Skip to main content

Research · Digital humanities · Preprint 2026

What the NO-DO talked about, and in which languages

The first computational analysis of the complete archive of Francoist Spain's state newsreel (1943–1981): 6,607 videos split into 39,957 news items classified into 12 themes and 61 sub-themes, and the language of every voiced fragment identified from the audio.

What this study is

The NO-DO (Noticiarios y Documentales) was the state newsreel of Francoist Spain, compulsorily screened in every cinema between January 1943 and May 1981. For three generations it was, until television spread, the only moving image of current affairs and the regime's most direct audiovisual voice. The archive is published online today by RTVE and the Filmoteca Española.

The historiography of the NO-DO is large (a recent bibliography lists some 300 works between 1987 and 2023), but the quantitative picture of what it showed rests on manual coding of partial samples: one period, one region, one topic. The archive exceeds 1,000 hours and 30,000 newsreel items. This study asks the whole archive two simple questions: what did the NO-DO talk about, and in which languages. It answers them with speech and text technology and large language models, stressing two things: that every aggregate figure can be traced back to a specific news item and timecode, and that the error of the automatic annotation is measured.

  1. A map of the spoken language and of the on-screen caption language across the archive, showing that the existing automatic transcription (forced to Spanish) had translated 84% of foreign speech into Spanish.
  2. The segmentation of all 6,607 videos into 39,957 timed news items, each classified into 12 themes and 61 sub-themes, with flags for scope (domestic or foreign) and for Franco on screen.
  3. A validation protocol (pilot, review of low-confidence items and a blind random sample) and a comparison with independent manual coding.
  4. A description of how the newsreel's themes changed between 1943 and 1981 and of how themes and languages relate.

The archive

The corpus is the complete NO-DO archive as RTVE publishes it: 6,607 videos in seven series. For every video the study uses RTVE's summary (a description written by archivists that lists the items in order), an earlier automatic transcription of the narration with timestamps (140,554 segments), OCR of on-screen captions (58,894, newsreel only) and the audio track, downloaded again to identify the language and then discarded.

The archive
SeriesVideosYearsItemsHours
NO-DO newsreel 4,064 1943–1981 31,232 664
Revista Imágenes 1,200 1945–1969 4,712 198
Colour documentaries 519 1949–1983 1,276 137
Historical Archive 460 1896–1970 1,172 93
Black-and-white documentaries 215 1939–1977 1,140 65
Imágenes del deporte 88 1968–1977 286 15
Alfonso XIII Royal Archive 61 1903–1930 139 11
Total 6,607 1896–1983 39,957 1,183

681 videos (almost all of the Alfonso XIII Royal Archive and more than half of the Historical Archive) are silent prints.

How it was done

01

Spoken language

The existing transcription was made with the language forced to Spanish: when a foreign voice was heard, the recogniser usually produced a translation. So the language was identified again from the audio: on the speech frames detected by Silero VAD, the language-identification module of Whisper large-v3 is applied. A fragment gets a language if p ≥ 0.6, or if p ≥ 0.4 and a fastText text classifier agrees with p ≥ 0.6; Catalan, Galician, Basque and Portuguese, which the model confuses with accented Spanish, require p ≥ 0.8. A language is said to appear in a video if it adds up to at least 5 seconds.

02

On-screen text

Captions were deduplicated per video (54,915 unique strings) and only given a language when fastText and Lingua agreed, or when they contained at least two words specific to Catalan, Basque or Galician. Most are proper names or too short and stay unassigned: 3,395 get a language.

03

Segmentation and classification

A two-level taxonomy of 12 themes and 61 sub-themes, designed through an iterative pilot. Items are classified by subject, not by place or by who appears: a dam opened by Franco is Economy and public works, with a separate flag for Franco's presence. For every video, Claude Haiku 5.5 receives the summary, the narration and the captions with timestamps and returns the list of contiguous items, each with title, start and end second, sub-theme, confidence, scope, Franco flag, places and people. The output is constrained by a JSON schema. The whole archive cost about 6 dollars (18.4 million input tokens, 40.1 million cached and 18.6 million output).

