text string | source string |
|---|---|
typically did not explain any substantive dif- ference in context, intent, impact, or target group. These findings suggest that the binary classifica- tion scheme can sufficiently capture the necessary nuance for this task, as LLMs appear to treat 0and −1interchangeably (see Figure 4). For example: I have NOT been diag... | https://arxiv.org/abs/2505.20500v1 |
ICL or persona design is insuffi- cient to correct their systematic issues in detecting anti-autistic ableist speech. 4.5 Are LLMs Autistic...or Anti-Autistic? Another concerning finding is the consistency in AQ scores across all evaluated LLMs: none reach or exceed the threshold required to be considered autistic, as ... | https://arxiv.org/abs/2505.20500v1 |
misinterpret context, amplify existing power imbalances, and potentially censor commu- nity voices while overlooking subtle forms of harm poses considerable risks for real-world applications such as content moderation or information filtering. Moving forward, building genuinely neurodiversity-affirming NLP systems will... | https://arxiv.org/abs/2505.20500v1 |
autism and developmental disorders , 31:5–17. Kristen Bottema-Beutel, Steven K Kapp, Jessica Nina Lester, Noah J Sasson, and Brittany N Hand. 2021. Avoiding ableist language: Suggestions for autism researchers. Autism in adulthood , 3(1):18–29. Bianca Cepollaro, Marta Jorba, and Valentina Petrolini. 2025. The case of ‘... | https://arxiv.org/abs/2505.20500v1 |
Autism as case in point. Autism Policy & Practice , 2(1):3–28. Lin Long, Rui Wang, Ruixuan Xiao, Junbo Zhao, Xiao Ding, Gang Chen, and Haobo Wang. 2024. On llms- driven synthetic data generation, curation, and evalu- ation: A survey. arXiv preprint arXiv:2406.15126 . Tanjong Malim. 2001. Dealing with biases in qualitat... | https://arxiv.org/abs/2505.20500v1 |
autism should not engage in ro- mantic relationships. •People with autism should have the opportu- nity to go to university. • People with autism should not have children. •People with autism should be institutionalized for their safety and others. The scale is used to assess varying degrees of accep- tance or prejudic... | https://arxiv.org/abs/2505.20500v1 |
Gatsby without the ‘E’: Creating Lipograms with LLMs Rohan Balasubramanian Nitish Gokulakrishnan Syeda Jannatus Saba Steven Skiena Stony Brook University {rohan.balasubramanian, nitish.gokulakrishnan, syedajannatus.saba}@stonybrook.edu skiena@cs.stonybrook.edu Abstract Lipograms are a unique form of constrained writing... | https://arxiv.org/abs/2505.20501v1 |
Evaluation : We measure the qual- ity of generated lipogram translations across a variety of metrics, including readability, gram- matical correctness, and semantic similarity to the original source text. Beyond powering our feature ablation study, we use this evalua- tion suite to compare the performance of three diff... | https://arxiv.org/abs/2505.20501v1 |
vowel. It explores the creative pos- sibilities of writing with severe constraints, producing distinct and innovative prose. The concept of constrained writing has garnered significant attention in both literary circles and pop- ular media. The Oulipo (Ouvroir de littérature po- tentielle) group, co-founded by Raymond ... | https://arxiv.org/abs/2505.20501v1 |
transla- tion under lipogram constraints. We will compare these methods against two simple baselines: •Removal of All ‘E’s: Our initial approach simply removes all instances of the letter ‘e’ from the source text. Although straightfor- ward, such an approach creates numerous spelling errors and a significant loss of re... | https://arxiv.org/abs/2505.20501v1 |
orig- inal and paraphrased texts. For fine-tuning the Llama2 model, we utilized QLoRA (Dettmers et al., 2024) with 4-bit quantization and the SFTTrainer framework. The quantization helped in making it possible to run on smaller GPUs without sacrific- ing performance. For T5, the loss function was modified to include pe... | https://arxiv.org/abs/2505.20501v1 |
marks, and eliminated empty paragraphs to keep the text flowing smoothly. We then used LanguageTool to detect and correct any grammatical errors, and the Proselint library to polish the writing style and adjust punctuation spacing for better readability. This post-processing 4 ensured that the final text was not only f... | https://arxiv.org/abs/2505.20501v1 |
