pith. sign in

arxiv: 2606.23321 · v1 · pith:SGBKT7LNnew · submitted 2026-06-22 · 💻 cs.CL

Tmax: A simple recipe for terminal agents

Hamish Ivison , Junjie Oscar Yin , Rulin Shao , Teng Xiao , Nathan Lambert , Hannaneh Hajishirzi This is my paper

Pith reviewed 2026-06-26 08:14 UTC · model grok-4.3

classification 💻 cs.CL
keywords terminal agentsreinforcement learningdata generation taxonomylanguage model trainingoutcome-only RLopen datasetterminal environments
0
0 comments X

The pith

A simple data-generation taxonomy and outcome-only RL lets 9B models hit 27 percent on terminal benchmarks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a recipe called Tmax for training language models as terminal agents using reinforcement learning. It generates large amounts of training data by combining difficulty control, personas, and verifier diversification in a novel taxonomy. This data supports both supervised fine-tuning and RL with only outcome rewards. The approach yields models that outperform much larger prior models on a standard benchmark while releasing a dataset more than twice as large as previous ones.

Core claim

By generating terminal environments through a taxonomy that mixes difficulty control, personas, and verifier diversification, and then training with a simple outcome-only RL procedure on the resulting data, 9B parameter models reach 27 percent success on Terminal-Bench 2.0 and surpass larger models from earlier work.

What carries the argument

The novel taxonomy that combines difficulty control, personas, and verifier diversification to cheaply produce large volumes of terminal environments for training.

If this is right

  • Open-weight 9B models trained this way achieve 27 percent on the benchmark.
  • The released dataset exceeds prior terminal-agent datasets by more than 2.5 times.
  • Outcome-only RL on this data produces effective terminal agents without complex reward shaping.
  • Releasing the data, models, and code establishes a reproducible baseline for further research.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar taxonomy-based data generation might improve training for other kinds of agents beyond terminals.
  • Outcome-only signals could prove sufficient for many agent tasks if paired with diverse enough environments.
  • Scaling up the number of generated environments might push performance higher without increasing model size.

Load-bearing premise

The generated environments from the taxonomy create training signals that transfer effectively to real terminal tasks when used with outcome-only RL.

What would settle it

Training a model with the described data and recipe and then measuring success below the performance of larger prior models on Terminal-Bench 2.0 would falsify the claim.

Figures

Figures reproduced from arXiv: 2606.23321 by Hamish Ivison, Hannaneh Hajishirzi, Junjie Oscar Yin, Nathan Lambert, Rulin Shao, Teng Xiao.

Figure 1
Figure 1. Figure 1: Performance of TMAX models compared to prior work with open data and selected closed and open￾weight models on Terminal-Bench 2.0. For TMAX and Qwen models, we report scores using our simple harness. TMAX outperforms prior work with open data (especially prior open RL recipes) and dominates the Pareto curve for models under 32B parameters. For further details see §E.1. Abstract Terminal-using agents have q… view at source ↗
Figure 2
Figure 2. Figure 2: TMAX Data Pipeline. Each task is composed by hierarchically sampling from 9 structured axes, after which a data generator instantiates into a Dockerfile, unit-test verifier, source files, and task instructions. Tasks are built atop a pre-built per-domain base image and served through a mini-SWE-agent harness. Composing axes yields combinatorially many task signatures with explicit, per-axis control over di… view at source ↗
Figure 3
Figure 3. Figure 3: Data Composition. Domain distribution of tasks across terminal datasets. Prior datasets skew heavily toward one or two domains, whereas our compositional sampler yields balanced coverage across all nine domains. Data Dataset size Pass@1 Gemini Pass@4 Gemini Pass@8 Gemini Mean turns Mean tokens / run (K) Domain balance Skill-type balance TMax (Ours) 15k 42% 50% 53% 16.3 120K 0.998 0.732 Endless Terminals 2.… view at source ↗
Figure 4
Figure 4. Figure 4: Maximum difference between inference (vLLM) and trainer (HuggingFace) logprobs during the first 100 steps of RL training for Qwen 3.5 9B. Qwen 3.5 without the FP32 LM head displays larger and more frequent spikes. Qwen 3 8B training also does not dis￾play spikes, even though we do not apply the FP32 LM head. LM head reduces the maximum difference dramat￾ically. Interestingly, we find this is less important… view at source ↗
Figure 5
Figure 5. Figure 5: Average step count over RL training when [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Average length (in tokens) of assistant turns [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Using DPPO limits training collapse. Aver￾age training reward when doing RL training on TMAX￾15K using GRPO or DPPO. See §D.4 for GRPO details. 0 100 200 300 350 Training steps 0.0 0.2 0.4 0.6 0.8 Avg Train Reward G=32 G=8 [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Using a larger group size improves stability. Average training reward when doing RL training on TMAX-15K using DPPO with varied group sizes (8, 32). ence more common. To reduce the impact of mismatches, we used an FP32 LM head, which aided in reducing mismatches as seen in [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Pass@k difficulty curves. Mean pass@k (k = 1–8) for Gemini-3-Flash-Preview on a fixed 250- task subsample per dataset (8 rollouts each); lower is harder. TMAX occupies the hardest band together with CLI-Gym and attains the lowest pass@8 of any dataset, showing that its difficulty persists as k grows, whereas easy datasets such as Endless-Terminals saturate near the ceiling. B.3 Balance score The balance sc… view at source ↗
Figure 11
Figure 11. Figure 11: Number of all-zero samples filtered over the [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Number of samples filtered for perfect solv [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
read the original abstract

