December 2025 - Sanghani-faculty

LDC data survey
by Brian Mayer 15 Dec '25

15 Dec '25

Please take this survey regarding the continuation of the LDC data subscription. It is 3 questions and should take less than a minute to complete. https://virginiatech.qualtrics.com/jfe/form/SV_2gcrxfoThNqNEOy If you don't complete it, we will assume you have no interest in maintaining the data source (which is fine). Thanks, Brian

1 0

FW: December 2025 Newsletter - LDC
by Brian Mayer 15 Dec '25

15 Dec '25

FYI From: Ldc-customers1 <ldc-customers1-bounces(a)ldc.upenn.edu> On Behalf Of Penn LDC Sent: Monday, December 15, 2025 11:16 AM To: Penn LDC <ldc(a)ldc.upenn.edu> Subject: December 2025 Newsletter - LDC In this newsletter: LDC 2026 membership discounts now available LDC's 1000th corpus Approaching deadline for Spring 2026 data scholarship applications LDC closed for Winter Break December 25 - January 2 New publications: <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcatalog.l dc.upenn.edu%2FLDC2025S11&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c243563 6a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124374649 541%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAi OiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=Xti89uf%2F5G CpQyTti3tyVdrHCu2ItbXpE3Oo%2B9ntBIE%3D&reserved=0> 2021 NIST Speaker Recognition Evaluation Development and Test Set <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcatalog.l dc.upenn.edu%2FLDC2025T17&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c243563 6a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124374681 734%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAi OiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=0ObmByO3w91O %2FHew4bseL9ZvuuYOTW5yt7fITYqtvYo%3D&reserved=0> LORELEI Sinhala Incident Language Pack _____ LDC 2026 membership discounts now available Now through March 2, 2026, any organization that joins the Consortium or renews their membership will receive a 10% discount off the 2026 membership fee. Membership remains the most economical way to access current and past LDC releases. Consult <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.ldc.u penn.edu%2Fmembers%2Fjoin-ldc&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c24 35636a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C63901412437 4701766%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIs IlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=wlqN54Qg nOLbGMjRtkheVdNyW0gfHWYaXdMxpkZrsJM%3D&reserved=0> Join LDC for details on membership options and benefits. LDC's 1,000th corpus LDC is delighted to announce the release of the 1,000th corpus into the Catalog! This milestone represents the commitment we made over thirty years ago to provide large quantities of diverse data, robust research program support, and exceptional member services. We are grateful for the continued support and collaboration of our members, friends, and the community. Approaching deadline for Spring 2026 data scholarship applications Attention students: don't miss out on the chance to receive no-cost access to LDC data for your research. Applications for Spring 2026 data scholarships are due January 15, 2026. For more information on requirements and program rules, see <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.ldc.u penn.edu%2Flanguage-resources%2Fdata%2Fdata-scholarships&data=05%7C02%7Cmaye r2%40vt.edu%7C16b5e9dbe9c2435636a008de3bf5de21%7C6095688410ad40fa863d4f32c1e 3a37a%7C0%7C0%7C639014124374727193%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiO nRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D% 7C0%7C%7C%7C&sdata=Z5EMqwSJG0GsxLD%2F%2B5Uy5fOfU3436N%2FMJvEjw20EFLs%3D&rese rved=0> LDC Data Scholarships. LDC closed for Winter Break December 25-January 2 LDC will be closed from Thursday, December 25, 2025, through Friday, January 2, 2026, in accordance with the University of Pennsylvania Winter Break Policy. Our offices will reopen on Monday, January 5, 2026. Requests received by the Membership Office during Winter Break will be processed when the office reopens. _____ New publications: <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcatalog.l dc.upenn.edu%2FLDC2025S11&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c243563 6a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124374981 015%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAi OiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=6Ev414eFN22p 5AF8Quw5yh93JKxMGLh%2BwAjJVWDKcFk%3D&reserved=0> 2021 NIST Speaker Recognition Evaluation Test Set was developed by LDC and NIST (National Institute of Standards and Technology). It contains approximately 447 hours of Cantonese, Mandarin, and English conversational telephone speech, audio from video, and selfie image data for development and test, along with answer keys, enrollment, trial files, and documentation from the NIST-sponsored <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.nist. gov%2Fitl%2Fiad%2Fmig%2Fnist-2021-speaker-recognition-evaluation-sre21&data= 05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c2435636a008de3bf5de21%7C6095688410ad4 0fa863d4f32c1e3a37a%7C0%7C0%7C639014124375003480%7CUnknown%7CTWFpbGZsb3d8eyJ FbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIld UIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=QLpSK7IxMlkVeb75zP1KT%2F9S3nx73nBAIFZULaTtL VM%3D&reserved=0> 2021 Speaker Recognition Evaluation (SRE). The SRE task is speaker detection, that is, to determine whether a specified target speaker was speaking during a segment of speech. SRE21 focused on telephone speech and audio from video and included close-up images of participants. The evaluation also featured cross-lingual trials, that is, enrollment and test segments spoken in different languages. The data was drawn from the WeCanTalk corpus collected by LDC in which speakers called friends or relatives who agreed to record their telephone conversations lasting between 8-10 minutes. Subjects contributed multiple conversational telephone speech recordings and audio recordings in which they were talking, plus a single selfie image. Recordings were manually audited to verify speaker, language, and quality. 2025 members can access this corpus through their LDC accounts. Non-members may license this data for $750. * <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcatalog.l dc.upenn.edu%2FLDC2025T17&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c243563 6a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124375028 780%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAi OiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=dawOMGXpbBE2 LWMCtcay9dwxaLntBm%2Fi%2FjoMzFVLR9U%3D&reserved=0> LORELEI Sinhala Incident Language Pack was developed by LDC and is comprised of 8.1 million words of Sinhala monolingual text, 700,00 words of English monolingual text, 6.4 million words of parallel Sinhala- English text, and 50,000 words annotated for entity discovery and linking and situation frames. It constitutes all of the text data, annotations, supplemental resources, and related software tools for the Sinhala language used in the <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.nist. gov%2Fitl%2Fiad%2Fmig%2Florehlt-evaluations&data=05%7C02%7Cmayer2%40vt.edu%7 C16b5e9dbe9c2435636a008de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0 %7C639014124375058815%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOi IwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C& sdata=s%2FaWXYuvajEkALtwY7%2FyA1iZqizFoJlH6vs56mTn1MI%3D&reserved=0> DARPA LORELEI / LoReHLT 2018 Evaluation. The LORELEI (Low Resource Languages for Emergent Incidents) program was concerned with building human language technology for low resource languages in the context of emergent situations. In the evaluation scenario, an unforeseen event triggered a need for humanitarian and logistical support in a region where the incident language had received little or no attention in NLP research. Evaluation participants provided NLP solutions, including information extraction and machine translation, with limited resources and limited development time. Data was collected from news, social network, weblog, newsgroup, discussion forum, and reference material. Entity discovery and linking annotation identified entities to be detected by systems for scoring purposes. Situation frame analysis was designed to extract basic information about needs and relevant issues for planning a disaster response effort. 2025 members can access this corpus through their LDC accounts. Non-members may license this data for $250. To unsubscribe from this newsletter, log in to your <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcatalog.l dc.upenn.edu%2Flogin&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c2435636a008 de3bf5de21%7C6095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124375086127%7 CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXa W4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=wZWb7cu332FyfUurc g%2Bemt4RMZbbH8yArCbHDhtsEuU%3D&reserved=0> LDC account and uncheck the box next to "Receive Newsletter" under Account Options or contact LDC for assistance. Membership Coordinator <https://nam04.safelinks.protection.outlook.com/?url=http%3A%2F%2Fldc.upenn. edu%2F&data=05%7C02%7Cmayer2%40vt.edu%7C16b5e9dbe9c2435636a008de3bf5de21%7C6 095688410ad40fa863d4f32c1e3a37a%7C0%7C0%7C639014124375113735%7CUnknown%7CTWF pbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjo iTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=GFYcb7Tt0Raxu%2Bt9OvpxWUYFCrDG% 2BYEwlMrqJJbiRcw%3D&reserved=0> Linguistic Data Consortium University of Pennsylvania T: +1-215-573-1275 E: <mailto:ldc@ldc.upenn.edu> ldc(a)ldc.upenn.edu M: 3600 Market St. Suite 810 Philadelphia, PA 19104

