Dutch Angle measures with data that others collected and shared. This is what we use, who made it and on what terms.
Subtitles. The dialogue comes from subtitles on OpenSubtitles.org, written by volunteers, as collected in the OpenSubtitles corpus of OPUS (version v2024). Source: P. Lison and J. Tiedemann (2016), OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles, LREC 2016. We do not share the subtitles themselves. The site only shows short quotes, 25 words at most, as evidence for an answer. Every film says which subtitle file we used.
Plot summaries. We judge the story questions from the plot summary on Wikipedia, usually the English one, sometimes the Dutch one. The quotes with those questions come from Wikipedia, sometimes shortened. Wikipedia text is licensed under CC BY-SA 4.0, and we share those quotes under the same licence. Every film page links to the article they come from, where you can also see who wrote it.
Cast and film data. Who is in the cast, whether a character is a woman or a man, dates of birth, director, running time, country and year come from Wikidata, free to use under CC0.
Screenplays. For some films we compare our share of women's words with Film Dialogue by The Pudding (Hanah Anderson and Matt Daniels, 2016; article, data). The data comes without a licence. We use it only as a check, with credit, and take just the percentage per film.
Bechdel test. The Bechdel results come from bechdeltest.com, where volunteers rate films. The list and the ratings are licensed under CC BY-NC 3.0: with credit, and not for commercial use. Dutch Angle does not use them commercially. The Bechdel test does not count in the angle.
Ideas. The writers, critics and researchers the questions build on are listed under How we measure. Their work is inspiration, not data.
Type and code. The typefaces are Anton, Inter and JetBrains Mono (SIL Open Font License, via Google Fonts). The balls fall with Matter.js (MIT licence).
Missing something? Is there a source that belongs here, or a credit that is wrong? Write to info@dutchangle.org.
", "overx": "Films teach us who gets to speak, to want things and to decide. Dutch Angle measures how a film shares that room between women and men.
Who. Dutch Angle is made by Astrid Bussink, documentary maker and film tutor. It is a project of The Missing Measure.
What you can do here. The archive holds {n} well-known films, each at the angle its story stands at. Open a film for the questions, the answers and the quotes they rest on. You can also measure your own film; in this design that is still a demo.
Which films. Well-known films from different decades: American and Dutch films and world cinema. They were chosen to test the method on films many people know, not as a ranking of the worst films.
What the angle does and does not say. The angle measures room in the story: who speaks, whose story it is, who decides and how the story treats them. It says nothing about whether a film is good. We measure gender only, and we never guess anyone's identity. We read the dialogue and the plot summary, not yet the image. How we measure exactly is under How we measure.
Independent. No film company, broadcaster or streamer pays for it or has a say in the results. We do not accept money from parties whose films we measure. Funding: own resources.
Something wrong? Did you make one of these films, or do you think we measured something wrong? Write to info@dutchangle.org, ideally with the scene or the timecode. We look at every report and correct what is wrong.
What is coming. Really measuring your own film, more films in the archive, and a measurement of the image next to that of the dialogue.
Privacy. Your film stays on your own computer. Only the transcribed text goes to a language model (AI). A full privacy statement will follow.
", "nav_back": "← Back", "h1": "How tilted is your film?", "lede": "Who gets to speak, to want things and to decide? Dutch Angle measures which way a film tilts.", "namex": "A dutch angle is a tilted camera shot. We measure how tilted the story itself is.", "drop": "Drop your film here", "dropsub": "Your film stays on your computer · mp4 or mov", "s1h": "Drop in your film", "s1d": "It stays on your own computer.", "s2h": "The balance is measured", "s3h": "Check your film's Dutch Angle", "s3d": "From −45° (all her way) through 0° (balanced) to 45° (all his way).", "samples": "Find a film in the archive", "ownh": "Or measure your own film", "missH": "We can't measure this film yet", "missDemo": "See an example", "hitH": "This film is already in the archive", "hitView": "View the result", "hitAgain": "Measure again", "searchPh": "Title, year or director", "searchMeasure": "Measure it yourself", "sortTilt": "Most his way", "sortLevel": "Level", "sortHer": "Most her way", "grpAll": "All", "grpWhere": "Where", "grpWhen": "When", "meterH": "Archive average", "meterHs": "Average", "meterTip": "Point at a ball for the title; click to open the film.", "showLess": "Show less", "demo_run": "Note: this is a design. Your film is not really measured yet; next you will see an example.", "runh": "Reading your film", "fast": "Sped up. In reality it takes about 25 minutes.", "skip": "Skip", "heard": "Just heard", "film": "Film", "again": "← Archive", "provisional": "provisional", "chipHand": "Scored by hand", "chipErr": "No margin of error yet", "again2": "Drop again", "wiphint": "Tap a ball for the question and the evidence.", "resHow": "Each ball is one question, such as who talks, who works