unstructuredio/SCORE-Bench · Datasets at Hugging Face

text stringlengths 0 2.18k
Climate change can result in a reduction in crop yields, livestock production and fish catch. This will place added stress on global food security and reinforce the existing challenges in food production. It is also likely that there will be greater volatility in food availability and price, as well as greater uncertainty and higher risks associated with access to food in international markets. This could encourage hoarding or greater speculative behavior, further destabilizing food markets.


--------------------------------------------------- Unstructured Sub-Title Begin
Increase in food prices due to rising production costs
--------------------------------------------------- Unstructured Sub-Title End
Food production costs are likely to increase due to increasing costs of climate adaptation and mitigation. Given the increase in temperature and drought occurrences, it is likely that feed prices will increase. Water scarcity, rise in feed prices and increase in demand for quality feed and energy for climate adaptation will drive up production costs. This will increase food prices and hence access to food.
Global commitment in mitigating climate change may alter the costs of energy and the way farmers farm their livestock. In Australia, and
in all likelihood, the United States and Canada, mitigation costs will be borne by farmers, as they have to avoid business-as-usual production. In all regions including Southeast Asia, India and China, adaptation costs may increase. As a result, there will be an increase in cost of energy and water inputs.
Given the increase in temperature and drought occurrences, it is also likely that animal feed prices will increase. The implementation of carbon tax in exporting nations will also push up food prices. The use of biofuel as a substitute for oil may continue to artificially inflate demand for already reduced grain supplies. Carbon tax on high emissions from livestock production will also increase production costs, which may be transferred to consumers.


--------------------------------------------------- Unstructured Sub-Title Begin
Shifting food production centres
--------------------------------------------------- Unstructured Sub-Title End
To avoid higher costs of adaptation, agriculture and livestock production are likely to shift to regions with more favourable climate conditions — regions of higher latitude or altitude. This will change the global distribution of food production and export, potentially opening up new food source countries and new supply chains. The balance-of-power between food exporters and importers will shift, with repercussions for regional and bilateral relations.
--------------------------------------------------- Unstructured Plain Text Format 1.0.4
--------------------------------------------------- Unstructured Page Header Begin
--------------------------------------------------- Unstructured Page Number Block Begin
8
--------------------------------------------------- Unstructured Page Number Block End
--------------------------------------------------- Unstructured Page Header End


--------------------------------------------------- Unstructured Title Begin
Key Recommendations
--------------------------------------------------- Unstructured Title End


--------------------------------------------------- Unstructured Image Begin
SALINE
--------------------------------------------------- Unstructured Image End
--------------------------------------------------- Unstructured Caption Begin
An IRRI test field where saline resistant rice varieties are cultivated
--------------------------------------------------- Unstructured Caption End
--------------------------------------------------- Unstructured Caption Begin
Source: Deutsche Welle/ flickr
--------------------------------------------------- Unstructured Caption End


Exporting countries will prioritise local markets and needs if there are production failures, particularly as a result of climate change. Therefore, exporting governments will likely enhance export restrictions in times of food emergency. For importing countries, this could translate to reduced stability of food supply and access, and greater price volatility of cereals and vegetables, as well as increase in prices of meat and eggs. Specifically, price volatility can affect the availability of key commodities such as rice and wheat which are stockpiled by numerous countries. Against these scenarios, it is critical that importing countries adapt to future climate change by developing forward looking strategies:


--------------------------------------------------- Unstructured Sub-Title Begin
1. Adopt a ‘no-regrets’ approach
--------------------------------------------------- Unstructured Sub-Title End
Both exporting and importing countries should adopt a “no-regrets” approach to adaptation actions in food systems. “No regrets” approach refers to the need to take proactive adaptation actions. This is to preempt adverse conditions given the lack of accuracy in future climate projections. Most importantly, as climate impacts will affect domestic production, it is necessary
for producing countries to identify potential impacts and possible adaptation actions on local production centres. Early identification of the impacts of climate change on current crop yields, livestock production and fish in producing countries will be important for these countries’ food strategy decisions. For importing countries, such an approach ensures minimum supplies in the least and cuts down on fears of price volatility.


