The word, and what it was doing there

Between 1992 and 2023 the UN Security Council produced 6,582 meeting records holding 106,302 speeches. In 3,273 of them — 3.08% — someone said genocide. This site asks which speeches those were, and what the word was doing in them.

Speeches in the corpus: 106,302
Speeches using genocid*: 3,273
Occurrences of the word: 6,092
Peak year by share: 1994

Why 2014 tops the raw count

2014 carries more occurrences of the word than 1994 does: 659 against 228. That is true, and it is mostly a side effect of a Council that talks more about everything. The number of speeches held each year grew roughly 7.45× across the period, so a raw count can rise while the habit behind it stays flat. Dividing by the speeches actually held is what every figure below does.

Occurrences and share of speeches, 1992–2023

Did the Council come to talk about genocide more, or simply to talk more?

Drawing…

View annual values as a table

Year	Occurrences	Share of speeches
1992	38	2.35%
1993	100	3.87%
1994	228	6.46%
1995	131	4.95%
1996	121	3.59%
1997	24	1.31%
1998	75	3.54%
1999	146	4.77%
2000	219	3.35%
2001	137	2.98%
2002	114	2.68%
2003	119	2.08%
2004	178	3.62%
2005	156	4.42%
2006	150	3.57%
2007	120	4.10%
2008	127	2.57%
2009	131	2.85%
2010	114	2.44%
2011	109	1.81%
2012	130	2.65%
2013	202	4.12%
2014	659	5.35%
2015	460	4.43%
2016	269	3.04%
2017	188	2.44%
2018	221	2.32%
2019	226	1.98%
2020	201	2.88%
2021	236	2.30%
2022	390	2.79%
2023	373	2.62%

How to read this

The bars count every occurrence of genocid* in a year, against the left-hand axis. The line is the share of that year's speeches that use it, against the right-hand axis. Select a year to open the individual lines behind it.

A statistical test that allows for how many speeches each year held asks whether the line is better described as one steady rate or as two. It settles on two, splitting at 2017, with the later rate 0.71× the earlier one.

What it does not show

A share of speeches says nothing about intensity: a speech that says the word once counts the same as a speech that repeats it twenty times. Both views are here for that reason.

The test re-runs its whole search on 2,000 simulated series in which the rate never changes, and reports how often chance alone produces a split this strong. It cannot tell an abrupt break from a gradual trend, treats each year as independent of the last, and ignores the way speeches cluster into meetings. The split describes the series; it is not a date on which something happened.

Where the rate does change

For genocide on its own, the best-supported split falls at 2017, and the rate after it is 0.71× the rate before. The wider atrocity core gathers five phrases at once: genocide, ethnic cleansing, crimes against humanity, war crimes and mass atrocity. Its split falls at 1996, with a ratio of 0.71×. Both results compare one stretch of years with another. Neither shows that Council language turned a corner in the year named.

Register share, 1992–2023

Which family of words does genocide sit in, and does that mix change over time?

View

accountability

11.92%

legal

8.59%

preventive

2.44%

commemorative

6.04%

core

3.08%

contentious

2.08%

35 reference dates

199220082023

View register shares as a table

Year	accountability	commemorative	contentious	core	legal	preventive
1992	1.96%	1.08%	1.56%	2.35%	10.65%	0.10%
1993	3.36%	0.64%	1.15%	3.87%	9.88%	0.07%
1994	5.09%	1.36%	1.36%	6.46%	10.90%	0.30%
1995	7.79%	4.80%	0.58%	4.95%	8.22%	0.36%
1996	5.88%	1.22%	1.58%	3.59%	3.30%	0.14%
1997	5.59%	1.64%	1.31%	1.31%	4.16%	0.33%
1998	8.93%	1.01%	1.60%	3.54%	6.07%	1.35%
1999	8.80%	3.78%	1.61%	4.77%	11.47%	2.60%
2000	9.37%	2.99%	2.57%	3.35%	7.31%	1.32%
2001	6.86%	1.02%	2.07%	2.98%	5.08%	1.09%
2002	6.48%	2.21%	1.88%	2.68%	6.01%	1.45%
2003	10.32%	1.92%	1.82%	2.08%	6.62%	0.83%
2004	13.62%	1.99%	2.34%	3.62%	8.83%	2.63%
2005	17.08%	3.94%	1.33%	4.42%	8.20%	3.30%
2006	14.24%	2.71%	1.96%	3.57%	7.38%	2.79%
2007	14.49%	2.85%	1.06%	4.10%	6.62%	4.64%
2008	13.95%	3.60%	2.17%	2.57%	7.34%	2.14%
2009	15.48%	6.22%	1.13%	2.85%	7.45%	3.47%
2010	14.20%	6.16%	1.48%	2.44%	5.83%	3.16%
2011	12.23%	3.38%	1.38%	1.81%	6.19%	2.91%
2012	16.49%	4.68%	1.54%	2.65%	9.16%	2.75%
2013	20.07%	6.31%	1.87%	4.12%	11.69%	3.03%
2014	14.89%	7.72%	2.62%	5.35%	11.32%	3.96%
2015	14.34%	9.98%	2.28%	4.43%	11.94%	3.59%
2016	10.77%	5.51%	2.54%	3.04%	9.45%	3.28%
2017	11.29%	5.25%	2.22%	2.44%	9.07%	2.29%
2018	12.65%	5.93%	2.19%	2.32%	8.78%	2.48%
2019	10.59%	9.37%	2.14%	1.98%	6.73%	2.29%
2020	14.22%	15.48%	2.90%	2.88%	9.54%	2.92%
2021	14.00%	11.10%	2.76%	2.30%	9.19%	3.23%
2022	11.24%	7.81%	2.31%	2.79%	9.81%	2.15%
2023	9.96%	8.21%	2.52%	2.62%	9.46%	1.99%

Each row is scaled to its own maximum · the number at the right is the share across the whole period

How to read this

The word list behind this site is sorted into six registers — families of vocabulary that do similar work in a speech. Each row is the share of a year's speeches using at least one word from one family: the core word itself; legal language that names the crime formally; preventive language; commemorative; contentious (denial, glorification); and accountability (courts, tribunals, impunity).

Accountability and legal language dominate throughout. The core word runs about ten times lower than either.

What it does not show

These rows do not add up to a whole. One speech can use four registers at once, and it is counted in all four. Stacking these rows on top of one another would be wrong.

Each row is scaled to its own maximum, so the shapes are comparable and the levels are not — read the level off the share at the right of each row.

Sorting the words into six registers is a proposal about how this vocabulary groups. It is written down in config/lexicon.yml, open to disagreement, and a starting point rather than a result.

Why 2014 tops the raw count

Occurrences and share of speeches, 1992–2023

Where the rate does change

Register share, 1992–2023

Where to go from here