or who gets rescued. It falls towards the side that comes off better: the clearer, the further out. The colour is the subject. Then the picture tilts with the weight.", "wiplevel": "This film is in balance. Tap a shape to see what was measured.", "sideW": "the women", "sideM": "the men", "wipnone": "Too few answers for an angle.", "sound_on": "Sound: on", "sound_off": "Sound: off", "women": "WOMEN", "men": "MEN", "notyet": "Not measured yet", "share": "Share the result", "copied": "Link copied", "shareH": "Your share card", "shareNative": "Share…", "shareDl": "Download the card", "shareCopy": "Copy the link", "shareBy": "Measured with Dutch Angle", "shareNote": "In this design the link points to the prototype. On the real site every result gets a page of its own, and the card also shows up in your messaging app as the link preview.", "more": "All questions and answers", "moreLink": "All questions as a list ↓", "conclh": "What this means", "notquality": "The angle says nothing about whether a film is good. It shows how the room is shared.", "erah": "Compared with its time:", "era": "coming after calibration. You will then see how this film compares with other films from the same years.", "pats": "Open a subject. For every question you see the answer, how it was measured and the words from the film it rests on.", "notmeasured": "not measured yet", "redflag": "Red flag: more than 5 years apart", "report": "Is this wrong?", "reportDone": "Thank you. In this design your report does not go anywhere yet. On the real site you can report a mistake here or respond as a filmmaker.", "foot_by": "Dutch Angle is a project by The Missing Measure.", "foot_privacy": "Privacy: your film stays on your own computer. Only the transcribed text goes to a language model (AI). A full privacy statement will follow.", "foot_more": "Contact: info@dutchangle.org · The measurement handbook will follow.", "support": "Support Dutch Angle", "supportResult": "Found this useful? Every measurement costs computing power.", "supportNote": "Thank you. In this design the button does not go anywhere yet; on the real site you can support Dutch Angle here with a small amount.", "foot_indep": "Dutch Angle is independent. No film company, broadcaster or streamer pays for it or has a say in the results. We do not accept money from parties whose films we measure.", "foot_fund": "Funding: own resources. We will update this as soon as there are donations or grants.", "method": "Method", "from": "inspired by ", "askh": "Did you see yourself in this film?", "asks": "We never guess colour, age, body or disability. Whether you recognise yourself is something we would rather ask you.", "a1": "Yes", "a2": "A little", "a3": "No", "askh2": "Were the women mostly looked at, or did they do the looking?", "b1": "Mostly looked at", "b2": "Both", "b3": "They did the looking", "askdone": "Thank you. This design does not store your answer anywhere; on the real site it will count.", "abouth": "How we measure", "treeh": "What we build on", "trees": "Dutch Angle asks its own questions, but we stand on the shoulders of writers, critics and researchers. These are our sources of inspiration.", "proto": "Design version · measuring your own film is still a demo", "foot_version": "Design version · question set v0.5 · {n} films measured on the subtitles and the Wikipedia plot", "archNote": "The archive holds {n} films, all measured on the subtitles of the complete film and the Wikipedia plot summary; every measurement with the tool adds one. In the real version, search is helped by Wikidata and you can filter by year, country, language, genre and angle.", "chipScript": "Measured on the subtitles and the Wikipedia plot", "subjects8": "8 dimensions · 16 questions", "path": "The eight dimensions", "s2d": "16 questions about who is there, whose story it is, who decides and how the story treats them.", "aboutx": "Dutch Angle measures how a film shares room between women and men, in eight dimensions that count equally: PRESENCE (who speaks and who is central), STORY (whose story it is, who has a goal of their own, who is essential), POWER (who decides, who has the money), VOICE (do women talk with each other, and about what), IDENTITY, WORK, AGE and FATE.
Every question shows its source. Counts come from the subtitles and the cast (Wikidata); the story questions are judged from the Wikipedia plot summary twice and separately, by Claude, a language model; the second time without the first result. Where that has happened only once, it says so. A check by people is still to come. Gender always comes from the cast, never from a face or a voice. The plot summaries are written by Wikipedia editors and can be one-sided themselves; that is why every judgement shows the sentence it rests on.
Dutch Angle measures gender only. We know that is not the whole story: women of colour, older women, queer and disabled women often disappear even further from the screen, and that intersection matters. We do not guess skin colour, sexuality or any other identity, for the same reason we never guess gender. Once there is reliable, public data on it, we will include it.
Per dimension we take the average of its questions, from −1 (all her way) to 1 (all his way). The angle is 45 × the average of the dimensions. A film gets an angle from {MIN_DIM_HOEK_W} dimensions and a verdict from {MIN_DIM_OORDEEL_W}, provided the speaker is known for at least {PCT_WD}% of the words; otherwise it is provisional. Every angle comes with a margin.