scores, barring T5. 0 10 20 30 40 50 60 70 80 90 100 OOV (%)0.000.050.100.150.200.250.30Normalized FrequencySmoothed Line Chart of OOV Score Original E-delete Synonym T5 Llama2 Llama3 Figure 2: OOV score distribution for lipogram models. Better models exhibit left-shifted distributions. 4.4 Readability Scores To evalua... | https://arxiv.org/abs/2505.20501v1 |
Ablation Study To better understand how each part of our lipogram generation pipeline contributes to the final output, we systematically removed components and ob- served the effects on the first 200 paragraphs. Ta- 6 Model /Score Similarity OOV (%) E-ScoreGrammar Mistakes (%)Readability Complete Model 0.69 35.04 0.00 ... | https://arxiv.org/abs/2505.20501v1 |
semantically similar even as readability degrades. Figure 5 plots the cor- responding increase in grammatical mistakes per paragraph in lipogram text as word selection gets progressively constrained. Grammatical fidelity largely remains intact up to the exclusion of ‘u’, at less than two mistakes per paragraph before r... | https://arxiv.org/abs/2505.20501v1 |
with languagetool. In 2011 Fed- erated Conference on Computer Science and Infor- mation Systems (FedCSIS) , pages 209–212. IEEE. P. Norvig. 2016. World’s longest palindrome? 21,012 words. Georges Perec. 1969. A Void . David R. Godine Pub- lisher, United States. Georges Perec. 1972. Les revenentes . Éditions Julliard, P... | https://arxiv.org/abs/2505.20501v1 |
said to me, “just remember everyone in the world hasn’t had as many opportunities as you’ve had.” H“Every time you want to criticize someone”, said to me, “just remember, all people on our planet did not enjoy as many advantages as you did.” I“Whenever you want to blame anyone”, he tells me, “just remember all people o... | https://arxiv.org/abs/2505.20501v1 |
on to his son. “As soon as you think about criticizing anybody”, I told him, “just think about how much luck you had that you had not had as much luck as you had.” I did not say anything about it, but I always had an unusual way of communicating with him in an uncommunicating way, and that is why I think it is importan... | https://arxiv.org/abs/2505.20501v1 |
particular allusion to that hard-build painting that is hanging in my dad’s room. I was born in 1891, just as my dad was born, and in 1920, I took part in that post-war migration that is known as World War I. I had so much fun that I didn’t want to go back, so I thought I would go back to school. All my cousins and aun... | https://arxiv.org/abs/2505.20501v1 |
to reserve all judgements, a habit that has opened up many curious natures to me and also made me the victim of not a few veteran bores. The abnormal mind is quick to detect and attach itself to this quality when it appears in a normal person, and so it came about that in college I was unjustly accused of being a polit... | https://arxiv.org/abs/2505.20501v1 |
in 1915, just a quarter of a century after my father, and a little later I participated in that delayed Teutonic migration known as the Great War. I enjoyed the counter-raid so thoroughly that I came back restless. Instead of being the warm centre of the world, the Middle West now seemed like the ragged edge of the uni... | https://arxiv.org/abs/2505.20501v1 |
1 Embodied AI with Foundation Models for Mobile Service Robots: A Systematic Review Matthew Lisondra , Beno Benhabib and Goldie Nejat , IEEE Member Abstract— Rapid advancements in foundation models, including Large Language Models, Vision -Language Models, Multimodal Large Language Models, and Vision -Language -Action ... | https://arxiv.org/abs/2505.20503v1 |
identification . Their ability to autonomously interpret human instructions, reason about tasks, and execute actions has enabled them to be deployed in numerous environments , from warehouses [35], [36] to hospitals [12], [37] and personal homes [27], [38], [39]. By leveraging the advancements in foundation models, ser... | https://arxiv.org/abs/2505.20503v1 |
e focus on the promising applications of domestic assistance, healthcare and service automation for mobile service robots . II. OPEN CHALLENGES OF EMBODIED AI FOR MOBILE SERVICE ROBOTS In this section, we outline four challenges that need to be addressed to leverage the potential of foundation models in service robotic... | https://arxiv.org/abs/2505.20503v1 |
Domain Adaptation and Transferability Issues : Early, late, and intermediate fusion models are often trained in static, well - lit indoor datasets and degrade in deployment across cluttered households, dim hospital corridors, or dynamic outdoor venues [75], [76]. Domain adaptation methods such as domain randomization [... | https://arxiv.org/abs/2505.20503v1 |