--------------------------------------------------- Unstructured Sub-Title Begin
2. Adopt an ‘adaptation without borders’ approach
--------------------------------------------------- Unstructured Sub-Title End
Food importing countries can also promote ‘adaptation without borders’²⁴ (AWB) as a framework to continuously monitor global and regional food production and trade. AWB suggests that no country (either consumers or producers) can survive without looking beyond their borders. Importing countries that have the capacity to invest in research and development should provide long term support in scientific research and technological innovation to improve crop yields in potential or emerging producers and existing exporting countries. These include investments in the construction of better weather and climate monitoring and early-warning systems, both regionally and nationally. Importing countries should also provide the expertise or technology required for the construction of adequate drying and storage equipment and facilities to reduce postharvest losses which are about 20 per cent for rice in Southeast Asia,²⁵ making more food available for populations.
Capable Asian governments should also support international food research centres such as International Rice Research Institute (IRRI), International Livestock Research
²⁴ Magnus Benzie, Oskar Wallgren and Marion Davis 2013. Adaptation without borders? How understanding indirect impacts could change countries’ approach to climate risks. Stockholm Environment Institute. Available at: http://www.sei-international.org/mediamanager/documents/Publications/Climate/SEI-DB-2013-Adaptation-Without-Borders.pdf
²⁵ Lassa JA, 2012, Emerging ‘Agricultural Involution’ in Indonesia: Impact of Natural Hazards and Climate Extremes on Agricultural Crops and Food System in Sawada, Y. and S. Oum (eds.), Economic and Welfare Impacts of Disasters in East Asia and Policy Responses. ERIA Research Project Report 2011-8, Jakarta: ERIA. pp.601-640.
--------------------------------------------------- Unstructured Plain Text Format 1.0.0


--------------------------------------------------- Unstructured Image Begin
Step 1: Generate a structured metadata representation of the document.
GPTriage: Question Answering over Long, Structured Documents
Anonymous ACL submission
Abstract
Large Language Models (LLMs) have recently shown remarkable abilities to generate human-like text for tasks such as question answering (Q/A). These models have issues with document Q/A where the document or collection of relevant documents are unable to fit in the small context length of an LLM. To overcome this issue, most existing works focus on retrieving the relevant context from the document(s), representing them as a stream of text. However, documents such as PDFs, web pages, and presentations are naturally structured with different pages, tables, figures and so on. Representing such structured documents as a text stream is incongruous with a users’ mental model of these documents with rich structure. When a system has to query the document for context, this incongruity is brought to the fore, and seemingly trivial questions can trip up the QA system. To bridge this fundamental gap in existing LLMs, we propose an approach called GPTriage that enables models to retrieve context based on either structure or content. Experiments demonstrate the effectiveness of the proposed GPTriage-augmented models across several classes of questions that existing retrieval-augmented LLMs fail. To facilitate further research on this fundamental problem, we release our benchmark dataset consisting of 900+ human-generated questions over 90 structured documents from 10 different categories of question types for document Q/A.


1 Introduction
When a document or collection of documents does not fit in the limited context window of an LLM, different strategies can be deployed to fetch relevant context. Current approaches often rely on a pre-retrieval step to fetch the relevant context from documents (Pereira et al., 2023; Gao et al., 2022). These pre-retrieval steps tend to represent the document as a stream of text chunks, that can be retrieved with natural language queries. However, many document types include rich structure, like web pages, PDFs, presentations, and so on. For these structured documents, representing the document as a text stream is often incongruous with users’ mental model of a structured document. This can lead to questions that, to users, may be trivially answerable, but fail with common/current approaches to document Q/A using LLMs. For instance, consider the following two questions:


Q1 “Can you summarize the key takeaways from pages 5-7?”


Q2 “What year [in table 3] has the maximum revenue?”


In the first question, document structure is explicitly referenced (“pages 5-7”). In the second question, document structure is implicitly referenced (“in table 3”). In both cases, a representation of document structure is necessary to identify the salient context and answer the question. Considering the document as a text stream discards the relevant structure needed to answer these questions.


We propose addressing this by allowing models to retrieve context based on either structure or content. Our approach, which we refer to as GPTriage, gives models access to metadata about the structure of the document. We accomplish this by augmenting prompts with both document structure metadata and a set of model-callable retrieval functions over various types of structure. For instance, we introduce the fetch_pages(pages: list[int]) function, which allows the model to fetch a list of pages. We show that by providing the structure and the ability to issue queries over that structure, GPTriage-augmented models can reliably answer several classes of questions that plain retrieval-augmented LLMs could not.


In order to evaluate our approach, we construct a dataset of roughly 900 human-generated questions


1
Document Section Section Section … Section
H1 P UL P
L1 L1
Q1: “Can you summarize the key takeaways from pages 5-7?”
Q2: “What year [in table 3] has the maximum revenue?”
Document Metadata Representation
Pages [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]
…
Section Title: "2 Related Works" Pages: [2, 3]
Section Title: "2.1 Tool and Retrieval Augmented LLMs" Pages: [2]
…
Table Caption: "Table 1: GPTriage functions for Document QA" Pages: [4]

Climate change can result in a reduction in crop yields, livestock production and fish catch. This will place added stress on global food security and reinforce the existing challenges in food production. It is also likely that there will be greater volatility in food availability and price, as well as greater uncertainty and higher risks associated with access to food in international markets. This could encourage hoarding or greater speculative behavior, further destabilizing food markets.

