The Knesset Corpus Project

Dr. Nurit Melnik


The Knesset Corpus Project builds directly upon the foundational Knesset Corpus created by Goldin, Howell, Ordan, Rabinovich & Wintner (2025) - a large digital collection of Israeli parliamentary plenary and committee protocols comprising over 30 million sentences covering the period from 1992 to 2024. Our project focuses on the most linguistically diverse, unscripted, and spontaneous section of the parliamentary record – the committee meeting protocols – with data spanning the 15th to 25th terms of the Knesset (1999–2026).

To make this valuable spoken-language resource fully accessible to researchers across disciplines, my colleague Noam Ordan and I have accomplished three major milestones:

1. Improved morphosyntactic and socio-political annotations 

2. Full integration in Sketch Engine

3. Matching reference corpus 


Major Achievements & Key Features 

1. Improved morphosyntactic and socio-political annotations 

• The committee protocols are enriched using Noam Ordan's PARSAN, a unified model that performs morphological segmentation, lemmatization, POS tagging, and Universal Dependencies (UD) parsing in a single pass. 

• This structural annotation enables precise, deep syntactic searches (e.g., querying relations like nsubj, obj, nmod) that go far beyond simple word or string lookups. 

• Building on the metadata schema of Goldin et al. (2025), every utterance is linked to speaker and faction attributes at the moment of speaking (including speaker gender, ethnic group, faction affiliation, coalition vs. opposition status, and political orientation). • Language use can be analyzed across all socio-political categories.

2. Full Integration in Sketch Engine 

• The corpus is hosted on the user-friendly Sketch Engine platform, making it searchable via Corpus Query Language (CQL) without requiring programming skills. 

3. The Hebrew Reference Corpus 

To enable systematic comparisons between parliamentary speech and general written Hebrew, we have built and uploaded a dedicated Hebrew Reference Corpus, based on HPLT data: 

• The corpus contains approximately 100 million words across 165,000 web documents, sampled across eight distinct text genres. • Balanced sampling imposes strict caps on document counts per site to ensure a representative sample of written (web) Hebrew.

• The Hebrew reference corpus is parsed using the exact same PARSAN architecture as the Knesset committee corpus, allowing researchers to run comparative keyword, and syntactic distribution analyses against an identical analytical baseline.


Research Applications

The combination of PARSAN morphosyntactic annotations, Sketch Engine CQL, speaker metadata, and the reference corpus enables researchers to address a wide array of empirical questions. The following are just a few illustrative examples: 

1. Language, Style & Spoken Discourse (Linguistics)

• Interactional features of spoken Hebrew: How do speakers navigate speech phenomena such as discourse markers, turn-taking, and stance constructions? 

• Spoken vs. written Hebrew: Which keywords, phrases, and grammatical constructions uniquely characterize spoken language (in the parliamentary committee setting) when compared against general written Hebrew? 

• Language change & stylistic shifts: How have Hebrew grammar, vocabulary, and syntactic structures evolved over nearly thirty years of committee protocols? 

2. Politics, Society & Communication (Social Sciences) 

• Ideology, sector & faction alignments: How do different ideological blocs (Left, Right, Center), ethno-religious sectors, and political factions frame key social, legal, and economic issues?

• Gender & agenda setting: Do male and female lawmakers prioritize different legislative topics (e.g., welfare, healthcare, defense, finance) across different committee settings? 

• Diachronic policy shifts: How has the political rhetoric surrounding major national issues (e.g., governance, judicial reform, conscription, equality) evolved across parliamentary terms from 1992 to the present?