Fine-Tuned Llama for Multilingual Text-to-Text Coreference Resolution Article Swipe

View

Association for Computational Linguistics 2025 ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48448/y1p8-qm18

This paper describes our approach to the CRAC 2025 Shared Task on Multilingual Coreference Resolution. We compete in the LLM track, where the systems are limited to generative text-to-text approaches. Our system is based on Llama 3.1-8B, fine-tuned to tag the document with coreference annotations. We have made one significant modification to the text format provided by the organizers: The model relies on the syntactic head for mention span representation. Additionally, we use joint pre-training, and we train the model to generate empty nodes. We provide an in-depth analysis of the performance of our models, which reveals several implementation problems. Although our system ended up in last place, we achieved the best performance on 10 datasets out of 22 within the track. By fixing the discovered problems in the post-evaluation phase, we improved our results substantially, outperforming all the systems in the LLM track and even some unconstrained track systems.

Related Topics

Computer Science

Artificial Intelligence

Generative Grammar

Training, Validation, And Test Data Sets

Machine Learning

Concepts

Coreference Computer science Artificial intelligence Task (project management) Natural language processing Generative grammar Resolution (logic) Joint (building) Training set Track (disk drive) Head (geology) Generative model Machine learning Task analysis Language model Structured prediction Span (engineering) Discriminative model

Metadata

Type: other
Landing Page: https://doi.org/10.48448/y1p8-qm18
OA Status: green
OpenAlex ID: https://openalex.org/W7106847480

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W7106847480

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48448/y1p8-qm18

Digital Object Identifier
Title: Fine-Tuned Llama for Multilingual Text-to-Text Coreference Resolution

Work title
Type: other

OpenAlex work type
Publication year: 2025

Year of publication
Publication date: 2025-10-25

Full publication date if available
Authors: Association for Computational Linguistics 2025

List of authors in order
Landing page: https://doi.org/10.48448/y1p8-qm18

Publisher landing page
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://doi.org/10.48448/y1p8-qm18

Direct OA link when available
Concepts: Coreference, Computer science, Artificial intelligence, Task (project management), Natural language processing, Generative grammar, Resolution (logic), Joint (building), Training set, Track (disk drive), Head (geology), Generative model, Machine learning, Task analysis, Language model, Structured prediction, Span (engineering), Discriminative model