Terminal-using agents have quickly become the most popular downstream application of language models (LMs). Despite their prevalence, relatively little academic work has examined RL-based training of these models, likely due to difficult benchmarks, a lack of data, and a lack of simple baseline recipes. We present Tmax, the strongest open RL recipe for terminal agents to date, bringing open data recipes closer to the frontier. While simple, our recipe achieves 27\% on Terminal-Bench 2.0 with only 9B parameters, outperforming much larger models from prior work. Concretely, we generate data using a novel taxonomy, combining difficulty control, personas, and verifier diversification, which allows us to cheaply generate large amounts of terminal environments for RL and SFT training. We open-source our terminal dataset, which is over 2.5x larger than previously released terminal-agent datasets. We then train open-weight models using RL with our data, using a simple, outcome-only recipe. We release our data, models, and code as a strong baseline for future open academic work on terminal agents at https://github.com/hamishivi/tmax.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The paper presents Tmax, a simple RL-based recipe for training terminal agents. It proposes a novel taxonomy for generating terminal environments that combines difficulty control, personas, and verifier diversification to efficiently create large amounts of training data for SFT and RL. Using this data with outcome-only RL, the authors train a 9B parameter model that achieves 27% on Terminal-Bench 2.0, outperforming larger models from prior work. The dataset released is over 2.5x larger than previous ones, and the authors open-source the data, models, and code.

Significance. This work provides a strong, reproducible baseline for open research on terminal agents by demonstrating competitive performance with a relatively small model through careful data generation and simple training. The release of extensive artifacts (dataset, models, code) is a notable strength that facilitates verification and further development in the field.

minor comments (2)
  1. [Abstract] Abstract: the claim of outperforming 'much larger models from prior work' would be strengthened by naming the specific models, their parameter counts, and citations for direct comparison.
  2. [Abstract] Abstract: no evaluation details (baselines, number of runs, error bars, or verification of comparisons) are supplied, which limits immediate assessment of the 27% result even if these appear in later sections.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary, significance assessment, and recommendation of minor revision. No major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an empirical recipe: a data-generation taxonomy (difficulty control + personas + verifier diversification) used to create a large terminal-agent dataset, followed by outcome-only RL/SFT training of 9B models that reach 27% on Terminal-Bench 2.0. No equations, fitted parameters renamed as predictions, self-definitional loops, or load-bearing self-citations appear in the derivation chain. The central claim is a reproducible performance number backed by released artifacts, making the work externally falsifiable rather than internally circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are described beyond the high-level claim of a novel taxonomy.

pith-pipeline@v0.9.1-grok · 5744 in / 1110 out tokens · 30930 ms · 2026-06-26T08:14:21.001876+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

