The Thirty-Third Text REtrieval Conference
(TREC 2024)

NeuCLIR (Cross language and multilingual search) Cross-language technical docs retrieval task Appendix

RuntagOrgIs this run manual or automatic?What form of the documents did you use?What topic fields did you use?What form of the queries did you use?Please provide a short description of this run, including info about anything checked "Other" above.Please give this run a priority for inclusion in manual assessments.
tech-coordinators-1MNCS-patapscoBM25htnoRM3desc (ir_measures) (paper)coordinators
manual
['Original non-English text']
['description']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization
20
tech-coordinators-1MNCS-patapscoBM25htnoRM3td (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization
6
tech-coordinators-1MNCS-patapscoBM25htnoRM3title (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization
14
tech-coordinators-1MNCS-patapscoBM25htRM3desc (ir_measures) (paper)coordinators
manual
['Original non-English text']
['description']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
17
tech-coordinators-1MNCS-patapscoBM25htRM3td (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
1 (top)
tech-coordinators-1MNCS-patapscoBM25htRM3title (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title']
['Track-provided human translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
10
tech-coordinators-MNES-fast_psq_t (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title']
['English']
Probabilistic structured queries using fast-psq implementation with recommended parameters for creating statistical translation table and index
11
tech-coordinators-MNES-fast_psq_td (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['English']
Probabilistic structured queries using fast-psq implementation with recommended parameters for creating statistical translation table and index
3
tech-coordinators-MNES-patapscoBM25qtnoRM3desc (ir_measures) (paper)coordinators
manual
['Original non-English text']
['description']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization
22
tech-coordinators-MNES-patapscoBM25qtnoRM3td (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization
8
tech-coordinators-MNES-patapscoBM25qtnoRM3title (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization
16
tech-ISI_SEARCHER-ANE_run1 (ir_measures) ISI
automatic
['Original non-English text']
['title', 'description', 'narrative']
['English']
ISI's TREC 2024 SEARCHER submission is a two-stage model consisting of an initial retrieval stage and a reranking stage. The first stage is based on a neural cross-lingual model utilizing a sparse index (see https://aclanthology.org/2020.clssts-1.4.pdf); the second stage re-ranks the top 1000 documents using a cross encoder based on a fine-tuned Mistral model.
1 (top)
tech-coordinators-MNES-patapscoBM25qtRM3desc (ir_measures) (paper)coordinators
manual
['Original non-English text']
['description']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
19
tech-coordinators-MNES-patapscoBM25qtRM3td (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
6
tech-coordinators-MNES-patapscoBM25qtRM3title (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title']
['Track-provided Google translation']
Patapsco BM25 run using spacy tokenization and RM3 with 10 terms. Original query weight is 50%.
13
tech-coordinators-MTES-patapscoBM25dtnoRM3desc (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['description']
['English']
Patapsco BM25 run using spacy tokenization and stemming
21
tech-coordinators-MTES-patapscoBM25dtnoRM3td (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['title', 'description']
['English']
Patapsco BM25 run using spacy tokenization and stemming
7
tech-coordinators-MTES-patapscoBM25dtnoRM3title (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['title']
['English']
Patapsco BM25 run using spacy tokenization and stemming
15
tech-coordinators-MTES-patapscoBM25dtRM3desc (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['description']
['English']
Patapsco BM25 run using spacy tokenization and stemming and RM3 with 10 terms. Original query weight is 50%.
18
tech-coordinators-MTES-patapscoBM25dtRM3td (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['title', 'description']
['English']
Patapsco BM25 run using spacy tokenization and stemming and RM3 with 10 terms. Original query weight is 50%.
4
tech-coordinators-MTES-patapscoBM25dtRM3title (ir_measures) (paper)coordinators
manual
['Track-provided translations']
['title']
['English']
Patapsco BM25 run using spacy tokenization and stemming and RM3 with 10 terms. Original query weight is 50%.
12
tech-coordinators-MNCD-plaid_distill_mono_ht (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['Track-provided human translation']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries and documents translated into Chinese using distillation. Default ColBERT parameters with 1 bit for residual index
2
tech_coordinators-topic_dev (ir_measures) (paper)coordinators
manual
['Original non-English text', 'Track-provided translations']
['title', 'description', 'narrative']
['Other']
These were the documents identified as somewhat or very valuable during topic development. Assessors use any query they wanted to identify the documents.
23
tech-hltcoe-MNED-plaid_distill_engeng_zs2engzho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries and documents in English using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index
8
tech-hltcoe-MNED-plaid_distill_engeng_zs2zhozho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries and documents in English using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index
13
tech-hltcoe-MNED-plaid_distill_engzho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries in English and documents translated into Chinese using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index
7
tech-hltcoe-MNED-plaid_distill_engzho.mt5rerank (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries in English and documents translated into Chinese using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index. Reranked using mt5
3
tech-hltcoe-MNED-plaid_distill_engzho.mt5rerank.gpt4rerank (ir_measures) (paper)coordinators
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries in English and documents translated into Chinese using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index. Reranked using mt5 and top 30 reranked using gpt4.