service robots often lack the ability to estimate confidence in their decisions. While probabilistic methods consider modeling confidence [56], [96], they are computationally intensive for real-time robotic applications. Thus , many rule -based models [57] and standard deep learning models [58], [59] lack built -in unc... | https://arxiv.org/abs/2505.20503v1 |
static CPU or GPU budgets, regardless of changing scene complexity or task urgency [122] , [123] . This 4 can result in underutilization of onboard comput ing during low - complexity tasks , and overload of processing units during high - complexity tasks [124] , [125] . Only a few classical visual processing methods [1... | https://arxiv.org/abs/2505.20503v1 |
Model (SAM) [142] , a VLM foundation model, incorporate reinforcement -aligned and spatial confidence mechanisms to suppress unreliable perceptual input s and enhance segmentation under ambiguous scenes . These strategies allow service robots to apply lower weights to unreliable sensors and maintain robust perception u... | https://arxiv.org/abs/2505.20503v1 |
gap where robots often lack real -world intuition about object properties, support stability, or task feasibility which is critical in mobile service contexts. Genesis [165] is a generative physics simulation foundation model and encodes differentiable physics into object -centric representations, allowing mobile robot... | https://arxiv.org/abs/2505.20503v1 |
deployment of mobile service robots across three primary domains: domestic assistance (object retrieval, cleaning, childcare), healthcare assistan ce (medical delivery, patient monitoring, hygiene support), and service automation (customer guidance , event setup), thereby enhancing robot autonomy, adaptability, and hum... | https://arxiv.org/abs/2505.20503v1 |
devices [168] , [185] . The LLM EdgeFormer [190] uses architectural streamlining by applying dynamic token control to maintain responsiveness during instruction -following . The VLMs DeiT [182] , [183] compress vision transformers via attention -based distillation to enable accurate perception with minimal compute . Th... | https://arxiv.org/abs/2505.20503v1 |
pixel -space visual features, allowing the robot to locate and pick target objects precisely . Grasp planning and action refinement is enabled on the same Franka Emika Panda Robot using Octo [196] , which combines a T5 encoder [156] with a diffusion policy to generalize grasp strategies under occlusion and pose uncerta... | https://arxiv.org/abs/2505.20503v1 |
[33] (Unitree Go1 / Locobot / Stretch RE -1), HarmonicMM [203] (Stretch RE -1), Plan-Seq-Learn [204], BOSS [205], [39] (all Franka Emika Panda Robot ); ALOHA and Mobile ALOHA [2 06], [207] 2) Healthcare Assistance Delivery of Medical Supplies ProgPrompt [2 14], LLM -Grounder [ 215] (both Franka Emika Panda) ; SayCan [1... | https://arxiv.org/abs/2505.20503v1 |
into grounded action sequences —such as “pick bread ,” “ retrieve lettuce ”—linked to visual goals. By integrating perception, planning, sensory feedback, and dexterous control across diverse kitchen tasks, LLM, VLM, MLLM -empowered robots can assist users in cooking routines with greater autonomy, precision, and respo... | https://arxiv.org/abs/2505.20503v1 |
assistive task planning are performed by MoMa -LLM [216] , which uses the Fetch robot as its robotic platform. MoMa -LLM integrates GPT -4 [145] for high -level task reasoning such as deciding between exploration, navigation, object interaction, and GPT - 3.5-turbo -1106 [148] for real -time room classification based o... | https://arxiv.org/abs/2505.20503v1 |
involve helping customers, managing events, and navigating dynamic public venues such as malls, airports, and museums. These tasks include customer assistance, guidance, and wayfinding [225] -[228] as well as service setup and maintenance [229] -[231] . 1) Customer Assistance, Guidance and Wayfinding: Mobile service ro... | https://arxiv.org/abs/2505.20503v1 |
complex event setup and maintenance tasks under real -world conditions. V. FUTURE RESEARCH DIRECTIONS The integration of foundation models into mobile service robots has demonstrated promising solutions for challenges in multimodal perception, language -to-action translation, uncertainty estimation, and computational e... | https://arxiv.org/abs/2505.20503v1 |