Top concepts (fields/topics) attached by OpenAlex
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W7106847480
doi	https://doi.org/10.48448/y1p8-qm18
ids.doi	https://doi.org/10.48448/y1p8-qm18
ids.openalex	https://openalex.org/W7106847480
fwci
type	other
title	Fine-Tuned Llama for Multilingual Text-to-Text Coreference Resolution
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C28076734
concepts[0].level	3
concepts[0].score	0.9715749621391296
concepts[0].wikidata	https://www.wikidata.org/wiki/Q63087
concepts[0].display_name	Coreference
concepts[1].id	https://openalex.org/C41008148
concepts[1].level	0
concepts[1].score	0.8551696538925171
concepts[1].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[1].display_name	Computer science
concepts[2].id	https://openalex.org/C154945302
concepts[2].level	1
concepts[2].score	0.6902992129325867
concepts[2].wikidata	https://www.wikidata.org/wiki/Q11660
concepts[2].display_name	Artificial intelligence
concepts[3].id	https://openalex.org/C2780451532
concepts[3].level	2
concepts[3].score	0.6775925755500793
concepts[3].wikidata	https://www.wikidata.org/wiki/Q759676
concepts[3].display_name	Task (project management)
concepts[4].id	https://openalex.org/C204321447
concepts[4].level	1
concepts[4].score	0.6507432460784912
concepts[4].wikidata	https://www.wikidata.org/wiki/Q30642
concepts[4].display_name	Natural language processing
concepts[5].id	https://openalex.org/C39890363
concepts[5].level	2
concepts[5].score	0.5594608783721924
concepts[5].wikidata	https://www.wikidata.org/wiki/Q36108
concepts[5].display_name	Generative grammar
concepts[6].id	https://openalex.org/C138268822
concepts[6].level	2
concepts[6].score	0.5368044972419739
concepts[6].wikidata	https://www.wikidata.org/wiki/Q1051925
concepts[6].display_name	Resolution (logic)
concepts[7].id	https://openalex.org/C18555067
concepts[7].level	2
concepts[7].score	0.5174639821052551
concepts[7].wikidata	https://www.wikidata.org/wiki/Q8375051
concepts[7].display_name	Joint (building)
concepts[8].id	https://openalex.org/C51632099
concepts[8].level	2
concepts[8].score	0.46453720331192017
concepts[8].wikidata	https://www.wikidata.org/wiki/Q3985153
concepts[8].display_name	Training set
concepts[9].id	https://openalex.org/C89992363
concepts[9].level	2
concepts[9].score	0.462985634803772
concepts[9].wikidata	https://www.wikidata.org/wiki/Q5961558
concepts[9].display_name	Track (disk drive)
concepts[10].id	https://openalex.org/C2780312720
concepts[10].level	2
concepts[10].score	0.42834988236427307
concepts[10].wikidata	https://www.wikidata.org/wiki/Q5689100
concepts[10].display_name	Head (geology)
concepts[11].id	https://openalex.org/C167966045
concepts[11].level	3
concepts[11].score	0.3551815152168274
concepts[11].wikidata	https://www.wikidata.org/wiki/Q5532625
concepts[11].display_name	Generative model
concepts[12].id	https://openalex.org/C119857082
concepts[12].level	1
concepts[12].score	0.2954264283180237
concepts[12].wikidata	https://www.wikidata.org/wiki/Q2539
concepts[12].display_name	Machine learning
concepts[13].id	https://openalex.org/C175154964
concepts[13].level	3
concepts[13].score	0.29412785172462463
concepts[13].wikidata	https://www.wikidata.org/wiki/Q380077
concepts[13].display_name	Task analysis
concepts[14].id	https://openalex.org/C137293760
concepts[14].level	2
concepts[14].score	0.2702753245830536
concepts[14].wikidata	https://www.wikidata.org/wiki/Q3621696
concepts[14].display_name	Language model
concepts[15].id	https://openalex.org/C22367795
concepts[15].level	2
concepts[15].score	0.2597981095314026
concepts[15].wikidata	https://www.wikidata.org/wiki/Q7625208
concepts[15].display_name	Structured prediction
concepts[16].id	https://openalex.org/C2778753569
concepts[16].level	2
concepts[16].score	0.2525843679904938
concepts[16].wikidata	https://www.wikidata.org/wiki/Q1960395
concepts[16].display_name	Span (engineering)
concepts[17].id	https://openalex.org/C97931131
concepts[17].level	2
concepts[17].score	0.25112003087997437
concepts[17].wikidata	https://www.wikidata.org/wiki/Q5282087
concepts[17].display_name	Discriminative model
keywords[0].id	https://openalex.org/keywords/coreference
keywords[0].score	0.9715749621391296
keywords[0].display_name	Coreference
keywords[1].id	https://openalex.org/keywords/task
keywords[1].score	0.6775925755500793
keywords[1].display_name	Task (project management)
keywords[2].id	https://openalex.org/keywords/generative-grammar
keywords[2].score	0.5594608783721924
keywords[2].display_name	Generative grammar
keywords[3].id	https://openalex.org/keywords/resolution
keywords[3].score	0.5368044972419739
keywords[3].display_name	Resolution (logic)
keywords[4].id	https://openalex.org/keywords/joint
keywords[4].score	0.5174639821052551
keywords[4].display_name	Joint (building)
keywords[5].id	https://openalex.org/keywords/training-set
keywords[5].score	0.46453720331192017
keywords[5].display_name	Training set
keywords[6].id	https://openalex.org/keywords/track
keywords[6].score	0.462985634803772
keywords[6].display_name	Track (disk drive)
language
locations[0].id	doi:10.48448/y1p8-qm18
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S7407053148
locations[0].source.type	repository
locations[0].source.is_oa	False
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	Underline Science Inc.
locations[0].source.host_organization
locations[0].source.host_organization_name
locations[0].license
locations[0].pdf_url
locations[0].version
locations[0].raw_type	article
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published
locations[0].raw_source_name
locations[0].landing_page_url	https://doi.org/10.48448/y1p8-qm18
indexed_in	datacite
authorships[0].author.id
authorships[0].author.orcid
authorships[0].author.display_name	Association for Computational Linguistics 2025
