Powerful Language Research Articles

PILL: Plug into LLM with adapter expert and attention gate

Due to the remarkable capabilities of powerful Large Language Models (LLMs) in effectively following instructions, significant progress has recently been made in the development of Vision Language Models (VLMs), expanding the capabilities of LLMs in multi-modal learning and enabling them to be fine-tuned in a parameter-efficient manner. Promising as the existing pre-trained vision-language models might be, there are two intractable issues including modal entanglement and insufficient capability of LLMs in handling visual information. To tackle these two issues, we introduce an architecture called PILL which Plugs Into LLM with adapter expert and attention gate. To maintain the capability of the original LLMs and the more progressive injection of visual modality, one Modality-Attention-Gating (MAG) module is introduced, enabling adaptive control of the contribution of modality tokens to the overall representation. Specifically, one Mixture-of-Modality-Adapter-Expert (MoMAE) module is proposed to handle different modalities with the dedicated adapters. In addition, further improvement is made to the adapter to enhance its learning and expressive capabilities. We adopt a two-stage training paradigm to optimize different modules of our model. Experimental results demonstrate that our approach exhibits competitive performance compared to other mainstream methods for modality fusion with much lower resources. We provide free access to the code and models.22https://github.com/DsaltYfish/PILL.

Optimizing GPT Technology in Educational Administration: Strategies, Practices, and Action Research Insights

With the rapid advancement of information technology, artificial intelligence (AI) has shown broad application prospects across various fields, particularly in educational administration. GPT technology, with its powerful natural language processing and understanding capabilities, can enhance the efficiency of educational administration, optimize resource allocation, and promote managerial innovation. This study aims to explore the practical applications of GPT technology in educational administration through action research, analyzing its potential value and challenges, and proposing targeted solutions. The findings reveal significant benefits of GPT in course content creation, contributing to improved efficiency in educational administration. However, the study also identifies challenges such as the complexity of technical integration, data security and privacy issues, dependency and misuse risks, and technology acceptance. To address these challenges, this paper offers a series of detailed strategies and operational recommendations, providing evidence-based support and practical reference for educational administrators to promote intelligent and efficient development in educational management.

Cascaded cross-modal transformer for audio–textual classification

Speech classification tasks often require powerful language understanding models to grasp useful features, which becomes problematic when limited training data is available. To attain superior classification performance, we propose to harness the inherent value of multimodal representations by transcribing speech using automatic speech recognition models and translating the transcripts into different languages via pretrained translation models. We thus obtain an audio–textual (multimodal) representation for each data sample. Subsequently, we combine language-specific Bidirectional Encoder Representations from Transformers with Wav2Vec2.0 audio features via a novel cascaded cross-modal transformer (CCMT). Our model is based on two cascaded transformer blocks. The first one combines text-specific features from distinct languages, while the second one combines acoustic features with multilingual features previously learned by the first transformer block. We employed our system in the Requests Sub-Challenge of the ACM Multimedia 2023 Computational Paralinguistics Challenge. CCMT was declared the winning solution, obtaining an unweighted average recall of 65.41% and 85.87% for complaint and request detection, respectively. Moreover, we applied our framework on the Speech Commands v2 and HVB dialog data sets, surpassing previous studies reporting results on these benchmarks. Our code is freely available for download at: https://github.com/ristea/ccmt.

Introduction: Languages of Power and Elite Legitimisation on the Periphery, Poland and Norway, 1000–1300

Introduction: Languages of Power and Elite Legitimisation on the Periphery, Poland and Norway, 1000–1300

Concluding Remarks: Languages of Power and Elite Legitimisation in Poland and Norway, 1000–1300

Concluding Remarks: Languages of Power and Elite Legitimisation in Poland and Norway, 1000–1300