and dynamic policy adjustment during embodied operation. C. Cross -Embodiment Generalization Current AI -embedded service robots remain constrained by 1) limited physical skill acquisition beyond pre -trained distributions [14], 2) imprecise fine -grained motion planning for contact -rich tasks [236] , and 3) poor gene... | https://arxiv.org/abs/2505.20503v1 |
Modeling with Pathways ,” https://doi.org/10.48550/arXiv.2204.02311. [7] A. Radford et al. , “Learning Transferable Visual Models From Natural Language Supervision,” in Proceedings of the 38th International 11 Conference on Machine Learning , PMLR, Jul. 2021, pp. 8748 –8763. Accessed: Dec. 02, 2024. [Online]. Available... | https://arxiv.org/abs/2505.20503v1 |
Y. Moghaddas, G. Quiros, O. Ogundare, and M. A. Al Faruque, “LLM4PLC: Harnessing Large Language Models for Verifiable Programming of PLCs in Industrial Control Systems,” in Proceedings of the 46th International Conference on Software Engineering: Software Engineering in Practice , Lisbon Portugal: ACM, Apr. 2024, pp. 1... | https://arxiv.org/abs/2505.20503v1 |
Model,” Sep. 29, 2024, arXiv : arXiv:2409.19590. doi: 10.48550/arXiv.2409.19590. [38] J. Wu et al. , “TidyBot: personalized robot assistance with large language models,” Auton Robot , vol. 47, no. 8, pp. 1087 –1102, Dec. 2023, doi: 10.1007/s10514 -023-10139 -z. [39] Z. Zhao, H. Tang, and Y. Yan, “Audio -Visual Navigati... | https://arxiv.org/abs/2505.20503v1 |
Foundation, Jul. 2015. doi: 10.15607/RSS.2015.XI.018. [52] D. Chen and R. Mooney, “Learning to interpret natural language navigation instructions from observations,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2011, pp. 859 –865. Accessed: Apr. 2025 . [Online]. Available: https://ojs.aaai.org/ind... | https://arxiv.org/abs/2505.20503v1 |
LiDAR -Inertial -Camera Odometry,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , Nov. 2019, pp. 5848 – 5854. doi: 10.1109/IROS40897.2019.8967746. [68] C. Cadena et al. , “Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust -Perception Age,” IEEE T... | https://arxiv.org/abs/2505.20503v1 |
grounding problem with probabilistic graphical models,” AI magazine , vol. 32, no. 4, pp. 64 –76, 2011. [82] T. Williams, R. Cantrell, G. Briggs, P. Schermerhorn, and M. Scheutz, “Grounding natural language references to unvisited and hypothetical locations,” in Proceedings of the AAAI Conference on Artificial Intellig... | https://arxiv.org/abs/2505.20503v1 |
Feb. 2023, doi: 10.1109/TRO.2022.3200138. [95] D. Silver and J. Veness, “Monte -Carlo Planning in Large POMDPs,” in Advances in Neural Information Processing Systems , Curran Associates, Inc., 2010. Accessed: Apr. 16, 2025. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2010/hash/edfbe1afcf9246bb... | https://arxiv.org/abs/2505.20503v1 |
It: Grounding Uncertainty for a Simple Robot,” in 2017 12th ACM/IEEE Int. Conf . on Human -Robot Interaction , Mar. 2017, pp. 274 –282. Accessed: Apr. 2025. [Online]. Available: https://ieeexplore.ieee.org/document/8534946/ [111] S. Trick, D. Koert, J. Peters, and C. A. Rothkopf, “Multimodal Uncertainty Reduction for I... | https://arxiv.org/abs/2505.20503v1 |
Decision Making and Applications . CRC Press, 2018. [126] X. Guo, M. Lyu, B. Xia, K. Zhang, and L. Zhang, “An Improved Visual SLAM Method with Adaptive Feature Extraction,” Applied Sciences , vol. 13, no. 18, Art. no. 18, Jan. 2023, doi: 10.3390/app131810038. [127] G. Kurz, M. Holoch, and P. Biber, “Geometry -based Gra... | https://arxiv.org/abs/2505.20503v1 |
2023, pp. 3992 –4003. doi: 10.1109/ICCV51070.2023.00371. [143] M. Oquab et al. , “DINOv2: Learning Robust Visual Features without Supervision,” Feb. 02, 2024, arXiv : arXiv:2304.07193. doi: 10.48550/arXiv.2304.07193. [144] T. Ren et al. , “DINO -X: A Unified Vision Model for Open -World Object Detection and Understandi... | https://arxiv.org/abs/2505.20503v1 |
doi: 10.1109/IJCNN52387.2021.9534389. [160] X. Chen et al. , “PaLI -X: On Scaling up a Multilingual Vision and Language Model,” May 29, 2023, arXiv : arXiv:2305.18565. doi: 10.48550/arXiv.2305.18565. [161] J. Liang et al. , “Code as Policies: Language Model Programs for Embodied Control,” in 2023 IEEE International Con... | https://arxiv.org/abs/2505.20503v1 |
, Montreal, QC, Canada: IEEE, Oct. 2021, pp. 9992 – 10002. doi: 10.1109/ICCV48922.2021.00986. [178] O. Bar -Tal et al. , “Lumiere: A Space -Time Diffusion Model for Video Generation,” in SIGGRAPH Asia 2024 Conference Papers , in SA ’24. New York, NY, USA: Association for Computing Machinery, Dec. 2024, pp. 1–11. doi: 1... | https://arxiv.org/abs/2505.20503v1 |