04

Review of uncertain items

The 2,683 items with confidence below 0.6 (6.7%) were re-examined one by one by Claude Opus 5.5 agents, with the context restricted to the item's span. The sub-theme changed in 1,036 cases, mostly section titles treated as items, “colour page” inserts and breaches of the subject-over-place rule. The corrections barely moved the aggregate distributions (under one point per theme and decade), because uncertain items are short.

Validation

In a 100-video pilot, Haiku agreed with a classification made with Claude Opus on 82.7% of screen seconds at theme level and 75.6% at sub-theme level. After the review, a simple random sample of 400 items (seed 2026) was drawn and two Opus agents classified them from scratch without seeing the automatic labels.

Validation
MeasureAgreementCohen's κ
Theme (12 classes) 88.0% [84.4, 90.8] 0.86
Sub-theme (61 classes) 84.0% [80.1, 87.3] 0.84
Theme, duration-weighted 89.2% —
Sub-theme, duration-weighted 86.7% —
Scope (domestic / foreign / colonial / unknown) 94.8% —
Franco on screen 99.8% —

This is agreement between two models of the same family with the same inputs, not accuracy against a human gold standard: it bounds the instability of the labels, not their validity. A human-coded sample is the first item of future work. In brackets, 95% Wilson intervals.

The languages of the NO-DO

Of the 564 hours of speech detected in the 5,924 videos with sound, 562.5 get a language. Spanish accounts for 95.8%. Then come English (2.5%, 13.9 hours in 441 videos), French, Italian, Portuguese and German. Catalan adds up to 27 minutes in 14 videos. 665 videos contain at least 5 seconds of a language other than Spanish.

A first finding concerns the archive itself: 84% of non-Spanish speech appears translated into Spanish in the existing transcription. Any language analysis based on the text would miss most foreign speech.

Minutes of speech by language, excluding Spanish

  • English : 831 min · 441 videos
  • French : 196 min · 70 videos
  • Italian : 119 min · 113 videos
  • Portuguese : 114 min · 26 videos
  • German : 79 min · 48 videos
  • Catalan : 27 min · 14 videos
  • Arabic : 22 min · 5 videos
  • Dutch : 12 min · 13 videos
  • Russian : 7 min · 6 videos
  • Galician (doubtful) : 5 min · 9 videos
  • Polish : 2 min · 2 videos
  • Japanese : 1 min · 1 videos
Whole archive. Whisper large-v3 language identification on the audio. Spanish (32,333 minutes, 95.8%) is left out so the others can be seen. Galician is doubtful: almost all of it is Spanish-language interviews misidentified.

Non-Spanish speech per year

0 20 40 60 19451950195519601965197019751980
Share of identified speech that is not Spanish, all series. Only years with at least half an hour of speech. The 1946–1951 peak is a property of the surviving prints, not of what audiences heard.
Show the data
Year% not Spanish
19434
19444.5
19454.2
194634
194739.7
194850.1
194954.5
195026
195125.2
19524.7
19535
19540.3
19551.5
19561
19570.9
19580
19591.3
19600.1
19611.1
19620.6
19630
19640.8
19653.8
19660.3
19671
19683.4
19690.3
19700.2
19710.5
19720.2
19730.4
19742.3
19750.7
19760.6
19770.2
19780
19790.5
19800.3
19810

The yearly share of non-Spanish speech is under 5% almost every year, except in 1946–1951, when it reaches 25–55%. The newsreel was not narrated in English: most surviving prints from those years lack the Spanish narration, and what is heard is the original sound of the American and British newsreel reports the NO-DO bought (Signal Corps, Movietone). Anyone using the audio from that period needs to bear this in mind.

On-screen captions by language, excluding Spanish

  • German : 244 captions
  • English : 241 captions
  • French : 153 captions
  • Italian : 63 captions
  • Dutch : 51 captions
  • Catalan : 21 captions
  • Hungarian : 11 captions
  • Portuguese : 10 captions
  • Latin : 2 captions
  • Norwegian : 2 captions
  • Turkish : 2 captions
Unique captions per video with an assigned language (3,395). 76% are in Spanish.

What can be read on screen tells another story. There is no Catalan caption before 1968 (21 in all); from the Transition onwards come election posters, banknotes of the republican Generalitat, comarques and exhibitions. The only Basque caption is the text of the 1978 Constitution. Foreign languages are posters, newspapers and signs in international reports.

