Swedish Historical Text NER Model
Budget: €30 – €250 EUR
Assignment: Named Entity Recognition
In this assignment, your goal is to build a Named Entity Recognition (NER) model capable of
identifying a set of specific entity types in historical texts (in Swedish). You are provided with
a training set (not included here, but shared separately) and a test set for final evaluation.
Dataset and Evaluation Metric
● Training Set: A set of historical text samples labeled with named entities. You may
train your model on this dataset only.
● Test Set: Contains the samples for final evaluation. You must generate predictions
(id, entities) for these examples.
● Evaluation: We will use Nervaluate in strict mode to compute precision, recall, and
F1-score. The final score used for ranking is the strict F1-score.
The entity types are as follows:
● EVN: Specific events (historical events, gatherings, incidents, etc.)
● LOC: Geographical locations (countries, cities, rivers, etc.)
● MSR-AREA: Surface/area measurements (square meters, hectares, etc.)
● MSR-DIST: Distance measurements (kilometers, miles, etc.)
● MSR-LEN: Length/height measurements (meters, centimeters, etc.)
● MSR-MON: Monetary values (any currency, interest rates, etc.)
● MSR-OTH: Other measurements not covered by other MSR categories
● MSR-VOL: Volume measurements (liters, cubic meters, etc.)
● MSR-WEI: Weight measurements (kilograms, pounds, etc.)
● OCC: Occupations or professions (doctor, carpenter, etc.)
● ORG-COMP: Company (any commercial enterprise, large or small)
● ORG-INST: Institutions (schools, universities, public authorities, NGOs, etc.)
● ORG-OTH: Other organizations not fitting the above categories
● PER: Persons (real, fictional, mythological)
● SYMP: Symptoms (medical or health-related conditions)
● TME-DATE: Specific calendar dates
● TME-INTRV: Time intervals with defined start/end
● TME-TIME: Time expressions that do not necessarily refer to specific dates/intervals
● WRK: Works of art (books, plays, brand names, newspapers, magazines, etc.)
Grading Policy
Restrictions
1. You must use only the provided training dataset.
2. No additional external data is allowed.
3. You may only use base models. In other words, no fine-tuned models that have
been trained on any other external dataset are permitted.
4. The final evaluation is performed on the test set only.
Submission Details
● Submission Format: A single CSV file including exactly two columns:
1. id (matching the test set IDs)
2. entities: A list of dictionaries (as a string) where each dictionary has label,
start, and end. Example row:
id,entities
123,"[{'label': 'PER', 'start': 36, 'end': 68}, {'label':
'LOC', 'start': 74, 'end': 86}]"
In this assignment, your goal is to build a Named Entity Recognition (NER) model capable of
identifying a set of specific entity types in historical texts (in Swedish). You are provided with
a training set (not included here, but shared separately) and a test set for final evaluation.
Dataset and Evaluation Metric
● Training Set: A set of historical text samples labeled with named entities. You may
train your model on this dataset only.
● Test Set: Contains the samples for final evaluation. You must generate predictions
(id, entities) for these examples.
● Evaluation: We will use Nervaluate in strict mode to compute precision, recall, and
F1-score. The final score used for ranking is the strict F1-score.
The entity types are as follows:
● EVN: Specific events (historical events, gatherings, incidents, etc.)
● LOC: Geographical locations (countries, cities, rivers, etc.)
● MSR-AREA: Surface/area measurements (square meters, hectares, etc.)
● MSR-DIST: Distance measurements (kilometers, miles, etc.)
● MSR-LEN: Length/height measurements (meters, centimeters, etc.)
● MSR-MON: Monetary values (any currency, interest rates, etc.)
● MSR-OTH: Other measurements not covered by other MSR categories
● MSR-VOL: Volume measurements (liters, cubic meters, etc.)
● MSR-WEI: Weight measurements (kilograms, pounds, etc.)
● OCC: Occupations or professions (doctor, carpenter, etc.)
● ORG-COMP: Company (any commercial enterprise, large or small)
● ORG-INST: Institutions (schools, universities, public authorities, NGOs, etc.)
● ORG-OTH: Other organizations not fitting the above categories
● PER: Persons (real, fictional, mythological)
● SYMP: Symptoms (medical or health-related conditions)
● TME-DATE: Specific calendar dates
● TME-INTRV: Time intervals with defined start/end
● TME-TIME: Time expressions that do not necessarily refer to specific dates/intervals
● WRK: Works of art (books, plays, brand names, newspapers, magazines, etc.)
Grading Policy
Restrictions
1. You must use only the provided training dataset.
2. No additional external data is allowed.
3. You may only use base models. In other words, no fine-tuned models that have
been trained on any other external dataset are permitted.
4. The final evaluation is performed on the test set only.
Submission Details
● Submission Format: A single CSV file including exactly two columns:
1. id (matching the test set IDs)
2. entities: A list of dictionaries (as a string) where each dictionary has label,
start, and end. Example row:
id,entities
123,"[{'label': 'PER', 'start': 36, 'end': 68}, {'label':
'LOC', 'start': 74, 'end': 86}]"