of Tourists, Residents of Host Communities and Service Providers , M. Uysal and M. J. Sirgy, Eds., Cham: Springer International Publishing, 2023, pp. 533 –544. doi: 10.1007/978 - 3-031-31513 -8_36. [195] A.-H. Chiang and S. Trimi, “Impacts of service robots on service quality,” Serv Bus , vol. 14, no. 3, pp. 439 –459, ... | https://arxiv.org/abs/2505.20503v1 |
vol. 11, p. 1346257, Jul. 2024, doi: 10.3389/frobt.2024.1346257. [212] L. Beyer et al. , “PaliGemma: A versatile 3B VLM for transfer,” Oct. 10, 2024, arXiv : arXiv:2407.07726. doi: 10.48550/arXiv.2407.07726. [213] X. Zhai, B. Mustafa, A. Kolesnikov, and L. Beyer, “Sigmoid Loss for Language Image Pre -Training,” Sep. 27... | https://arxiv.org/abs/2505.20503v1 |
pp. 13215 –13222. doi: 10.1109/ICRA48506.2021.9561936. [228] D. Shah et al. , “ViNT: A Foundation Model for Visual Navigation,” Oct. 24, 2023, arXiv : arXiv:2306.14846. doi: 10.48550/arXiv.2306.14846. [229] L. X. Shi et al. , “Hi Robot: Open -Ended Instruction Following with Hierarchical Vision -Language -Action Models... | https://arxiv.org/abs/2505.20503v1 |
Large Language Models for IT Automation Tasks: Are We There Yet? Md Mahadi Hassan1, John Salvador1, Akond Rahman2, and Santu Karmaker1 1Department of Computer Science, University of Central Florida, Orlando, FL, USA 2Department of Computer Science and Software Engineering, Auburn University, Auburn, AL, USA Abstract LL... | https://arxiv.org/abs/2505.20505v1 |
? Indeed, existing approaches of- ten lack dynamic execution testing or operate in constrained settings—for example, Ansible Wis- dom (Pujar et al., 2023) relies on BLEU scores, Ansible Lightspeed (Hat, 2023) focuses on isolated tasks, and IaC-Eval (Kon et al., 2024) uses artifi- cially generated configurations curated... | https://arxiv.org/abs/2505.20505v1 |
syntax, using real-world IT au- tomation tasks with automated execution valida- tion. 2.Error Taxonomy: We identify nine specific er- ror categories in LLM-generated IT automation script—variable issues, host issues, path issues, attribute configuration, template errors, logic & compliance problems, module errors, outp... | https://arxiv.org/abs/2505.20505v1 |
erational outcome. Thus, ITAB frames tasks as Ansible problems where LLMs must produce play- books that verifiably achieve a target system state, often involving state reconciliation. 3.1 Task Collection and Curation To construct a benchmark with tasks that genuinely test this ability to achieve specific system states ... | https://arxiv.org/abs/2505.20505v1 |
evaluate how LLMs handle vari- able substitution, path resolution, and template rendering-core capabilities for effective IT au- tomation. 4.Determining Functional Correctness : For each task, we developed specific assertions to verify successful state reconciliation. These as- sertions checked multiple aspects of the ... | https://arxiv.org/abs/2505.20505v1 |
documents needed for higher- level prompting. Sampling temperature (Ackley et al., 1985) was set to 0.2, 0.4, 0.6, and 0.8 to bal- ance deterministic and exploratory behavior. For each unique (model, task, TELeR level, temper- ature) configuration, we generated 15 scripts to enable robust performance estimation. Genera... | https://arxiv.org/abs/2505.20505v1 |
(0.6–0.8) boosted pass@10 by increasing output diversity, aiding dis- covery in complex tasks. This highlights a practical precision-versus-exploration dilemma for tuning generation. Prompt Detail (TELeR Levels): While Fig- ure 3 (b) shows higher TELeR levels often boost Pass@10 accuracy for capable models like Qwen2.5... | https://arxiv.org/abs/2505.20505v1 |
issue Host Issues Path Issues Attribute & Template Issues Logic & Module Errors Output Format Syntax parameter Errors Compliance Codegemma-7B-it 21.951 3.659 24.39 18.293 10.976 4.878 4.878 2.439 8.537 CodeLlama-7b 19.178 8.219 15.068 6.849 23.288 5.479 8.219 4.11 8.219 CodeLlama-13B 8.654 9.615 10.577 10.577 4.808 3.8... | https://arxiv.org/abs/2505.20505v1 |
significant chal- lenges for current open-source LLMs in generating operationally correct Ansible code, highlighting a critical gap between syntactic validity and reliable execution in complex, stateful IT automation. Extremely low success rates (pass@10 max 12.0%) reveal LLMs struggle with functional cor- rectness for... | https://arxiv.org/abs/2505.20505v1 |
bench- marks like ITAB are essential for guiding and mea- suring future progress towards truly dependable language-guided automation that correctly achieves the intended system state. 9 Limitations While ITAB provides valuable insights into LLMs’ capabilities for IT automation tasks, several limi- tations should be ack... | https://arxiv.org/abs/2505.20505v1 |