Where the other languages appear

Catalan: the republican newsreel and the Transition

Catalan narration is concentrated in the republican material from 1937–1938 held by the Historical Archive: Espanya al dia, Laya Films' Catalan newsreel, and Quan el camperol és soldat. In the NO-DO proper there are only live snippets, such as Mario Cabré's words at his 1977 tribute.

Versions for audiences abroad

Among the documentaries there are editions cut for other audiences: in French (“Edición para Vichy”, “Actualités espagnoles” for Cannes, “España y el Líbano”), in Portuguese (Coimbra, Lisbon, Fátima, Kubitschek's visit) and in Arabic (the visits of the King of Jordan and of Saud of Arabia).

Foreign war footage

German comes mainly from the “Alemania 1940” series and other German war reports in the Historical Archive; French from Belgian and French productions of the 1930s; Russian from Soviet films of 1934 and 1937.

What does not appear

There is no spoken Basque. The Galician the model detects is almost always Spanish-language interviews misclassified. The newsreel's official narration is in Spanish throughout.

What the NO-DO talked about

The newsreel was split into 31,232 items (7.7 per video). Average length grows from about a minute in the 1940s and 1950s to two minutes in the 1970s and three and a half in 1980–81: the newsreel moves from many short pieces to fewer reports. The measure is screen time: a three-minute item weighs three times as much as a one-minute item.

Share of each theme by decade

Share of each theme by decade
Theme 1940s1950s1960s1970s1980–81
Culture and entertainment 17.519.124.836.845.6
Sport 20.623.123.721.27.2
Economy and public works 10.811.511.110.613.4
Society 7.710.410.38.111.7
International 18.78.56.90.71.8
Regime and politics 7.06.55.14.92.6
Science and technology 3.85.26.36.47.7
Armed forces 7.46.83.41.94.1
Religion 3.74.83.31.90.8
Tourism and landscape 1.11.32.85.94.3
Oddities and humour 1.52.52.31.50.7
Items5,8349,73710,6444,817199
Hours10617021815812
NO-DO newsreel. Percentage of each decade's news time (no headers or credits). The 1980s are 1980–81 only. Table 3 of the preprint.

Share of each theme by year, 1943–1981

  • Culture and entertainment

    max 48.8% · 1979

  • Sport

    max 26.7% · 1949

  • Economy and public works

    max 14.9% · 1980

  • International

    max 38.5% · 1945

  • Society

    max 29.3% · 1981

  • Regime and politics

    max 11.6% · 1978

  • Armed forces

    max 13.5% · 1946

  • Religion

    max 6.6% · 1958

  • Science and technology

    max 9.6% · 1980

  • Tourism and landscape

    max 8.2% · 1972

  • Oddities and humour

    max 3.5% · 1950

NO-DO newsreel. Share of each year's news time. Each theme has its own scale; the figure marks its maximum. Exact values are in the decade table.

From war to entertainment

International falls the most: 18.7% in the 1940s (71% of it the Second World War and its aftermath), 8.5% and 6.9% in the 1950s and 1960s (foreign politics and the Cold War, plus royalty and celebrities) and 0.7% in the 1970s. Culture and entertainment rises from 17.5% to 36.8% and becomes the late newsreel's dominant theme. Folklore leads the theme in the 1940s (29%), bullfighting peaks in the 1950s (16%) and then fades, cinema recedes, and art and heritage grow to almost half of the theme in 1980–81.

Sport, a constant

Sport takes a steady fifth of the time (20.6–23.7%) until the 1970s. Football is only 14–18% of it: the bulk is a long tail of other sports (boxing, tennis, equestrianism, sailing, pelota…), motor sport, athletics and winter sports. It only falls in the final years, when the NO-DO stops being compulsory.

The regime, inside other themes

Regime and politics as a theme of its own is small and shrinking (from 7.0% to 4.9%). Events devoted to Franco and state visits dominate it until the 1960s, Falange and Movement organisations fall from 26% to 7% of the theme, and the 1970s bring the Royal Household (28%) and the Transition (15%). Armed forces drops from 7.4% to 1.9%; in the 1940s and 1950s, half of the military time is foreign armies and equipment.