300 extracted references · 7 canonical work pages · 2 internal anchors

  1. [1]

    Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces

    Terminal-bench: Benchmarking agents on hard, realistic tasks in command line interfaces , author=. arXiv preprint arXiv:2601.11868 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Team, OpenThoughts-Agent , month = Dec, title =

  3. [3]

    Endless terminals: Scaling rl environments for terminal agents.arXiv preprint arXiv:2601.16443,

    Endless Terminals: Scaling RL Environments for Terminal Agents , author=. arXiv preprint arXiv:2601.16443 , year=

    work page arXiv

  4. [4]

    NL 2 B ash: A Corpus and Semantic Parser for Natural Language Interface to the Linux Operating System

    Lin, Xi Victoria and Wang, Chenglong and Zettlemoyer, Luke and Ernst, Michael D. NL 2 B ash: A Corpus and Semantic Parser for Natural Language Interface to the Linux Operating System. Proceedings of the Eleventh International Conference on Language Resources and Evaluation ( LREC 2018). 2018

    2018

  5. [5]

    DeepSWE: Training a State-of-the-Art Coding Agent from Scratch by Scaling RL , author=

  6. [6]

    2025 , howpublished =

    Anthropic , title =. 2025 , howpublished =

    2025

  7. [7]

    2026 , eprint=

    Composer 2 Technical Report , author=. 2026 , eprint=

    2026

  8. [8]

    2026 , eprint=

    Computer Environments Elicit General Agentic Intelligence in LLMs , author=. 2026 , eprint=

    2026

  9. [9]

    Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles , year=

    Efficient Memory Management for Large Language Model Serving with PagedAttention , author=. Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles , year=

  10. [10]

    2025 , month = nov, howpublished =

    Interleaved Thinking Unlocks Reliable. 2025 , month = nov, howpublished =

    2025

  11. [11]

    2025 , eprint=

    SWE-smith: Scaling Data for Software Engineering Agents , author=. 2025 , eprint=

    2025

  12. [12]

    2025 , eprint=

    DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models , author=. 2025 , eprint=

    2025

  13. [13]

    Wang, Junli and Cheng, Zhoujun and Zhang, Yuxuan and Hao, Shibo and Tang, Yao and Hu, Zhiting and Ammanabrolu, Prithviraj and Zhang, Hao , year =

  14. [14]

    OpenThoughts-Agent team, Snorkel AI, Bespoke Labs , month = Feb, title =

  15. [15]

    2026 , eprint=

    CLI-Gym: Scalable CLI Task Generation via Agentic Environment Inversion , author=. 2026 , eprint=

    2026

  16. [16]

    2025 , eprint=

    MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention , author=. 2025 , eprint=

    2025

  17. [17]

    2026 , eprint=

    Olmo 3 , author=. 2026 , eprint=

    2026

  18. [18]

    2025 , eprint=

    DAPO: An Open-Source LLM Reinforcement Learning System at Scale , author=. 2025 , eprint=

    2025

  19. [19]

    2024 , eprint=

    DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models , author=. 2024 , eprint=

    2024

  20. [20]

    2025 , eprint=

    Scaling Synthetic Data Creation with 1,000,000,000 Personas , author=. 2025 , eprint=

    2025

  21. [21]

    2026 , eprint=

    Rethinking the Trust Region in LLM Reinforcement Learning , author=. 2026 , eprint=

    2026

  22. [22]

    2025 , eprint=

    Tulu 3: Pushing Frontiers in Open Language Model Post-Training , author=. 2025 , eprint=

    2025

  23. [23]

    Guo, Daya and Yang, Dejian and Zhang, Haowei and Song, Junxiao and Wang, Peiyi and Zhu, Qihao and Xu, Runxin and Zhang, Ruoyu and Ma, Shirong and Bi, Xiao and Zhang, Xiaokang and Yu, Xingkai and Wu, Yu and Wu, Z. F. and Gou, Zhibin and Shao, Zhihong and Li, Zhuoshu and Gao, Ziyi and Liu, Aixin and Xue, Bing and Wang, Bingxuan and Wu, Bochao and Feng, Bei ...