authorships[0].author_position	first
authorships[0].raw_author_name	Association for Computational Linguistics 2025
authorships[0].is_corresponding	True
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://doi.org/10.48448/y1p8-qm18
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-11-28T00:00:00
display_name	Fine-Tuned Llama for Multilingual Text-to-Text Coreference Resolution
has_fulltext	False
is_retracted	False
updated_date	2025-11-28T02:12:24.556248
primary_topic
cited_by_count	0
locations_count	1
best_oa_location.id	doi:10.48448/y1p8-qm18
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S7407053148
best_oa_location.source.type	repository
best_oa_location.source.is_oa	False
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	Underline Science Inc.
best_oa_location.source.host_organization
best_oa_location.source.host_organization_name
best_oa_location.license
best_oa_location.pdf_url
best_oa_location.version
best_oa_location.raw_type	article
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	https://doi.org/10.48448/y1p8-qm18
primary_location.id	doi:10.48448/y1p8-qm18
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S7407053148
primary_location.source.type	repository
primary_location.source.is_oa	False
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	Underline Science Inc.
primary_location.source.host_organization
primary_location.source.host_organization_name
primary_location.license
primary_location.pdf_url
primary_location.version
primary_location.raw_type	article
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	https://doi.org/10.48448/y1p8-qm18
publication_date	2025-10-25
publication_year	2025
referenced_works_count	0
abstract_inverted_index.10	114
abstract_inverted_index.22	118
abstract_inverted_index.By	122
abstract_inverted_index.We	15, 45, 84
abstract_inverted_index.an	86
abstract_inverted_index.by	56
abstract_inverted_index.in	17, 105, 127, 140
abstract_inverted_index.is	32
abstract_inverted_index.of	89, 92, 117
abstract_inverted_index.on	11, 34, 62, 113
abstract_inverted_index.to	5, 26, 38, 51, 80
abstract_inverted_index.up	104
abstract_inverted_index.we	71, 76, 108, 131
abstract_inverted_index.LLM	19, 142
abstract_inverted_index.Our	30
abstract_inverted_index.The	59
abstract_inverted_index.all	137
abstract_inverted_index.and	75, 144
abstract_inverted_index.are	24
abstract_inverted_index.for	66
abstract_inverted_index.one	48
abstract_inverted_index.our	3, 93, 101, 133
abstract_inverted_index.out	116
abstract_inverted_index.tag	39
abstract_inverted_index.the	6, 18, 22, 40, 52, 57, 63, 78, 90, 110, 120, 124, 128, 138, 141
abstract_inverted_index.use	72
abstract_inverted_index.2025	8
abstract_inverted_index.CRAC	7
abstract_inverted_index.Task	10
abstract_inverted_index.This	0
abstract_inverted_index.best	111
abstract_inverted_index.even	145
abstract_inverted_index.have	46
abstract_inverted_index.head	65
abstract_inverted_index.last	106
abstract_inverted_index.made	47
abstract_inverted_index.some	146
abstract_inverted_index.span	68
abstract_inverted_index.text	53
abstract_inverted_index.with	42
abstract_inverted_index.Llama	35
abstract_inverted_index.based	33
abstract_inverted_index.empty	82
abstract_inverted_index.ended	103
abstract_inverted_index.joint	73
abstract_inverted_index.model	60, 79
abstract_inverted_index.paper	1
abstract_inverted_index.track	143, 148
abstract_inverted_index.train	77
abstract_inverted_index.where	21
abstract_inverted_index.which	95
abstract_inverted_index.Shared	9
abstract_inverted_index.fixing	123
abstract_inverted_index.format	54
abstract_inverted_index.nodes.	83
abstract_inverted_index.phase,	130
abstract_inverted_index.place,	107
abstract_inverted_index.relies	61
abstract_inverted_index.system	31, 102
abstract_inverted_index.track,	20
abstract_inverted_index.track.	121
abstract_inverted_index.within	119
abstract_inverted_index.3.1-8B,	36
abstract_inverted_index.compete	16
abstract_inverted_index.limited	25
abstract_inverted_index.mention	67
abstract_inverted_index.models,	94
abstract_inverted_index.provide	85
abstract_inverted_index.results	134
abstract_inverted_index.reveals	96
abstract_inverted_index.several	97
abstract_inverted_index.systems	23, 139
abstract_inverted_index.Although	100
abstract_inverted_index.achieved	109
abstract_inverted_index.analysis	88
abstract_inverted_index.approach	4
abstract_inverted_index.datasets	115
abstract_inverted_index.document	41
abstract_inverted_index.generate	81
abstract_inverted_index.improved	132
abstract_inverted_index.in-depth	87
abstract_inverted_index.problems	126
abstract_inverted_index.provided	55
abstract_inverted_index.systems.	149
abstract_inverted_index.describes	2
abstract_inverted_index.problems.	99
abstract_inverted_index.syntactic	64
abstract_inverted_index.discovered	125
abstract_inverted_index.fine-tuned	37
abstract_inverted_index.generative	27
abstract_inverted_index.Coreference	13
abstract_inverted_index.Resolution.	14
abstract_inverted_index.approaches.	29
abstract_inverted_index.coreference	43
abstract_inverted_index.organizers:	58
abstract_inverted_index.performance	91, 112
abstract_inverted_index.significant	49
abstract_inverted_index.Multilingual	12
abstract_inverted_index.annotations.	44
abstract_inverted_index.modification	50
abstract_inverted_index.text-to-text	28
abstract_inverted_index.Additionally,	70
abstract_inverted_index.outperforming	136
abstract_inverted_index.pre-training,	74
abstract_inverted_index.unconstrained	147
abstract_inverted_index.implementation	98
abstract_inverted_index.substantially,	135
abstract_inverted_index.post-evaluation	129
abstract_inverted_index.representation.	69
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	1
citation_normalized_percentile