From developmentalism to tourism

Economy and public works stays around 11% throughout, but inside it moves from industry and agriculture (a quarter each in the 1940s) to trade fairs and commerce (a third in the 1960s and 1970s) and transport. Religion peaks in the 1950s (4.8%). Tourism and landscape grows from 1% to 5.9% in the 1970s, and Science and technology from 3.8% to 6.4%, with a space-race peak in the 1960s.

Turning inward

In the 1940s, events abroad take 52.8% of the time; the share falls to 45.5% in the 1950s, 40.8% in the 1960s and 27.2% in the 1970s, and to 7.2% in 1980–81. Colonial territories (Ifni, Sahara, Guinea and the Protectorate) never exceed about 1%.

Time devoted to events abroad, by year

0 20 40 60 80 19451950195519601965197019751980
NO-DO newsreel. Share of news time with foreign scope; the rest is Spain, the colonies or unknown.
Show the data
Year% foreign
194359.9
194461.6
194557.3
194639.3
194746.1
194849.8
194950.7
195047.4
195146.3
195248.4
195346.9
195445.3
195546.3
195646.6
195743.2
195844.1
195940.5
196042.6
196140.9
196243.3
196344
196445.3
196544
196639.8
196739.5
196834.6
196929.6
197026.8
197128.8
197228.5
197328
197429.6
197528.1
197634.7
197729.9
197811.8
19793.6
19803.7
198115.6

Franco on screen

Franco appears or attends in 1,860 of the 31,232 newsreel items (6.0%). As a share of screen time, his presence is strikingly stable: between 3.7% and 12.2% every year from 1943 to 1975 (mean 7.9%), with no clear trend, and zero after his death.

Only 42% of that time is in items whose theme is politics; the rest is public works, culture, armed forces and sport that he presides over. Politics is a minor theme, but the dictator is a recurring figure across all themes: the regime's presence is diffuse rather than thematic.

Franco on screen, by year

0 5 10 15 19451950195519601965197019751980 Franco's death (Nov. 1975)
NO-DO newsreel. Share of news time in items where Franco appears or attends, whatever the theme (a mere mention does not count).
Show the data
Year% with Franco
19438.6
19445
19456.1
194611.3
194711.1
19488.8
19498.1
19507.4
19516.7
19529.2
195312.2
195411.1
19559.4
19569.3
195710.3
19587.7
19599
19606.2
19618.9
19626.7
19637.1
19647.5
19656
19668.6
196710.1
19687.1
19696.1
19706.3
19717.4
19725
19737.1
19743.7
19755.1
19760
19770
19780
19790
19800
19810

Themes and languages

Crossing both analyses (non-Spanish speech spans of at least 5 seconds inside an item), foreign speech concentrates in culture, international, sport and armed forces. English dominates in sport, international, military and cultural items; German and French are heard mostly in international items, and Italian and Portuguese in cultural ones. Foreign voices enter the newsreel mainly through bought-in sport and world-news footage and through performances.

Which themes carry other languages

  • Culture and entertainment : 521 min · mostly English, Portuguese
  • International : 481 min · mostly English, German
  • Sport : 433 min · mostly English, French
  • Armed forces : 357 min · mostly English, French
  • Economy and public works : 216 min · mostly English, German
  • Society : 211 min · mostly English, French
  • Science and technology : 162 min · mostly English, French
  • Tourism and landscape : 126 min · mostly French, Portuguese
  • Regime and politics : 109 min · mostly English, Arabic
  • Oddities and humour : 82 min · mostly English, Italian
  • Religion : 73 min · mostly English, Italian
Minutes of non-Spanish speech inside items of each theme. Whole archive.

The other series

The series differ a great deal. Revista Imágenes is the most cultural (31% culture and 18% economy); the colour documentaries give a quarter to culture and another to tourism and landscape; in the Royal and Historical Archives the military and official events dominate, and Imágenes del deporte is 90% sport.

Share of each theme by series

Share of each theme by series
Series Culture and entertainmentSportEconomy and public worksInternationalRegime and politicsArmed forcesTourism and landscape
NO-DO newsreel 2522118653
Revista Imágenes 3113183248
Colour documentaries 25813311326
Black-and-white documentaries 17211011897
Historical Archive 155111392211
Alfonso XIII Royal Archive 121314325271
Imágenes del deporte 49010000
Share of each series' time, main themes.