DRM LISTS CREATED WITH CHATGPT: ANALYSING RECOGNITION MEMORY WITH CLIMATE-CHANGE THEMED LISTS

Machine learning and natural language processing have led to the development of powerful language models such as ChatGPT, which can generate consistent and human-like responses to a wide range of queries. In many domains, ChatGPT provides appropriate responses to given commands. One of the aims of this study is to investigate the use of these association lists, such as the Deese-Roediger-McDermott (DRM) lists popular in cognitive psychology studies, by ChatGPT by giving the necessary instructions. The same method was then used to create association lists around a specific topic (climate change). The results of the first study showed that participants gave more false answers when discriminating whether critical words were presented during the test phase than when related and unrelated words were presented. This finding shows that DRM lists generated by ChatGPT can be used to search for memory errors. In line with the literature, false answers for critical words were predominantly rated as ‘remember’. The results of the second study, which was applied to the lists created on the topic of climate change and compared the responses of the groups with the climate denial scores, show that there is no significant difference in the emergence of false memories between the two groups. The level of climate change denial did not significantly affect the participants’ responses to the critical words in the climate- related lists. The low level of climate denial in the sample is a limitation of this study. It is recommended that future studies compare memory performance across an appropriate sample.

Descriptional Complexity of Finite Automata – Selected Highlights

The state complexity, respectively, nondeterministic state complexity of a regular language L is the number of states of the minimal deterministic, respectively, of a minimal nondeterministic finite automaton for L. Some of the most studied state complexity questions deal with size comparisons of nondeterministic finite automata of differing degree of ambiguity. More generally, if for a regular language we compare the size of description by a finite automaton and by a more powerful language definition mechanism, such as a context-free grammar, we encounter non-recursive trade-offs. Operational state complexity studies the state complexity of the language resulting from a regularity preserving operation as a function of the complexity of the argument languages. Determining the state complexity of combined operations is generally challenging and for general combinations of operations that include intersection and marked concatenation it is uncomputable.

Generative AI enhances individual creativity but reduces the collective diversity of novel content.

Creativity is core to being human. Generative artificial intelligence (AI)-including powerful large language models (LLMs)-holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on generative AI ideas. We study the causal impact of generative AI ideas on the production of short stories in an online experiment where some writers obtained story ideas from an LLM. We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers. However, generative AI-enabled stories are more similar to each other than stories by humans alone. These results point to an increase in individual creativity at the risk of losing collective novelty. This dynamic resembles a social dilemma: With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced. Our results have implications for researchers, policy-makers, and practitioners interested in bolstering creativity.

Unveiling ‘HEREDIAQUA’: An Exploration of Association Picture Storytelling in Documentary Filmmaking

This study explores the application of the “association picture story” genre in documentary filmmaking, particularly focusing on the film “Herediaqua,” to effectively communicate water sustainability issues. The research employs a three-stage methodology: exploration (field research, observation, interviews, and collection of visual materials), experimentation (applying association picture story techniques in short film formats and testing with audiences), and creation (synthesizing findings to produce a cohesive and impactful film).Results indicate that “Herediaqua” successfully uses strong visual symbolism and narrative to convey messages about the interconnectedness between humans and nature, effectively enhancing audience awareness and understanding of water sustainability issues. The film employs powerful visual language to communicate its environmental message, evoking emotional responses from viewers. Positive audience feedback confirms the film’s success in communicating its message, demonstrating the effectiveness of the association picture story technique in visual narration. The conclusion asserts that the “association picture story” technique is a powerful tool for conveying environmental messages through visual storytelling. “Herediaqua” serves not only as a viewing experience but also as an educational medium and a catalyst for inspiring environmental conservation efforts. By leveraging compelling visual narration and engaging storytelling, the film promotes awareness and encourages concrete actions for sustaining the planet’s life. This research contributes to advancing documentary filmmaking by highlighting the effectiveness of combining visual symbolism with strong narrative techniques.

Tensions of perceiving and practice: an intersectional raciolinguistic DisCrit inquiry

ABSTRACT This longitudinal qualitative case study contextualizes the dynamic interplay of race, language, and disability through a conceptual framework grounded in the principles of Raciolinguistics, DisCrit theory, and Intersectionality to surface the tensions between the way a language-racialized student labeled as a ‘long-term English learner’ and disabled is institutionally perceived by mechanisms of education policy and practice and the ways the student perceives herself. Findings shed light on how racism, linguicism, and ableism intersect and overlap to disenfranchise one student through dynamic encounters with institutional mechanisms of education policy and practice. In centering student voice, we bear witness to a narrative that speaks against the institutional record that depicts language use and disability as deficient. Findings illustrate two critical tensions: 1) the student perceives herself as a powerful language user who, unlike the process and practices of the school, see her as not fully capable of engaging with grade-level content, and 2) the student’s classification as having a disability contrasts with her belief that she is just like other students, highlighting the tension between her self-identification and the institutional classification ascribed to her.

