Introduction

• There has been a recent turn toward public participation in constitutional design (Choudhry and Tushnet 2020).

• Citizen consultations produce voluminous textual data that are challenging to structure and analyze (Houlihan and Bisarya 2021).

• The value of such consultations has been largely symbolic (Blount 2011; Ginsburg, Elkins, and Blount 2009)—until now.

• Using data from Chile’s 2016 consultation process, we develop an approach that we call conceptual ecology.

Constitutions

Research questions

1. Idea Representation: How do the ideas suggested by Chilean participants compare with the universe of ideas in the world’s constitutions?

2. Consultation Structure: What is the relationship between topics raised by participants and the administrative level at which consultations were conducted?

3. Residential Context: What is the relationship between topics raised by participants and the type of municipality where consultation participants live?

Data

1. Reference ontology: 334 constitutional topics defined by the Comparative Constitutions Project (CCP) (Elkins and Ginsburg 2007).

2. National constitutions: 192 current constitutions transformed into 163,102 text segments.

3. Chilean consultation data: 264,800 consultation responses collected at three administrative levels (local, provincial, regional).

Methodology

• Google’s Universal Sentence Encoder (USE-3) encodes multilingual textual data, including Spanish, as vectors in a 512-dimension space.

• The angular distance between two sentence-level embeddings provides a measure of the proximity and therefore semantic similarity of two sentences.

• Semantic similarity scores are values between 0.0 (no shared meaning) and 1.0 (identical words and meaning).

• Our threshold for identifying “significant” results is 0.7.

Results

Top-10 topics

All topics

Fig 1. Comparison of the ranked distribution of constitution sections (red) and aggregated consultation responses (green) for all 334 CCP topics.

Fig 2. Comparison of the ranked distribution of constitution sections (red) and aggregated consultation responses (green) across 12 categories of constitutional topics.

Rights-related topics

Fig 3. Comparison of the ranked distribution of constitution sections (red) and aggregated consultation responses (green) across the 30 rights topics most prevalent in national constitutions.

Fig 4. Comparison of the ranked distribution of consultation responses (green) and aggregated constitution sections (red) across the 30 rights topics that are most prevalent in consultation responses.

Institutions-related topics

Fig 5. Comparison of the ranked distribution of constitution sections (red) and consultation responses (green) across the 30 institutional topics that are most prevalent in national constitutions.

Fig 6. Comparison of the ranked distribution of consultation responses (green) and constitution sections (red) across the 30 institutional topics that are most prevalent in consultation responses.

Consultation vs. Constitutions

Conclusions

Our methodology:

• Analyzes and translates valuable consultation data into something actionable by constitutional practitioners, researchers, and civil society.

• Has the power to increase the reach and impact of public input into constitution-making, which is too often underutilized.

• Can be applied to many other projects, such as ontology alignment.

Ontology alignment