--------------------------------------------------- Unstructured Sub-Title Begin

Increase in food prices due to rising production costs

--------------------------------------------------- Unstructured Sub-Title End

Food production costs are likely to increase due to increasing costs of climate adaptation and mitigation. Given the increase in temperature and drought occurrences, it is likely that feed prices will increase. Water scarcity, rise in feed prices and increase in demand for quality feed and energy for climate adaptation will drive up production costs. This will increase food prices and hence access to food.

Global commitment in mitigating climate change may alter the costs of energy and the way farmers farm their livestock. In Australia, and

in all likelihood, the United States and Canada, mitigation costs will be borne by farmers, as they have to avoid business-as-usual production. In all regions including Southeast Asia, India and China, adaptation costs may increase. As a result, there will be an increase in cost of energy and water inputs.

Given the increase in temperature and drought occurrences, it is also likely that animal feed prices will increase. The implementation of carbon tax in exporting nations will also push up food prices. The use of biofuel as a substitute for oil may continue to artificially inflate demand for already reduced grain supplies. Carbon tax on high emissions from livestock production will also increase production costs, which may be transferred to consumers.

--------------------------------------------------- Unstructured Sub-Title Begin

Shifting food production centres

--------------------------------------------------- Unstructured Sub-Title End

To avoid higher costs of adaptation, agriculture and livestock production are likely to shift to regions with more favourable climate conditions — regions of higher latitude or altitude. This will change the global distribution of food production and export, potentially opening up new food source countries and new supply chains. The balance-of-power between food exporters and importers will shift, with repercussions for regional and bilateral relations.

--------------------------------------------------- Unstructured Plain Text Format 1.0.4

--------------------------------------------------- Unstructured Page Header Begin

--------------------------------------------------- Unstructured Page Number Block Begin

8

--------------------------------------------------- Unstructured Page Number Block End

--------------------------------------------------- Unstructured Page Header End

--------------------------------------------------- Unstructured Title Begin

Key Recommendations

--------------------------------------------------- Unstructured Title End

--------------------------------------------------- Unstructured Image Begin

SALINE

--------------------------------------------------- Unstructured Image End

--------------------------------------------------- Unstructured Caption Begin

An IRRI test field where saline resistant rice varieties are cultivated

--------------------------------------------------- Unstructured Caption End

--------------------------------------------------- Unstructured Caption Begin

Source: Deutsche Welle/ flickr

--------------------------------------------------- Unstructured Caption End

Exporting countries will prioritise local markets and needs if there are production failures, particularly as a result of climate change. Therefore, exporting governments will likely enhance export restrictions in times of food emergency. For importing countries, this could translate to reduced stability of food supply and access, and greater price volatility of cereals and vegetables, as well as increase in prices of meat and eggs. Specifically, price volatility can affect the availability of key commodities such as rice and wheat which are stockpiled by numerous countries. Against these scenarios, it is critical that importing countries adapt to future climate change by developing forward looking strategies:

--------------------------------------------------- Unstructured Sub-Title Begin

1. Adopt a ‘no-regrets’ approach

--------------------------------------------------- Unstructured Sub-Title End

Both exporting and importing countries should adopt a “no-regrets” approach to adaptation actions in food systems. “No regrets” approach refers to the need to take proactive adaptation actions. This is to preempt adverse conditions given the lack of accuracy in future climate projections. Most importantly, as climate impacts will affect domestic production, it is necessary

for producing countries to identify potential impacts and possible adaptation actions on local production centres. Early identification of the impacts of climate change on current crop yields, livestock production and fish in producing countries will be important for these countries’ food strategy decisions. For importing countries, such an approach ensures minimum supplies in the least and cuts down on fears of price volatility.

--------------------------------------------------- Unstructured Sub-Title Begin

2. Adopt an ‘adaptation without borders’ approach

--------------------------------------------------- Unstructured Sub-Title End

Food importing countries can also promote ‘adaptation without borders’²⁴ (AWB) as a framework to continuously monitor global and regional food production and trade. AWB suggests that no country (either consumers or producers) can survive without looking beyond their borders. Importing countries that have the capacity to invest in research and development should provide long term support in scientific research and technological innovation to improve crop yields in potential or emerging producers and existing exporting countries. These include investments in the construction of better weather and climate monitoring and early-warning systems, both regionally and nationally. Importing countries should also provide the expertise or technology required for the construction of adequate drying and storage equipment and facilities to reduce postharvest losses which are about 20 per cent for rice in Southeast Asia,²⁵ making more food available for populations.