1 0

Sanghani Center Faculty Meeting - Friday, Dec. 12 at 12 - 1:30 pm
by Shalaby, Wanawsha 12 Dec '25

12 Dec '25

Hello, Our next Sanghani Center Faculty meeting is scheduled for Friday, December 12 at 12:00 - 1:30 pm. As usual, if you have any recent news or updates since our last faculty meeting, please send them my way so that I may add them to our meeting slides. Updates include awards, proposals, grants, conference presentations, events, etc. Please send all updates by Wednesday, December 10. Those of you in Blacksburg will be meeting in D&DS 366 and those of us in the D.C. area will be meeting in IAC 5740. I have provided a Zoom link below for those who cannot attend in person. Please note that it's the same link for all of our Sanghani Faculty meetings this semester. Join Zoom Meeting Https://virginiatech.zoom.us/j/82689882396<https://virginiatech.zoom.us/j/82689882396> Meeting ID: 826 8988 2396 --- One tap mobile +13017158592,,82689882396# US (Washington DC) +13052241968,,82689882396# US Thanks, Wanawsha Wanawsha Shalaby Manager of Operations Sanghani Center for AI and Data Analytics - Virginia Tech 3625 Potomac Ave., Alexandria, VA 22305 E-Mail: wanah92(a)vt.edu<mailto:wanah92@vt.edu>

