FinCausal-2023 Shared Task: “Financial Document Causality Detection”

To be held at The 5th Financial Narrative Processing Workshop (FNP 2023), Sorrento, Italy, 15-18 December 2023.

New: Access the CodaLab by clicking here

Important Dates:

Call for participation and registration: 3rd June 2023
Registration deadline: 28 June
Training set release: 14 July 2023
Test set release: 5 September 2023 11 September 2023
Systems submission deadline: 15 September 2023 19 September 2023 –> 29 September 2023
Release of results: 20 September 2023 25 September 2023 –> 2 October 2023
Paper submission deadline: October 30, 2023 (anywhere in the world).
Notification of acceptance: November 12, 2023
Camera-ready of accepted papers: November 20, 2023
Financial Narrative Processing Workshop: December 2023

Introduction

Financial analysis needs factual data and an explanation of the variability of these data. Data state facts but need more knowledge regarding how these facts materialised. Furthermore, understanding causality is crucial in studying decision-making processes. 

The Financial Document Causality Detection Task aims to develop an ability to explain, from external sources, why a transformation occurs in the financial landscape as a preamble to generating accurate and meaningful financial narrative summaries. Its goal is to evaluate which events or chain of events can cause a financial object to be modified or an event to occur, regarding a given context. 

For this task, we propose two subtasks, one in English and one in Spanish. In both of them, participants will be asked to identify, in causal sentences, which elements of the sentence relate to the cause, and which relate to the effect.

In the English subtask, the dataset has been sourced from various 2019 financial news articles provided by Qwam, along with additional SEC data from the Edgar Database. Additionally, we have augmented the dataset from FinCausal 2022, adding 500 new segments.

In the case of Spanish, the task dataset has been extracted from a corpus of Spanish financial annual reports from 2014 to 2018. This is the first year where we introduce a subtask in Spanish.

However, there will be some differences between the two subtasks:

Regarding the focus of the task. In English it centers on detecting causes and effects when the effects are quantified. Conversely, the Spanish task aims to detect all types of causes and effects, not necessarily limited to quantified effects.

Regarding the text segments. The English segments are made up of up to three sentences, while the Spanish task will involve complete paragraphs.

Despite the differences, it is important to note that the dataset will be presented in the same format for both tasks.

English Subtask

This shared task focuses on determining causality associated with a quantified fact. An event is defined as the arising or emergence of a new object or context regarding a previous situation. So, the task will emphasise the detection of causality associated with the transformation of financial objects embedded in quantified facts. Participants will be provided with a sample of text blocks extracted from financial news, labelled through inter-annotator agreement.

Cause and Effect Detection

This task is a relation detection task. The aim is to identify, in a causal sentence or text block, the causal elements and the consequential ones. Only one causal element and one effect are expected in each segment.

Text	Cause	Effect
Boussard Gavaudan Investment Management LLP bought a new position in shares of GENFIT S A/ADR in the second quarter worth about $199,000. Morgan Stanley increased its stake in shares of GENFIT S A/ADR by 24.4% in the second quarter.Morgan Stanley now owns 10,700 shares of the company’s stock worth $211,000 after purchasing an additional 2,100 shares during the period	Morgan Stanley increased its stake in shares of GENFIT S A/ADR by 24.4% in the second quarter	Morgan Stanley now owns 10,700 shares of the company’s stock worth $211,000 after purchasing an additional 2,100 shares during the period.
Zhao found himself 60 million yuan indebted after losing 9,000 BTC in a single day (February 10, 2014)	losing 9,000 BTC in a single day (February 10, 2014)	Zhao found himself 60 million yuan indebted

Table 1: Cause and Effect Detection Sample for English

Spanish Subtask

This shared task focuses on determining causality associated with both events or quantified facts. For this task, a cause can be the justification for a statement or the reason that explains a result. Participants will be provided with a sample of paragraphs extracted from Spanish financial annual reports, labelled through inter-annotator agreement. 

Cause and Effect Detection

This task is also a relation detection task. The aim is to identify, in a paragraph, the causal elements and the consequential ones. Only one causal element and one effect are expected in each paragraph.

Text	Cause	Effect
El deterioro de activos financieros se incrementó en un 267,4%, debido a ciertos impactos negativos de la cartera de clientes mayoristas y a la actualización del escenario macroeconómico.	debido a ciertos impactos negativos de la cartera de clientes mayoristas y a la actualización del escenario macroeconómico	El deterioro de activos financieros se incrementó en un 267,4%
El resultado atribuido creció un 10,8% hasta los 1.522 millones de euros, gracias al buen comportamiento de las comisiones y sobre todo a la significativa reducción de los gastos y a los menores saneamientos y provisiones.	gracias al buen comportamiento de las comisiones y sobre todo a la significativa reducción de los gastos y a los menores saneamientos y provisiones	El resultado atribuido creció un 10,8% hasta los 1.522 millones de euros

Table 2: Cause and Effect Detection Sample for Spanish

For the English and Spanish tasks, participants can use any method they see fit (regex, corpus linguistics, entity relationship models, deep learning methods) to identify the causes and effects. 

FinCausal Shared Task Organisers

Antonio Moreno-Sandoval (UAM, Spain)
Blanca Carbajo Coronado (UAM, Spain)
Doaa Samy (UCM, Spain)
Jordi Porta (UAM, Spain)
Dominique Mariko (Yseop, France)

Shared Task Contact

Participants can register for this shared task by filling this form, and get access to the datasets.

For any questions, please contact the organisers at fincausal.2023@gmail.com