    work page doi:10.1038/s41586-025-09422-z

  24. [24]

    LiteCoder: Advancing Small and Medium-sized Code Agents , author=

  25. [25]

    2026 , month = jan, url =

    Qijia Shen and Jay Rainton and Aznaur Aliev and Ahmed Awelkair and Boyuan Ma and Zhiqi (Julie) Huang and Yuzhen Mao and Wendong Fan and Philip Torr and Bernard Ghanem and Changran Hu and Urmish Thakker and Guohao Li , title =. 2026 , month = jan, url =

    2026

  26. [26]

    2026 , eprint=

    SERA: Soft-Verified Efficient Repository Agents , author=. 2026 , eprint=

    2026

  27. [27]

    2026 , eprint=

    Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem , author=. 2026 , eprint=

    2026

  28. [28]

    2024 , eprint=

    Training Software Engineering Agents and Verifiers with SWE-Gym , author=. 2024 , eprint=

    2024

  29. [29]

    Advances in Neural Information Processing Systems , volume=

    Language Models are Few-Shot Learners , author=. Advances in Neural Information Processing Systems , volume=

  30. [30]

    2023 , eprint=

    Hugo Touvron and Thibaut Lavril and Gautier Izacard and Xavier Martinet and Marie-Anne Lachaux and Timoth. 2023 , eprint=

    2023

  31. [31]

    2025 , eprint=

    SWE-rebench: An Automated Pipeline for Task Collection and Decontaminated Evaluation of Software Engineering Agents , author=. 2025 , eprint=

    2025

  32. [32]

    arXiv preprint arXiv:2602.21193 , year=

    On Data Engineering for Scaling LLM Terminal Capabilities , author=. arXiv preprint arXiv:2602.21193 , year=

    work page arXiv

  33. [33]

    Termigen: High-fidelity environment and robust trajectory synthesis for terminal agents.arXiv preprint arXiv:2602.07274,

    TermiGen: High-Fidelity Environment and Robust Trajectory Synthesis for Terminal Agents , author=. arXiv preprint arXiv:2602.07274 , year=

    work page arXiv

  34. [34]

    arXiv preprint arXiv:2602.01244 , year=

    Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments , author=. arXiv preprint arXiv:2602.01244 , year=

    work page arXiv

  35. [35]

    Mountain View, CA: Google)

    A new era of intelligence with Gemini 3 , author=. Mountain View, CA: Google). Available online at: https://blog. google/products-andplatforms/products/gemini/gemini-3/(Accessed February 1, 2026) , year=

    2026

  36. [36]

    2026 , eprint=

    SWE-Universe: Scale Real-World Verifiable Environments to Millions , author=. 2026 , eprint=

    2026

  37. [37]

    2024 , url=

    Carlos E Jimenez and John Yang and Alexander Wettig and Shunyu Yao and Kexin Pei and Ofir Press and Karthik R Narasimhan , booktitle=. 2024 , url=

    2024

  38. [38]

    Kimi k1.5: Scaling Reinforcement Learning with LLMs

    Kimi k1. 5: Scaling reinforcement learning with llms , author=. arXiv preprint arXiv:2501.12599 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  39. [39]

    Advances in Neural Information Processing Systems , volume=

    Swe-agent: Agent-computer interfaces enable automated software engineering , author=. Advances in Neural Information Processing Systems , volume=

  40. [40]

    A Report on the First Native Language Identification Shared Task

    Tetreault, Joel and Blanchard, Daniel and Cahill, Aoife. A Report on the First Native Language Identification Shared Task. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  41. [41]

    Applying Unsupervised Learning To Support Vector Space Model Based Speaking Assessment

    Chen, Lei. Applying Unsupervised Learning To Support Vector Space Model Based Speaking Assessment. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  42. [42]

    Role of Morpho-Syntactic Features in E stonian Proficiency Classification

    Vajjala, Sowmya and L \ o o, Kaidi. Role of Morpho-Syntactic Features in E stonian Proficiency Classification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  43. [43]

    Automated Content Scoring of Spoken Responses in an Assessment for Teachers of E nglish