1 2

Final PhD defense of Satvik Chekuri at 11am, Tuesday, December 16 in 2030 Torgersen Hall and in Zoom
by Fox, Edward 02 Dec '25

02 Dec '25

Hi! I am pleased to invite you to Satvik Chekuri's final PhD defense at 11am on Tuesday, December 16, in 2030 Torgersen Hall. The Zoom link is: https://virginiatech.zoom.us/my/ed.fox Please note that topics covered include AI, NLP, LLMs, Question Answering, Knowledge Graphs, Information Retrieval (aka search engines), Digital Libraries, User Studies, ETDs, Topic Modeling, Question Generation, Object Detection, Ontologies, Automatic Classification, and Scholarly Publishing. Title: Scholarly Information System for Long Documents and their Elements: Structured Representation and Exploration using Knowledge Graphs and LLM Abstract: Electronic Theses and Dissertations (ETDs) contain some of the deepest and most detailed accounts of academic work, yet the systems used to access them remain limited. Traditional information retrieval emphasizes document-level search and provides only coarse entry points into long-form scholarly writing. Large language models (LLMs), although capable of fluent explanation, lack grounding in the structure of ETDs and cannot reliably reason over their internal elements. This dissertation addresses these challenges by developing a scholarly information system that integrates fine-grained retrieval, structured knowledge graph representations, and grounded question answering to support richer access to ETD knowledge. Instead of treating an ETD as a single monolithic text, the proposed ETD-IR system (Information Retrieval System for ETDs) extracts and indexes elements like figure captions, topic labels, and abstract sentences, allowing users to retrieve precise evidence. We introduce a formal ontology to capture semantic relationships in ETDs, leading to the creation of the first ETD-specific Knowledge Graph (ETD-KG). To enhance scholarly discovery and question-answering, this dissertation proposes ETD-OHKGR, an ontology-guided hybrid KG retrieval framework. ETD-IR retrieves relevant documents, ETD-IR + LLM synthesize results with a LLM, and ETD-OHKGR adds graph-based reasoning to generate answers based on textual and structural evidence. This research contributes an ETD Abstract Sentence Classification dataset (ETD-ASC), consisting of human-labeled and LLM-assisted silver annotations. Fine-tuned models trained on this dataset highlight that ETD abstracts exhibit rhetorical patterns distinct from those found in scientific papers, reinforcing the need for ETD-specific classification models. These class labels are used to improve knowledge extraction and structural coherence within ETD-KG. A comprehensive user study involving 47 participants across four engineering departments has compared the three methods on informativeness, accuracy, readability, and clarity metrics. The ETD-OHKGR method has achieved 40% higher overall user satisfaction than ETD-IR (p < 0.001). It also delivered major improvements in presentation quality, with a 53% gain in readability and a 48% gain in clarity. ETD-OHKGR significantly outperformed both baselines on the two metrics that users consistently valued the most—informativeness and clarity. These improvements held for 87% of participants, demonstrating robust performance. To analyze the relationship between the responses from different methods, paragraph-level natural language inference (NLI) was used, treating ETD-IR+LLM as the premise and ETD-OHKGR as the hypothesis. ETD-OHKGR’s answers mainly fell into the neutral class (50–100%), suggesting it adds complementary content without contradictions, aligning with its goal of generating structured, elaborative answers based on ETD-KG. Semantic similarity analysis showed moderate divergence (mean cosine similarity = 0.420), emphasizing the role of structured retrieval with diverse writing styles. Strong correlations (r > 0.70) among metrics confirmed that clarity and readability predict user satisfaction. Taken together, these findings demonstrate that fine-grained retrieval, ontological structure, and knowledge graph reasoning enhance access to scholarly content in ETDs. The proposed system offers capabilities that neither traditional IR nor LLMs alone can achieve, making long-form academic documents more discoverable and useful for researchers. All are welcome! PS Special thanks go to Satvik's doctoral committee and to collaborators in University Libraries. -- Edward A. Fox, Ph.D., Emeritus Professor of Computer Science Virginia Tech | CS (0106) 3160 Torgersen Hall (CS dept. office) Blacksburg, VA 24061 (540) 231-5113 | fox(a)vt.edu | http://fox.cs.vt.edu

1 0