Revealing the Unseen: AI Chain on LLMs for Predicting Implicit Data Flows to Generate Data Flow Graphs in Dynamically-Typed Code

Data flow graphs (DFGs) capture definitions (defs) and uses across program blocks, which is a fundamental program representation for program analysis, testing and maintenance. However, dynamically-typed programming languages like Python present implicit data flow issues that make it challenging to determine def-use flow information at compile time. Static analysis methods like Soot and WALA are inadequate for handling these issues, and manually enumerating comprehensive heuristic rules is impractical. Large pre-trained language models (LLMs) offer a potential solution, as they have powerful language understanding and pattern matching abilities, allowing them to predict implicit data flow by analyzing code context and relationships between variables, functions, and statements in code. We propose leveraging LLMs’ in-context learning ability to learn implicit rules and patterns from code representation and contextual information to solve implicit data flow problems. To further enhance the accuracy of LLMs, we design a five-step Chain of Thought (CoT) and break it down into an AI chain, with each step corresponding to a separate AI unit to generate accurate DFGs for Python code. Our approach’s performance is thoroughly assessed, demonstrating the effectiveness of each AI unit in the AI Chain. Compared to static analysis, our method achieves 82% higher def coverage and 58% higher use coverage in DFG generation on implicit data flow. We also prove the indispensability of each unit in the AI Chain. Overall, our approach offers a promising direction for building software engineering tools by utilizing foundation models, eliminating significant engineering and maintenance effort, but focusing on identifying problems for AI to solve.

Hubris occurrence, management and accountability in higher education institutions in Australia

Hubris refers to an individual's display of self-assurance, pride or excessive confidence and has been presented in the extant literature to describe the language used by hubristic leaders. Therefore, this research explores the characteristic features of hubristic discourse in higher education and its potential link to the organisation's financial outcomes. By using computational linguistics to analyse written discourse samples of 20 Australian Vice-Chancellor's Addresses in the university's annual reports for the 10-year period, this research identifies linguistic markers of Vice-Chancellor hubris (increased use of negations, discrepancy words, power and achievement-oriented language) and their link to financial outcomes. As the role of the Vice-Chancellor has been compared in literature to that of the CEO, exploring if and how hubris’ symptoms manifest in language use can contribute towards explaining agency problems, accountability issues and the ongoing ethical debate over pay for performance in higher education. This research has implications for university key stakeholders such as government bodies, for standard setters who oversee the preparation of university financial reports and those who assume leadership roles within tertiary institutions.

Translating Soviet Moldova’s Language Politics: Observations on Tatiana Ţîbuleac’s The Glass Garden in Spanish Translation

Grădina de sticlă (The Glass Garden) is the second novel by the prize-winning Moldovan and Romanian author Tatiana Ţîbuleac. Set in Soviet Chişinău towards the end of the communist era, the book is narrated by a Moldovan-speaking young girl who has been adopted by an elderly Russian-speaking woman. Language and identity play key roles in the book, as the child is encouraged to forgo the Moldovan language and to learn how to speak Russian, the language of Soviet prestige and power. This personal struggle occurs against the background of the major sociopolitical and linguistic changes of perestroika and beyond, including the independence of Moldova, and is reflected by the fact that the text of the novel itself (though originally written in Ro-manian) is sprinkled with words and phrases in untransliterated Russian. Accordingly, noting that this complex linguistic situation could conceivably pose a challenge for translators of the novel into other languages, this study contrasts the original with its Spanish version, El jardín de vidrio. The aim is to show how, as part of the legacy of Soviet rule, the enduring complexities of Moldova’s linguistic and socio-political situation can be rendered through translation. Focussing initially on how the Spanish translation deals with the general features of Moldova’s language situation, it then moves to detailing how the novel’s frequent usage of Russian and the Cyrillic script is accommodated in the translated version. Lastly, selected aspects centring on the interplay between languages (e.g. differences in script and lexis, wordplay, misspellings) are scrutinised.