    Zechner, Klaus and Wang, Xinhao. Automated Content Scoring of Spoken Responses in an Assessment for Teachers of E nglish. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  44. [44]

    Experimental Results on the Native Language Identification Shared Task

    Abu-Jbara, Amjad and Jha, Rahul and Morley, Eric and Radev, Dragomir. Experimental Results on the Native Language Identification Shared Task. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  45. [45]

    VTEX System Description for the NLI 2013 Shared Task

    Daudaravi c ius, Vidas. VTEX System Description for the NLI 2013 Shared Task. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  46. [46]

    Feature Space Selection and Combination for Native Language Identification

    Goutte, Cyril and L \'e ger, Serge and Carpuat, Marine. Feature Space Selection and Combination for Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  47. [47]

    Discriminating Non-Native E nglish with 350 Words

    Henderson, John and Zarrella, Guido and Pfeifer, Craig and Burger, John D. Discriminating Non-Native E nglish with 350 Words. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  48. [48]

    Maximizing Classification Accuracy in Native Language Identification

    Jarvis, Scott and Bestgen, Yves and Pepper, Steve. Maximizing Classification Accuracy in Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  49. [49]

    Recognizing E nglish Learners' Native Language from Their Writings

    Li, Baoli. Recognizing E nglish Learners' Native Language from Their Writings. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  50. [50]

    NLI Shared Task 2013: MQ Submission

    Malmasi, Shervin and Wong, Sze-Meng Jojo and Dras, Mark. NLI Shared Task 2013: MQ Submission. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  51. [51]

    NAIST at the NLI 2013 Shared Task

    Mizumoto, Tomoya and Hayashibe, Yuta and Sakaguchi, Keisuke and Komachi, Mamoru and Matsumoto, Yuji. NAIST at the NLI 2013 Shared Task. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  52. [52]

    Cognate and Misspelling Features for Natural Language Identification

    Nicolai, Garrett and Hauer, Bradley and Salameh, Mohammad and Yao, Lei and Kondrak, Grzegorz. Cognate and Misspelling Features for Natural Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  53. [53]

    Exploring Syntactic Representations for Native Language Identification

    Swanson, Ben. Exploring Syntactic Representations for Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  54. [54]

    Simple Yet Powerful Native Language Identification on TOEFL 11

    Wu, Ching-Yi and Lai, Po-Hsiang and Liu, Yang and Ng, Vincent. Simple Yet Powerful Native Language Identification on TOEFL 11. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  55. [55]

    Prompt-based Content Scoring for Automated Spoken Language Assessment

    Evanini, Keelan and Xie, Shasha and Zechner, Klaus. Prompt-based Content Scoring for Automated Spoken Language Assessment. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  56. [56]

    Automated Scoring of a Summary-Writing Task Designed to Measure Reading Comprehension

    Madnani, Nitin and Burstein, Jill and Sabatini, John and O ' Reilly, Tenaha. Automated Scoring of a Summary-Writing Task Designed to Measure Reading Comprehension. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  57. [57]

    Inter-annotator Agreement for Dependency Annotation of Learner Language

    Ragheb, Marwa and Dickinson, Markus. Inter-annotator Agreement for Dependency Annotation of Learner Language. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  58. [58]

    Native Language Identification with PPM

    Bobicev, Victoria. Native Language Identification with PPM. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  59. [59]

    Using Other Learner Corpora in the 2013 NLI Shared Task

    Brooke, Julian and Hirst, Graeme. Using Other Learner Corpora in the 2013 NLI Shared Task. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  60. [60]

    Combining Shallow and Linguistically Motivated Features in Native Language Identification

    Bykh, Serhiy and Vajjala, Sowmya and Krivanek, Julia and Meurers, Detmar. Combining Shallow and Linguistically Motivated Features in Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  61. [61]

    Linguistic Profiling based on General -- purpose Features and Native Language Identification

    Cimino, Andrea and Dell ' Orletta, Felice and Venturi, Giulia and Montemagni, Simonetta. Linguistic Profiling based on General -- purpose Features and Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  62. [62]