Comparison with manual coding

For 1943–1956, Gámir Orueta (2026) counted 11,873 items in 1,460 issues from the printed programmes, 40.4% of them about Spain. The automatic segmentation of the 1,365 issues from that period available in the digital archive yields 12,645 (6.5% more, with 6.5% fewer issues), of which 41.1% are domestic. Both studies find Madrid over-represented (27.9% and 32.4% of domestic items).

For Catalonia in 1960–1975, Santacana Torres (2021) finds politics to be 5.2% of items; this study's matching figure is 3.0%. Both confirm politics as a minor theme.

Theme by theme

For each theme, the weight of its sub-themes in each decade of the newsreel and the longest confidently classified item of each sub-theme (from any series). The link opens the video on RTVE; the minute shows where the item starts. Item titles are those of the classification, in Spanish.

Culture and entertainment

Culture and entertainment: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Art and heritage 3.23.55.811.922.0
Folklore 5.14.84.45.710.1
Music and theatre 1.21.23.46.58.2
Bullfighting 2.33.13.43.30.5
Fashion 1.71.62.63.41.6
Cinema 2.02.32.12.42.1
Circus and variety 0.81.31.31.3·
Literature 0.80.91.21.61.2
Press and broadcasting 0.50.50.80.7·

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Sport

Sport: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Other sports 8.68.38.87.52.9
Football 3.53.83.43.90.6
Athletics 2.83.14.22.70.2
Motor sport 1.73.22.93.00.8
Winter sports 2.02.12.22.51.1
Cycling 1.41.71.31.00.7
Hunting and fishing 0.50.90.90.60.9

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Economy and public works

Economy and public works: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Trade and fairs 1.72.33.63.81.6
Industry 2.82.62.21.33.7
Agriculture 2.82.61.41.23.7
Dams and public works 1.62.01.61.41.4
Transport 1.41.21.42.32.6
Housing 0.60.90.70.60.6

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

International

International: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
World wars 13.30.30.2··
Foreign politics 2.53.12.80.10.5
Cold War 0.93.02.30.11.1
Royalty and celebrities 1.31.61.40.20.2
Latin America and the Philippines 0.50.30.20.2·
Life abroad 0.20.20.20.2·

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Society

Society: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Everyday life 1.81.83.12.96.1
Disasters 1.42.71.70.50.4
Schools 0.91.11.21.40.7
Obituaries 0.91.01.01.03.5
Health 1.10.90.91.10.4
Accidents 0.81.50.90.3·
Work 0.60.70.70.50.7
Police and justice 0.40.70.70.4·

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Regime and politics

Regime and politics: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
State visits 1.82.91.41.10.5
Franco 2.51.91.90.9·
Falange and the Movement 1.91.10.60.4·
Official events 0.90.60.90.50.4
Royal Household and Bourbons 0.10.10.31.41.7
Transition ···0.7·

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Armed forces

Armed forces: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Foreign armies 3.63.81.40.4·
Manoeuvres and weapons 1.41.20.80.71.8
Parades 1.81.00.80.71.7
Morocco and the colonies 0.30.50.2··
Spanish Civil War 0.30.20.20.10.6

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Religion

Religion: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Processions and feasts 2.82.91.91.3·
Hierarchy and the Vatican 0.71.61.10.3·
Churches and monasteries 0.30.30.30.40.8

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Science and technology

Science and technology: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Inventions 2.53.12.42.50.6
Nature 0.50.81.23.05.8
Space race 0.20.61.80.4·
Exploration 0.30.40.60.2·
Medicine and science 0.30.30.30.41.3

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Tourism and landscape

Tourism and landscape: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Landscapes 0.80.61.14.23.7
Tourism 0.30.71.71.70.6

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Oddities and humour

Oddities and humour: NO-DO newsreel. Share of each decade's news time taken by each sub-theme.
Sub-theme 1940s1950s1960s1970s1980–81
Oddities 1.22.12.01.10.8
Humour 0.30.50.30.5·

NO-DO newsreel. Share of each decade's news time taken by each sub-theme.

The longest item of each sub-theme

Discussion and limitations

The figures quantify, for the whole archive, a shift historians had described qualitatively: a newsreel born in the middle of a war as a window on the world, dependent on German and American footage, that gradually turned inward and towards entertainment, heritage and leisure, while sport stayed constant. The stability of Franco's presence and the small weight of explicit politics fit the newsreel as “banal nationalism” (Sánchez-Biosca, 2014): the regime is everywhere as frame and figure, and rarely as subject.