{VAL}The questions build on Woolf, Gerbner, Tuchman, Mulvey, the Bechdel, Mako Mori and Sexy Lamp tests, Lauzen, the Geena Davis Institute, Anderson and Daniels, Kagan et al. and Sap et al. This is a pilot version; the rationale is in the measurement handbook.
", "grp": { "nederland": "Netherlands", "europa": "Europe", "vs": "United States", "azie": "Asia", "latijns-amerika": "Latin America", "afrika": "Africa", "oceanie": "Oceania", "voor1970": "Before 1970", "1970-1999": "1970–1999", "2000-2019": "2000–2019", "sinds2020": "Since 2020" }, "steps": [ [ "Taking the sound from the film", "In small pieces, so your computer keeps running.", "2 min" ], [ "Transcribing the dialogue", "A speech model writes it down, on your own computer.", "21 min" ], [ "Matching speakers", "Every line gets a character. Whether that is a woman or a man comes from the cast list, never from the voice.", "3 min" ], [ "Answering the questions", "A language model reads only the text and backs every answer with quotes.", "2 min" ] ], "loading": "Loading the archive…", "loadError": "The archive could not be loaded. Please try again later.", "missTxt": "Dutch Angle is still a design: {name} is not really measured yet. Search below to see whether the film is already in the archive, or see what a measurement looks like with an example (Pretty Woman).", "hitTxt": "{film} has already been measured: {res}. View the result, or measure it again. Several measurements of the same film together give a margin of error.", "searchNone": "\"{q}\" is not in the archive yet.", "meterN": { "one": "{k} film", "other": "{k} films" }, "meterHeadMany": "In {him} of {total} films women get less room. The other way round: {her}.", "meterHeadSplit": "{him} of {total} films lean towards the men, {her} towards the women.", "meterTxt": "Each ball is a film, at its angle: right of zero the men get more room, left of zero the women. The light strip around zero is in balance, up to {BAND1}° either way.{prov} Together the archive tilts {h}.{none} These are the films we chose, not all films; the picture moves with every new measurement.", "meterTxtProv": " See-through balls are provisional: measured in fewer than {MIN_DIM_OORDEEL} of the 8 dimensions, or the speaker is known for less than {PCT_WD}% of the words.", "meterTxtNone": " {k} film without an angle (too little data) is left out.", "meterSides": [ "her way", "in balance", "his way" ], "meterDirHim": "his way", "meterDirHer": "her way", "meterDirLevel": "level", "margeLine": "Margin: ± {m}°. With these questions the angle most likely lies between {lo} and {hi}. {tail}", "margeHim": "Even at the low end, this film leans his way.", "margeHer": "Even at the high end, this film leans her way.", "margeOpen": "The margin crosses zero: the direction is not certain yet.", "castLine": "Who speaks: {women} and {men} ({pct}% women).{tail}", "castWomen": { "one": "{k} named woman", "other": "{k} named women" }, "castMen": { "one": "{k} named man", "other": "{k} named men" }, "castFewW": " With so few women in the cast, almost every question tips towards him.", "castFewM": " With so few men in the cast, almost every question tips towards her.", "showAll": "Show all {k}", "demo_res": "You dropped in \"{file}\". This design does not measure real films yet: what you see here is the example of {title}.", "mostTilted": "Most tilted: {list}.", "chipN": "{a} of {b} questions answered", "chipMarge": "Margin ± {m}°", "chipS": "spread over {k} of the 8 dimensions", "resHead": "In {him} of {total} questions the men come off better, {her}{level}", "resHeadHer": "in {k} the women", "resHeadHerNone": "in none the women", "resHeadLevel": { "one": "; {k} is level.", "other": "; {k} are level." }, "resHeadEnd": ".", "wiplight": "This film leans slightly towards: {side}, but stays on its feet. Tap a shape to see what was measured.", "wipslide": "This film is out of balance and tips over towards: {side}. Tap a shape to see what was measured.", "measured": "{a} of {b} questions answered", "nq": { "one": "{k} question", "other": "{k} questions" }, "and": "and", "provName": "Provisional", "provSd": "Not tested enough yet: we know which character speaks for only {sd}% of the subtitle time, and a verdict needs at least {PCT_SD}%. When a better source turns up, we measure again.", "provWd": "For {wd}% of the words the speaker is known (songs do not count). A verdict needs at least {PCT_WD}%.", "provDims": "Measured in {dims} of the 8 dimensions. That is too few for a verdict.", "complexName": "Level, but…", "complexTxt": "Level, but not because everything is equal: {up} pulls {u} his way, {dn} {d} her way. They cancel each other out.", "conclMain": "{title} tilts {h}, based on {k} of the {N_MEAS} questions, measured on the subtitles of the complete film: counts over all conversations and judgments from the dialogue, with quotes. Not scored by hand.", "conclTop": "The heaviest subjects are {list}.", "conclEnAll": "Measured on the English subtitles; there were no usable {taal} subtitles. Every question rests on the