    Improving Native Language Identification with TF - IDF Weighting

    Gebre, Binyam Gebrekidan and Zampieri, Marcos and Wittenburg, Peter and Heskes, Tom. Improving Native Language Identification with TF - IDF Weighting. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  63. [63]

    Native Language Identification: a Simple n-gram Based Approach

    Gyawali, Binod and Ramirez, Gabriela and Solorio, Thamar. Native Language Identification: a Simple n-gram Based Approach. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  64. [64]

    Feature Engineering in the NLI Shared Task 2013: C harles U niversity Submission Report

    Hladk \'a , Barbora and Holub, Martin and Kr \'i z , Vincent. Feature Engineering in the NLI Shared Task 2013: C harles U niversity Submission Report. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  65. [65]

    Native Language Identification: A Key N-gram Category Approach

    Kyle, Kristopher and Crossley, Scott and Dai, Jianmin and McNamara, Danielle S. Native Language Identification: A Key N-gram Category Approach. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  66. [66]

    Using N-gram and Word Network Features for Native Language Identification

    Lahiri, Shibamouli and Mihalcea, Rada. Using N-gram and Word Network Features for Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  67. [67]

    LIMSI ' s participation to the 2013 shared task on Native Language Identification

    Lavergne, Thomas and Illouz, Gabriel and Max, Aur \'e lien and Nagata, Ryo. LIMSI ' s participation to the 2013 shared task on Native Language Identification. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  68. [68]

    Native Language Identification using large scale lexical features

    Lynum, Andr \'e. Native Language Identification using large scale lexical features. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  69. [69]

    The Story of the Characters, the DNA and the Native Language

    Popescu, Marius and Ionescu, Radu Tudor. The Story of the Characters, the DNA and the Native Language. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  70. [70]

    Identifying the L 1 of non-native writers: the CMU -Haifa system

    Tsvetkov, Yulia and Twitto, Naama and Schneider, Nathan and Ordan, Noam and Faruqui, Manaal and Chahuneau, Victor and Wintner, Shuly and Dyer, Chris. Identifying the L 1 of non-native writers: the CMU -Haifa system. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  71. [71]

    Evaluating Unsupervised Language Model Adaptation Methods for Speaking Assessment

    Xie, Shasha and Chen, Lei. Evaluating Unsupervised Language Model Adaptation Methods for Speaking Assessment. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  72. [72]

    Improving interpretation robustness in a tutorial dialogue system

    Dzikovska, Myroslava and Farrow, Elaine and Moore, Johanna. Improving interpretation robustness in a tutorial dialogue system. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  73. [73]

    Detecting Missing Hyphens in Learner Text

    Cahill, Aoife and Chodorow, Martin and Wolff, Susanne and Madnani, Nitin. Detecting Missing Hyphens in Learner Text. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  74. [74]

    Applying Machine Translation Metrics to Student-Written Translations

    Michaud, Lisa and McCoy, Patricia Ann. Applying Machine Translation Metrics to Student-Written Translations. Proceedings of the Eighth Workshop on Innovative Use of NLP for Building Educational Applications. 2013

    2013

  75. [75]

    Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

  76. [76]

    Recent adventures with emotion-reading technology

    Picard, Rosalind. Recent adventures with emotion-reading technology. Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

  77. [77]

    Bootstrapped Learning of Emotion Hashtags \# hashtags4you

    Qadir, Ashequl and Riloff, Ellen. Bootstrapped Learning of Emotion Hashtags \# hashtags4you. Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

  78. [78]

    Fine-Grained Emotion Recognition in Olympic Tweets Based on Human Computation

    Sintsova, Valentina and Musat, Claudiu and Pu, Pearl. Fine-Grained Emotion Recognition in Olympic Tweets Based on Human Computation. Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

  79. [79]

    S panish DAL : A S panish Dictionary of Affect in Language

    Dell ' Amerlina R \'i os, Mat \'i as and Gravano, Agust \'i n. S panish DAL : A S panish Dictionary of Affect in Language. Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

  80. [80]

    The perfect solution for detecting sarcasm in tweets \# not

    Liebrecht, Christine and Kunneman, Florian and van den Bosch, Antal. The perfect solution for detecting sarcasm in tweets \# not. Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2013

    2013

Showing first 80 references.