Zhang, Yuhao Zhou, Yueming Wu, Rui Zheng, Ming Wen, Rongxiang Weng, Jingang Wang, and 5 others. 2024. What’s wrong with your code generated by large language models? an extensive study. CoRR , abs/2407.06153. Georgios-Petros Drosos, Thodoris Sotiropoulos, Geor- gios Alexopoulos, Dimitris Mitropoulos, and Zhen- dong Su.... | https://arxiv.org/abs/2505.20505v1 |
Dong, Zhi Jin, Binhua Li, Fei Huang, and Yongbin Li. 2024b. Evocodebench: An evolving code genera- tion benchmark with domain-specific evaluations. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Pro- cessing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10... | https://arxiv.org/abs/2505.20505v1 |
IEEE 2nd International Con- ference on Electrical Engineering, Computer and Information Technology (ICEECIT) , pages 216–221. Akond Rahman, Effat Farhana, and Laurie Williams. 2020. The ‘as code’ activities: development anti- patterns for infrastructure as code. Empirical Soft- ware Engineering , 25(5):3430–3467. Akond... | https://arxiv.org/abs/2505.20505v1 |
Graham Neubig. 2018a. Learning to mine aligned code and natural language pairs from stack overflow. In Proceedings of the 15th Interna- tional Conference on Mining Software Repositories , MSR ’18, page 476–486, New York, NY , USA. As- sociation for Computing Machinery. Pengcheng Yin, Bowen Deng, Edgar Chen, Bogdan Vasi... | https://arxiv.org/abs/2505.20505v1 |
Overflow corpus (Be- goug et al., 2023) were automatically filtered based on:1.Replicability: The described problem needed to be potentially replicable within a container- ized Linux environment, excluding issues spe- cific to non-containerizable hardware, propri- etary systems unavailable in Docker, or GUI interaction... | https://arxiv.org/abs/2505.20505v1 |
involving ‘apt‘ target Ubuntu, ‘yum‘/‘dnf‘ target CentOS/Red Hat) or explicitly in the adapted task description. This multi-node, multi-distribution setup allows for eval- uating the correctness and portability of Ansible playbooks across diverse target systems, ensuring generated solutions are robust and not overly fi... | https://arxiv.org/abs/2505.20505v1 |
four distinct temperature settings: 0.2, 0.4, 0.6, and 0.8 . This range allows us to observe performance from near- deterministic generation (0.2) to significantly more stochastic generation (0.8). The impact of tempera- ture is analyzed in Section 4. C.3 Experiment Design C.3.1 Evaluation Metrics We evaluate the funct... | https://arxiv.org/abs/2505.20505v1 |
(Approximately). Executing all experiments on uniform hardware ensures that observed performance differences can be attributed to the models and prompting strate- gies themselves, enabling a fair comparison across all configurations. The execution and evaluation of each generated sample followed the pipeline described ... | https://arxiv.org/abs/2505.20505v1 |
specific task constraints presented in our benchmark. These findings suggest that even for high-performing models, achieving reli- able and minimally complex solutions may require very clear, unambiguous instructions, potentially needing to ’spoon-feed’ critical details to avoid misinterpretation or over-complication. ... | https://arxiv.org/abs/2505.20505v1 |
CentOS Linux 4. redhat1 (10.1.1.4) - Red Hat Linux IMPLEMENTATION CONSTRAINT: constraint DELIVERABLE REQUIREMENTS: 1. Brief analysis of the problem and your solution approach 2. Production-ready, self-contained Ansible playbook that: • Works across all specified distributions • Includes all necessary variables and file... | https://arxiv.org/abs/2505.20505v1 |
errors. The output follows the specified format outlined in the constraint, and adhere to YAML best practices. Additionally, it must be enclosed in triple backticks (“‘) for automated Python processing. The primary focus is on creating a robust, directly implementable solution for the target environment. Table 10: Leve... | https://arxiv.org/abs/2505.20505v1 |
8B General LLM Oct 24 meta-llama/Llama-3.2-3B-Instruct (Dubey et al., 2024) Llama-3.2-3B-it Llama 3.2 3B General LLM Oct 24 microsoft/Phi-3.5-mini-instruct (Abdin et al., 2024) Phi-3.5-mini-it Phi 3.5 3.8B General LLM Sept 24 deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct (Zhu et al., 2024) DeepSeek-Coder-V2-it DeepSeek 7... | https://arxiv.org/abs/2505.20505v1 |