8
tech-hltcoe-MNED-plaid_distill_zhozho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries and documents translated into Chinese using distillation. Default ColBERT parameters with 1 bit for residual index
11
tech-hltcoe-MNED-plaid_syn_distill_engzho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on sythetix queries created by mixtral in English and documents from the collection using distillation of English, Chinese scores. Default ColBERT parameters with 1 bit for residual index
6
tech-hltcoe-MNTEH-kitchen_rankfuse (ir_measures) hltcoe
manual
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
Used rank-based fusion over the following runs: tech-hltcoe-MNED-plaid_distill_engzho, tech-hltcoe-MTED-plaid_distill_engeng, tech-coordinators-MNES-fast_psq_td, tech-coordinators-MNES-patapscoBM25qtRM3td, tech-coordinators-MTES-patapscoBM25dtRM3td, and tech-ISI_SEARCHER-ANE_run1
9
tech-hltcoe-MNTEH-kitchen_rankfuse.mt5rerank (ir_measures) hltcoe
manual
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
Reranked tech-hltcoe-MNTEH-kitchen_rankfuse run with mt5
4
tech-hltcoe-MNTEH-kitchen_rankfuse.mt5rerank.gpt4rerank (ir_measures) hltcoe
manual
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
Reranked tech-hltcoe-MNTEH-kitchen_rankfuse.mt5rerank run with gpt4
2
tech-hltcoe-MTED-plaid_distill_engeng (ir_measures) hltcoe
manual
['Track-provided translations']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on MSMARCO queries and documents in English using distillation of English,English scores. Default ColBERT parameters with 1 bit for residual index
10
tech-hltcoe-MNED-plaid_eqsynms_distill_engzho (ir_measures) hltcoe
manual
['Original non-English text']
['title', 'description']
['English']
PLAID-X implementation of ColBERT with model fine-tuned on 50% sythetix queries created by mixtral in English and documents from the collection using distillation of English, Chinese scores and 50% MSMARCO queries in English and MSMARCO documents translated in Chinese. Default ColBERT parameters with 1 bit for residual index
1 (top)
fs1_xlmr_monot5_listgalore (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR monoT5 - top-1K reranking with monoT5-3B-10k listgalore - RRF of top-100 Listwise RankGPT-4o + RankLLaMA70B - window (20, 10)
1 (top)
fs1_monot5_listgalore (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT) monoT5 - top-1K reranking with monoT5-3B-10k listgalore - RRF of top-100 Listwise RankGPT-4o + RankLLaMA70B - window (20, 10)
2
fs1_xlmr_monot5_rgpt-4o (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR monoT5 - top-1K reranking with monoT5-3B-10k rgpt-4o - top-100 Listwise RankGPT-4o - window (20, 10)
3
fs1_monot5_rgpt-4o (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT). monoT5 - top-1K reranking with monoT5-3B-10k rgpt-4o - top-100 Listwise RankGPT-4o - window (20, 10)
4
fs1_xlmr_monot5_rl3.1_70b (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR monoT5 - top-1K reranking with monoT5-3B-10k rl3.1 - top-100 Listwise RankLLaMA70B - window (20, 10)
6
fs1_monot5_rl3.1_70b (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT). monoT5 - top-1K reranking with monoT5-3B-10k rl3.1 - top-100 Listwise RankLLaMA70B - window (20, 10)
7
fs1_xlmr_monot5_rz (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR monoT5 - top-1K reranking with monoT5-3B-10k rz - top-100 Listwise RankZephyr - window (20, 10)
8
fs1_monot5_rz (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT). monoT5 - top-1K reranking with monoT5-3B-10k rz - top-100 Listwise RankZephyr - window (20, 10)
9
fs1_xlmr_monot5 (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR monoT5 - top-1K reranking with monoT5-3B-10k
10
fs1_monot5 (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT). monoT5 - top-1K reranking with monoT5-3B-10k
11
fs1_xlmr (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1_xlmr - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT), xlmr-based ColBERT-X trained with translate-distill for English-Chinese CLIR
12
fs1 (ir_measures) h2oloo
automatic
['Original non-English text', 'Track-provided translations']
['title', 'description']
['English', 'Track-provided Google translation']
fs1 - RRF of BM25+Rocchio-QT, BM25+Rocchio-DT, GTE QWEN2 (QT)
13
gte-qwen-desc+title (ir_measures) h2oloo
automatic
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
GTE QWEN2
14
bm25-rocchio-qt-desc+title (ir_measures) h2oloo
automatic
['Original non-English text']
['title', 'description']
['Track-provided Google translation']
Pyserini QT BM25 + Rocchio - Desc + Title
15
bm25-rocchio-dt-desc+title (ir_measures) h2oloo
automatic
['Track-provided translations']
['title', 'description']
['English']
Pyserini DT BM25 + Rocchio - Desc + Title
16