Let the algorithm speak: How to use neural networks for automatic item generation in psychological scale development.

Measurement is at the heart of scientific research. As many-perhaps most-psychological constructs cannot be directly observed, there is a steady demand for reliable self-report scales to assess latent constructs. However, scale development is a tedious process that requires researchers to produce good items in large quantities. In this tutorial, we introduce, explain, and apply the Psychometric Item Generator (PIG), an open-source, free-to-use, self-sufficient natural language processing algorithm that produces large-scale, human-like, customized text output within a few mouse clicks. The PIG is based on the GPT-2, a powerful generative language model, and runs on Google Colaboratory-an interactive virtual notebook environment that executes code on state-of-the-art virtual machines at no cost. Across two demonstrations and a preregistered five-pronged empirical validation with two Canadian samples (NSample 1 = 501, NSample 2 = 773), we show that the PIG is equally well-suited to generate large pools of face-valid items for novel constructs (i.e., wanderlust) and create parsimonious short scales of existing constructs (i.e., Big Five personality traits) that yield strong performances when tested in the wild and benchmarked against current gold standards for assessment. The PIG does not require any prior coding skills or access to computational resources and can easily be tailored to any desired context by simply switching out short linguistic prompts in a single line of code. In short, we present an effective, novel machine learning solution to an old psychological challenge. As such, the PIG will not require you to learn a new language-but instead, speak yours. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Gemini MultiPDF Chatbot: Multiple Document RAG Chatbot using Gemini Large Language Model

Abstract: The Gemini MultiPDF Chatbot represents a groundbreaking advancement in natural language processing (NLP) by integrating Retrieval-Augmented Generation (RAG) techniques with the Gemini Large Language Model. This innovative chatbot is designed to handle multiple document retrieval and generation tasks, leveraging the extensive knowledge base of the Gemini model. By harnessing RAG methods, the chatbot enhances its ability to acquire, comprehend, and generate responses across diverse knowledge sources contained within multiple PDF documents. The integration of Gemini's powerful language understanding capabilities with RAG facilitates seamless interaction with users, offering comprehensive and contextually relevant responses. This paper presents the design, implementation, and evaluation of the Gemini MultiPDF Chatbot, demonstrating its effectiveness in navigating complex information landscapes and delivering high-quality conversational experiences.

Unstructured Data Fusion for Schema and Data Extraction

Recently, there has been significant interest in extracting actionable insights from the abundance of unstructured textual data. In this paper, we introduce a novel problem, which we term Semistructured Schema and Data Extraction (SDE). This task aims to enhance and complete tables using information discovered from textual repositories, given partial table specifications in the form of queries. To effectively solve SDE, several challenges must be overcome, which involve transforming the partial table specifications into effective queries, retrieving relevant documents, discerning values for partially specified attributes, inferring additional attributes, and constructing an enriched output table while mitigating the influence of false positives from the retrieval. We propose an end-to-end pipeline for SDE, which consists of a retrieval component and an augmentation component, to address each of the challenges. In the retrieval component, we serialize the partial table specifications into a query and employ a dense passage retrieval algorithm to extract the top-k relevant results from the text repository. Subsequently, the augmentation component ingests the output documents from the retrieval phase and generates an enriched table. We formulate this table enrichment task as a unique sequence-to-sequence task, distinct from traditional approaches, as it operates on multiple documents during generation. Utilizing an interpolation mechanism on the encoder output, our model maintains a nearly constant context length while automatically prioritizing the importance of documents during the generation. Due to the novelty of SDE, we establish a validation methodology, adapting and expanding existing benchmarks with the use of powerful large language models. Our extensive experiments show that our method achieves high accuracy in enriching query tables through multi-document fusion, while also surpassing baseline methods in both accuracy and computational efficiency.