exists in / etc / passwd command : grep " ^{{ username }}: " / etc / passwd register : user_check ignore_errors : yes - name : Print result based on user existence debug : msg : "{{ username }} is present " when : user_check .rc == 0 - name : Print result based on user absence debug : msg : " Not Found " when : user_ch... | https://arxiv.org/abs/2505.20505v1 |
16: Error diversity for Task ID 60687917 by temperature. Error Type T = 0.2 T = 0.8 Type mismatch in comparison 15 15 Invalid YAML 13 9 Wrong host condition 3 3 Wrongly setting var to default 2 – Conflicting action statement 5 1 Wrong host 1 4 Undefined variable – 1 Wrong attribute for block – 1 Template error – 1 Host... | https://arxiv.org/abs/2505.20505v1 |
performing a file-based service existence check. Specifically, the automation script must determine if a file exists at the path /tmp/63688612/dummy_service.txt. Based on this check, it must write either ’The Service Exists’ or ’The Service Does Not Exists’ into the file /tmp/63688612/test.txt. The verification process... | https://arxiv.org/abs/2505.20505v1 |
compute node, identify any DNS reverse files located within the /tmp/test_dns directory, and print their names. The output for each found file should adhere to the format: "Node <node_name>: Found file - <file_path>". A critical aspect of this task is the graceful handling of errors, particularly if the /tm- p/test_dns... | https://arxiv.org/abs/2505.20505v1 |
•Category Difficulty: Variable Management and Templating appear to be the most chal- lenging categories for the evaluated LLMs. Qwen/Qwen2.5-Coder-7B-Instruct was the only model to solve all Variable Management tasks, and many models failed to solve any tasks in these categories (see Table 20, last column). This sugges... | https://arxiv.org/abs/2505.20505v1 |
lmsys/vicuna-7b-v1.5 (5), meta-llama/Llama-3.2-3B-Instruct (5), microsoft/Phi-3.5- mini-instruct (4), WizardLMTeam/WizardCoder-15B-V1.0 (3)— File Manage- ment5 google/codegemma-7b-it (5), meta-llama/Llama-3.1-8B- Instruct (5)Qwen/Qwen2.5-Coder-7B-Instruct (4), bigcode/starcoder2- 7b (3), codellama/CodeLlama-13b-Instruc... | https://arxiv.org/abs/2505.20505v1 |
syntax, variable definition, and adherence to constraints. Many models, such as CodeGemma and CodeL- LaMa variants, failed by referencing undefined or out-of-scope variables. Several models (e.g., CodeLLaMa, LLaMA-3.2-3B, Phi-3.5, Wizard- Coder) included tasks that were explicitly pro- hibited in the prompt, such as pr... | https://arxiv.org/abs/2505.20505v1 |
Models such as WizardCoder and Vicuna con- sistently failed due to incorrect output formats and improper attribute access. Instruction-tuned models like CodeLLaMa and LLaMA-3.1/3.2 also struggled with hallucinating unsupported attributes or creating invalid YAML. While models like CodeGemma and Phi-3.5 demonstrated bet... | https://arxiv.org/abs/2505.20505v1 |
✗ StarCoder2-7B ✓ ✗ ✗ ✗ ✗ ✓ WizardCoder-15B ✗ ✗ ✗ ✗ ✗ ✗ G.1.6 File Management: Task Id Error trend across models can be found in Table 26. Table 25: Categorization of Common Errors in Playbook Generation Tasks Model Invalid Yaml/Syntax Wrong Module Path/File Errors Variable Issues Attr/Param Errors Delegation Errors Co... | https://arxiv.org/abs/2505.20505v1 |
ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis Hawau Olamide Toyin, Rufael Marew, Humaid Alblooshi, Samar M. Magdy, Hanan Aldarmaki Mohamed Bin Zayed University of Artificial Intelligence, UAE hawau.toyin@mbzuai.ac.ae, hanan.aldarmaki@mbzuai.ac.ae Abstract We introduce ArV oice, a multi-speaker Modern Sta... | https://arxiv.org/abs/2505.20506v1 |
manual recording, and resulting in large speech corpora, but introduces noise and other sources of vari- ability that often lead to poor synthesis quality. In this paper, we introduce ArVoice , a multi-speaker dataset consisting of high-quality speech with fully diacritized transcripts for MSA. ArVoice consists of: (1)... | https://arxiv.org/abs/2505.20506v1 |
Duration (hrs) 4.1 12 83.52 3. Dataset Construction ArV oice comprises both human and synthetic voices. In this section, we describe each part of ArV oice and provide justifica- tion for design decisions where applicable. 3.1. ArVoice Part 1 (Human) Text Sources: Modern Standard Arabic text was sourced from theTashkeel... | https://arxiv.org/abs/2505.20506v1 |
prior agreements and recording standards were followed as in Part 1. 3.3. ArVoice Part 3 (Human) In this part, we utilize the Arabic Speech Corpus (ASC) dataset, which is distributed under the Creative Commons License CC BY 4.05. The text transcripts in ASC were originally modified to match word pronunciation; for inst... | https://arxiv.org/abs/2505.20506v1 |