Capable Asian governments should also support international food research centres such as International Rice Research Institute (IRRI), International Livestock Research

²⁴ Magnus Benzie, Oskar Wallgren and Marion Davis 2013. Adaptation without borders? How understanding indirect impacts could change countries’ approach to climate risks. Stockholm Environment Institute. Available at: http://www.sei-international.org/mediamanager/documents/Publications/Climate/SEI-DB-2013-Adaptation-Without-Borders.pdf

²⁵ Lassa JA, 2012, Emerging ‘Agricultural Involution’ in Indonesia: Impact of Natural Hazards and Climate Extremes on Agricultural Crops and Food System in Sawada, Y. and S. Oum (eds.), Economic and Welfare Impacts of Disasters in East Asia and Policy Responses. ERIA Research Project Report 2011-8, Jakarta: ERIA. pp.601-640.

--------------------------------------------------- Unstructured Plain Text Format 1.0.0

--------------------------------------------------- Unstructured Image Begin

Step 1: Generate a structured metadata representation of the document.

GPTriage: Question Answering over Long, Structured Documents

Anonymous ACL submission

Abstract

Large Language Models (LLMs) have recently shown remarkable abilities to generate human-like text for tasks such as question answering (Q/A). These models have issues with document Q/A where the document or collection of relevant documents are unable to fit in the small context length of an LLM. To overcome this issue, most existing works focus on retrieving the relevant context from the document(s), representing them as a stream of text. However, documents such as PDFs, web pages, and presentations are naturally structured with different pages, tables, figures and so on. Representing such structured documents as a text stream is incongruous with a users’ mental model of these documents with rich structure. When a system has to query the document for context, this incongruity is brought to the fore, and seemingly trivial questions can trip up the QA system. To bridge this fundamental gap in existing LLMs, we propose an approach called GPTriage that enables models to retrieve context based on either structure or content. Experiments demonstrate the effectiveness of the proposed GPTriage-augmented models across several classes of questions that existing retrieval-augmented LLMs fail. To facilitate further research on this fundamental problem, we release our benchmark dataset consisting of 900+ human-generated questions over 90 structured documents from 10 different categories of question types for document Q/A.

1 Introduction

When a document or collection of documents does not fit in the limited context window of an LLM, different strategies can be deployed to fetch relevant context. Current approaches often rely on a pre-retrieval step to fetch the relevant context from documents (Pereira et al., 2023; Gao et al., 2022). These pre-retrieval steps tend to represent the document as a stream of text chunks, that can be retrieved with natural language queries. However, many document types include rich structure, like web pages, PDFs, presentations, and so on. For these structured documents, representing the document as a text stream is often incongruous with users’ mental model of a structured document. This can lead to questions that, to users, may be trivially answerable, but fail with common/current approaches to document Q/A using LLMs. For instance, consider the following two questions:

Q1 “Can you summarize the key takeaways from pages 5-7?”

Q2 “What year [in table 3] has the maximum revenue?”

In the first question, document structure is explicitly referenced (“pages 5-7”). In the second question, document structure is implicitly referenced (“in table 3”). In both cases, a representation of document structure is necessary to identify the salient context and answer the question. Considering the document as a text stream discards the relevant structure needed to answer these questions.

We propose addressing this by allowing models to retrieve context based on either structure or content. Our approach, which we refer to as GPTriage, gives models access to metadata about the structure of the document. We accomplish this by augmenting prompts with both document structure metadata and a set of model-callable retrieval functions over various types of structure. For instance, we introduce the fetch_pages(pages: list[int]) function, which allows the model to fetch a list of pages. We show that by providing the structure and the ability to issue queries over that structure, GPTriage-augmented models can reliably answer several classes of questions that plain retrieval-augmented LLMs could not.

In order to evaluate our approach, we construct a dataset of roughly 900 human-generated questions

1

Document Section Section Section … Section

H1 P UL P

L1 L1

Q1: “Can you summarize the key takeaways from pages 5-7?”

Q2: “What year [in table 3] has the maximum revenue?”

Document Metadata Representation

Pages [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]

…

Section Title: "2 Related Works" Pages: [2, 3]

Section Title: "2.1 Tool and Retrieval Augmented LLMs" Pages: [2]

…

Table Caption: "Table 1: GPTriage functions for Document QA" Pages: [4]