CodeS: Towards Building Open-source Language Models for Text-to-SQL

Language models have shown promising performance on the task of translating natural language questions into SQL queries (Text-to-SQL). However, most of the state-of-the-art (SOTA) approaches rely on powerful yet closed-source large language models (LLMs), such as ChatGPT and GPT-4, which may have the limitations of unclear model architectures, data privacy risks, and expensive inference overheads. To address the limitations, we introduce CodeS, a series of pre-trained language models with parameters ranging from 1B to 15B, specifically designed for the text-to-SQL task. CodeS is a fully open-source language model, which achieves superior accuracy with much smaller parameter sizes. This paper studies the research challenges in building CodeS. To enhance the SQL generation abilities of CodeS, we adopt an incremental pre-training approach using a specifically curated SQL-centric corpus. Based on this, we address the challenges of schema linking and rapid domain adaptation through strategic prompt construction and a bi-directional data augmentation technique. We conduct comprehensive evaluations on multiple datasets, including the widely used Spider benchmark, the newly released BIRD benchmark, robustness-diagnostic benchmarks such as Spider-DK, Spider-Syn, Spider-Realistic, and Dr.Spider, as well as two real-world datasets created for financial and academic applications. The experimental results show that our CodeS achieves new SOTA accuracy and robustness on nearly all challenging text-to-SQL benchmarks.

Evaluating Datalog over Semirings: A Grounding-based Approach

Datalog is a powerful yet elegant language that allows expressing recursive computation. Although Datalog evaluation has been extensively studied in the literature, so far, only loose upper bounds are known on how fast a Datalog program can be evaluated. In this work, we ask the following question: given a Datalog program over a naturally-ordered semiring σ, what is the tightest possible runtime? To this end, our main contribution is a general two-phase framework for analyzing the data complexity of Datalog over σ: first ground the program into an equivalent system of polynomial equations (i.e. grounding) and then find the least fixpoint of the grounding over σ. We present algorithms that use structure-aware query evaluation techniques to obtain the smallest possible groundings. Next, efficient algorithms for fixpoint evaluation are introduced over two classes of semirings: (1) finite-rank semirings and (2) absorptive semirings of total order. Combining both phases, we obtain state-of-the-art and new algorithmic results. Finally, we complement our results with a matching fine-grained lower bound.

The Complexity of Why-Provenance for Datalog Queries

Datalog is a powerful rule-based language that allows us to express complex recursive queries and has found numerous applications over the years. Explaining why a result to a Datalog query is obtained is an essential task towards explainable and transparent data-intensive applications that rely on Datalog. A standard way of explaining a query result is the so-called why-provenance, which provides information about the witnesses to a query result in the form of subsets of the input database that as a whole can be used to derive that result. To our surprise, despite the fact that the notion of why-provenance for Datalog queries has been around for decades and intensively studied, its computational complexity remains unexplored. Our goal is to fill this gap in the why-provenance literature. Towards this end, we pinpoint the data complexity of why-provenance for Datalog queries and key subclasses thereof. The takeaway of our work is that why-provenance for recursive queries, even if the recursion is limited to be linear, is an intractable problem, whereas for non-recursive queries is highly tractable.

Critical Discourse Analysis on Language and Power in Nepali Public ELT Classroom

The study focuses on how teachers use language and power to share knowledge, specifically concerning the power language used in Nepali ELT classrooms. The study's primary goals are to analyze language power as it reveals itself in speech acts and examine how teachers' use of language power affects students' attitudes toward learning. The primary source of data for this study was the teaching activities carried out in the district of Nawalparasi, Nepal. Depending on attributes including gender, job position, and length of service, four public school teachers—two male and two female—were chosen. Using a qualitative ethnographic study approach, the teacher and students' interactions and language use are examined. The data is further evaluated using Norman Fairclough's critical discourse analysis, which comprises description, interpretation, and explanation. The findings suggest that teachers are imposing speech act forms on their students and that this power may have an impact on how they feel about learning. However, students are also motivated and become more engaged in class when teachers express their pleasure in speech act forms. Therefore, teachers should incorporate humanistic teaching into their teaching and learning activities, taking into account the potential problems of using language in an ELT classroom.