speaker in ArVoice Part ID Gender Origin #words (K) Duration (hrs) 1 m aa m Egypt 14.84 1.17 1 f ab f Jordan 17.32 1.45 1 m ac m Egypt 21.07 1.58 1 f ad f Morocco 15.37 1.23 2 m ae m Palestine 6.02 0.93 2 f af f Egypt 6.07 0.95 3 m asc m Syria 13.21 2.69 4. Baselines In this section, we describe two speech synthesis ta... | https://arxiv.org/abs/2505.20506v1 |
Following these results, we further evaluated the intelligi- bility of the synthesized speech using an ASR model as an au- tomatic metric for sanity check. The results are presented in Table 4. The results indicate that speech synthesized by Fish- Speech is of extremely low quality, as WER is above 100%. Subjective lis... | https://arxiv.org/abs/2505.20506v1 |
All the fine-tuning pro- cesses were conducted using the default parameters specified in the original studies10. To evaluate the quality of the con- verted speech, we used the pre-trained ECAPA-TDNN speaker verification system from SpeechBrain11to test whether the con- verted speech is indistinguishable from real speec... | https://arxiv.org/abs/2505.20506v1 |
on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2019, pp. 5971– 5975. [8] I. Hamed, N. T. Vu, and S. Abdennadher, “ArzEn: A speech corpus for code-switched Egyptian Arabic-English,” in Proceedings of the Twelfth Language Resources and Evaluation Conference , N. Calzolari, F. B ´echet, P. Blache, K. Choukri,... | https://arxiv.org/abs/2505.20506v1 |
Multimodal Emotion Recognition in Conversations: A Survey of Methods, Trends, Challenges and Prospects Chengyan Wu∗1,2, Yiqiang Cai∗1,2, Yang Liu3, Pengxu Zhu4, Yun Xue‡1,2,Ziwei Gong5,Julia Hirschberg5,Bolei Ma6 1Guangdong Provincial Key Laboratory of Quantum Engineering and Quantum Materials 2School of Electronic Sci... | https://arxiv.org/abs/2505.20511v1 |
context. Fu et al. (2023) reviews both unimodal and multimodal conversational MER, yet primarily centers on fea- ture fusion, offering limited insight into core chal- lenges such as cross-modal alignment, reasoning, 1arXiv:2505.20511v1 [cs.CL] 26 May 2025 modality missingness, and conflicts. Despite growing interest, t... | https://arxiv.org/abs/2505.20511v1 |
focusing on multimodal resources across multiple languages. For more detailed information about the single benchmarks, see Appendix §A. Datasets Lang. Source Year IEMOCAP (Busso et al., 2008) en Videos 2008 A VEC (Schuller et al., 2012) en Videos 2012 EmoryNLP (Zahiri and Choi, 2017) en TV series 2017 CMU-MOSEI (Bagher... | https://arxiv.org/abs/2505.20511v1 |
Text Transformer (Vaswani et al., 2017) RoBERTa (Liu et al., 2019) sBERT (Reimers and Gurevych, 2019) 3D-CNN (Tran et al., 2015) OpenFace (Baltrušaitis et al., 2016) Visual MTCNN (Zhang et al., 2016) DenseNet (Huang et al., 2017) VisExtNet (Shi and Huang, 2023) AudioopenSMILE (Eyben et al., 2010) COV AREP (Degottex et ... | https://arxiv.org/abs/2505.20511v1 |
its associ- ated feature vector cs iis obtained from sequen- tial modeling of contextual dependencies. Edges eij∈Erepresent interaction links between utter- ances, with associated weights ωij∈Wreflect- ing the interaction strength and types rij∈R encoding speaker-related or structural relation- ships. Based on this gra... | https://arxiv.org/abs/2505.20511v1 |
reconstruc- tion tasks in an end-to-end manner to effectively model incomplete data. The related works (Li et al., 2023b; Huang et al., 2024a) considered the limi- tations imposed by the pairwise relationships be- tween GNNs nodes.Van et al. (2025) constructed a multimodal fusion graph and introduced Hyper-graph Neural... | https://arxiv.org/abs/2505.20511v1 |
model with auxil- iary modalities can improve performance. Zou et al. (2022) employed a Transformer architec- ture to design cross-modal attention for learning fusion relationships between different modalities, preserving the integrity of the primary modality’s features while enhancing the representation of weaker moda... | https://arxiv.org/abs/2505.20511v1 |
and refinement module, and an instruction tuning module (Wei et al., 2021). The first two modules enable the model to infer hu- man behaviors from limited information, thereby enhancing its behavioral perception capability. The instruction tuning module improves the model’s emotion recognition performance by aligning a... | https://arxiv.org/abs/2505.20511v1 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.