The language map shows a newsreel monolingual by design: Catalan, Basque and Galician are practically absent from both sound and captions. Methodologically, the study shows that a complete, two-level, timecoded thematic annotation of a 1,200-hour audiovisual archive is now possible at negligible cost, and that the main difficulty lies elsewhere: in the taxonomy and its rules, in aligning human summaries with noisy transcripts, and in validation.

Limitations

  • No human gold standard

    Agreement was measured between models of the same family with the same inputs; shared biases would go undetected. The external checks against manual coding are encouraging but partial.

  • Inputs, not images

    Classification relies on RTVE's summaries, the narration and the captions, not on visual analysis. The summaries are themselves a modern cataloguing layer that shapes the segmentation.

  • Taxonomy

    Every taxonomy embeds decisions, and the percentages depend on them (for instance, trade fairs under economy or crafts under folklore).

  • Timing and prints

    16.7% of items have estimated boundaries. The archive is not the newsreel as screened: some issues are missing and the 1946–1951 prints largely lack the Spanish narration.

  • Language identification

    It is unreliable on short fragments and between closely related languages, and songs, anthems and Latin chant can be mislabelled: findings about minority languages should be checked by listening to the fragment before citing them.

This is the first complete thematic and linguistic map of the NO-DO archive, item by item and with timecodes. Beyond the broad trends (from war to entertainment, sport as a constant, politics as frame rather than subject, a monolingual voice), the main contribution is an item-level map that links every aggregate figure to a specific item and second of RTVE's footage, and a validation protocol for annotating audiovisual archives with language models.

Data, code and citation

The audiovisual material is public on RTVE's website; each video is identified by its RTVE id. The derived data (segmentation and classification of the 39,957 items with timecodes, language segments and aggregate tables), the taxonomy, prompts and scripts are available from the author on reasonable request to review or replicate the study (xavi@sorensen.ai). The audiovisual material belongs to RTVE and the Filmoteca Española.

The preprint is published under a CC BY-NC-ND 4.0 licence. It has not been peer reviewed. The full text is in Spanish.

Large language models (Anthropic Claude Haiku 5.5 and Opus 5.5) were used as research instruments for segmentation, classification, review and validation, as described in the method.

How to cite

Vinaixa Roselló, X. (2026). De qué hablaba el NO-DO, y en qué lenguas. Análisis computacional del archivo completo de noticiarios y documentales del franquismo (1943-1981). Preprint. https://doi.org/10.13140/RG.2.2.29332.26243

References

  1. Vinaixa Roselló, X. (2026). De qué hablaba el NO-DO, y en qué lenguas. Análisis computacional del archivo completo de noticiarios y documentales del franquismo (1943-1981). Preprint. https://doi.org/10.13140/RG.2.2.29332.26243 — ResearchGate https://www.researchgate.net/publication/416587484_De_que_hablaba_el_NO-DO_y_en_que_lenguas_Analisis_computacional_del_archivo_completo_de_noticiarios_y_documentales_del_franquismo_1943-1981
  2. RTVE y Filmoteca Española. NO-DO: noticiarios y documentales (archivo en línea). — rtve.es https://www.rtve.es/filmoteca/no-do/
  3. Radford, A. et al. (2023). Robust Speech Recognition via Large-Scale Weak Supervision (Whisper). ICML. — arXiv https://arxiv.org/abs/2212.04356
  4. Bain, M. et al. (2023). WhisperX: Time-Accurate Speech Transcription of Long-Form Audio. Interspeech. — arXiv https://arxiv.org/abs/2303.00747
  5. Silero Team (2021). Silero VAD: pre-trained enterprise-grade Voice Activity Detector. — GitHub https://github.com/snakers4/silero-vad
  6. Arnold, T. & Tilton, L. (2019). Distant viewing: analyzing large visual corpora. Digital Scholarship in the Humanities, 34. — DSH https://doi.org/10.1093/llc/fqz013
  7. Törnberg, P. (2024). Large Language Models Outperform Expert Coders and Supervised Classifiers at Annotating Political Social Media Messages. Social Science Computer Review. — SSCR https://doi.org/10.1177/08944393241286471