses. We map our strategies to the three dimensions of transitions posed in the multiple and multidimensional transitions (MMT) theory. While previous studies and common EMI practices have focused primarily on supporting students’ academic and social-cultural transitions, some of our strategies also serve to support students’ psychological transition from high school to university in an EMI context. Our experience has relevance for other educational institutions, particularly those in Japan, where English-taught computing courses are being implemented or expanded.

Recently, real-time affect-awareness has been applied in several commercial systems, such as dialogue systems and computer games. Real-time recognition of affective states, however, requires the application of costly feature extraction methods and/or labor-intensive annotation of large datasets, especially in the case of Asian languages where large annotated datasets are seldom available. To improve recognition accuracy, we propose the use of cognitive context in the form of “emotion-sensitive” intentions. Intentions are often represented through dialogue acts and, as an emotion-sensitive model of dialogue acts, a tagset of interpersonal-relations-directing interpersonal acts (the IA model) is proposed. The model's adequacy is assessed using a sentiment classification task in comparison with two well-known dialogue act models, the SWBD-DAMSL and the DIT++. For the assessment, five Japanese in-game dialogues were annotated with labels of sentiments and the tags of all three dialogue act models which were used to enhance a baseline sentiment classifier system. The adequacy of the IA tagset is demonstrated by a 9% improvement to the baseline sentiment classifier's recognition accuracy, outperforming the other two models by more than 5%.

Maximizing the quality of service experienced by passengers of public transportation systems is of great importance to service operators. Accurate and timely information about passenger load, both at stops and on board vehicles, can help in many ways to improve the quality of service. This information is currently not readily available using data collection techniques that are transparent to passengers. We describe a method and feasibility study for estimating transit passenger population at two bus stops by passively monitoring Wi-Fi network activity of mobile devices. The collected data is filtered and analyzed to distinguish passengers' mobile devices from non-passenger devices. Our experimental results show that filtered Wi-Fi activity correlates well with observed passenger population at bus stops and has the potential to provide near real-time information.

This paper introduces OMeta, a new object-oriented language for pattern matching. OMeta is based on a variant of Parsing Expression Grammars (PEGs) [5]---a recognition-based foundation for describing syntax---which we have extended to handle arbitrary kinds of data. We show that OMeta's general-purpose pattern matching provides a natural and convenient way for programmers to implement tokenizers, parsers, visitors, and tree transformers, all of which can be extended in interesting ways using familiar object-oriented mechanisms. This makes OMeta particularly well-suited as a medium for experimenting with new designs for programming languages and extensions to existing languages.

The VVM (virtual virtual machine) is a systematic approach to adaptability and reconfigurability for portable, object-oriented applications based on byte-coded languages such as Java and Smalltalk. The main objectives of the VVM are (i) to allow adaptation of language and system according to a particular application domain; (ii) to provide extensibility by allowing a live execution environment to evolve according to new protocols or language standards; and (iii) to provide a common substrate on which to achieve true interoperability between different languages. On the way to implement a VVM we have already implemented VVM1 (and its application to active networks) and VVM2 (and its application to flexible Web cache and distributed observation). The VVM2 is a highly-flexible language kernel which consists of a minimal, complete programming language in which the most important goal is to maximise the amount of reflective access and intercession that are possible, at the lowest possible software level.

Most existing reference-based distributed object systems include some kind of acyclic garbage collection, but fail to provide acceptable collection of cyclic garbage. Those that do provide such GC currently suffer from one or more problems: synchronous operation, the need for expensive global consensus or termination algorithms, susceptibility to communication problems, or an algorithm that does not scale. We present a simple, complete, fault-tolerant, asynchronous extension to the (acyclic) cleanup protocol of the SSP Chains system. This extension is scalable, consumes few resources, and could easily be adapted to work in other reference-based distributed object systems---rendering them usable for very large-scale applications.

Achieving good performance in bytecoded language interpreters is difficult without sacrificing both simplicity and portability. This is due to the complexity of dynamic translation ("just-in-time compilation") of bytecodes into native code, which is the mechanism employed universally by high-performance interpreters.We demonstrate that a few simple techniques make it possible to create highly-portable dynamic translators that can attain as much as 70% the performance of optimized C for certain numerical computations. Translators based on such techniques can offer respectable performance without sacrificing either the simplicity or portability of much slower "pure" bytecode interpreters.