translation.", "conclEnI2": "Measured in {taal}; only the question on names and roles (I2) was counted on the English subtitles.", "hardSameHim": "{s} leans his way.", "hardSameHer": "{s} leans her way.", "hardToo": "{s} too.", "hardLevel": "{s} is level.", "hardComplex": "{up} his way. {dn} her way. Level together.", "valHead": "How we check it", "valPlot": "Two judgements. For {films} films the story questions were judged twice and separately by Claude, a language model; the second time without the first result. For {pct}% of the {q} answers the two results differ by at most a quarter. A check by people is still to come.", "valPud": "Screenplays. For {films} films we compare our share of women's words with the screenplays counted by The Pudding (Anderson and Daniels, Film Dialogue, 2016). The correlation is {r} and the difference is {gem} percentage points on average; for {bin} films it is under 10 points.", "valSdh": "Speaker names. Some subtitles for the deaf and hard of hearing say who speaks a line. For {films} films we found one with enough names, and there we compare every named line with ours ({regels} lines). Before this check we had the same character for {p}% and a speaker of the same gender for {g}%. Where the label clearly said otherwise, we changed it: {fix} lines corrected and {vul} lines without a speaker filled in, in {ff} films. The gender comes from the cast list, or for a label such as MAN or WOMAN from that label. For the other {geen} films there is no such subtitle, or it names too few speakers.", "valBech": "Bechdel test. For {k} films there is a verdict on bechdeltest.com. Films that fail lean {d} further towards him on average than films that pass. The Bechdel test does not count in the angle: it is an outside check.", "valProv": "Provisional. A film where we know which character speaks for less than {PCT_SD}% of the subtitle time, or whether a woman or a man speaks for less than {PCT_WD}% of the words (songs do not count), or that is measured in fewer than {MIN_DIM_OORDEEL_W} dimensions, is listed as provisional ({k} films now). When a better source turns up, we measure it again. Every film page says under 'How this film was tested' what was checked for that film.", "testedH": "How this film was tested", "testedMore": "How we check the whole archive is under How we measure.", "tSubFile": " (OpenSubtitles, file {id})", "tSub": "Subtitle{file}, with a time for every line.", "tSd": "Characters: the lines were matched to a character by Claude, a language model, where possible. For {pct} of the subtitle time the speaker is known{low}.", "tSdLow": ". That is too little: at least {PCT_SD}% is needed", "tWd": "Woman or man: for {pct} of the words we know whether a woman or a man says them (from the cast list on Wikidata; songs do not count){low}.", "tWdLow": ". A verdict needs {PCT_WD}%", "tShDone": " Where the label clearly said otherwise, the line was changed automatically: {list}.", "tShFixed": { "one": "{k} line corrected", "other": "{k} lines corrected" }, "tShFilled": { "one": "{k} line without a speaker filled in", "other": "{k} lines without a speaker filled in" }, "tShNothing": " Nothing needed changing.", "tShX": "Speaker names: there is a subtitle with speaker names, but its names often did not fit the scene. Only lines where the content confirms the label were changed, by hand.", "tShOk": "Speaker names: checked against a subtitle for the deaf and hard of hearing, which says who speaks for {lines} lines{time}. {res}", "tShTime": " (an English subtitle, lined up with ours by time)", "tShSame": "We had the same character for {p}% of them and a speaker of the same gender for {g}%.", "tShGender": "For {g}% of them we had a speaker of the same gender.", "tShFew": "Speaker names: there is a subtitle with speaker names, but it names a character for only {lines} of our lines: too few to compare.", "tShNone": "Speaker names: no subtitle found that names enough of the speakers.", "tPr2": "Story questions: judged twice and separately by Claude, a language model. For {a} of the {b} questions the two results differ by at most a quarter. A check by people is still to come.", "tPr1": "Story questions: judged once by Claude, a language model; the second, separate judgement is still to come.", "tPr0": "Story questions: no plot summary found, so not judged.", "tPlotSrc": "Plot: the summary from {link} (CC BY-SA 4.0). The quotes with the story questions come from it.", "tPud": "Compared with the screenplay (The Pudding, Film Dialogue): we count {ons} of the words for women, the screenplay {zij}.{gap}", "tPudGap": " That gap is large; the screenplay may be a different version from the film.", "tPudNone": "Compared with the screenplay: this screenplay is not in The Pudding's dataset.", "tBdNone": "Bechdel test: this film is not on bechdeltest.com.", "tBd": "Bechdel test (bechdeltest.com): {res}. This is an outside check; it does not count in the angle.", "tBdPass": "passes", "tBdFail": "fails ({k} of 3 steps)", "shareQ": "questions", "shareCast": "cast {pct}% women", "getal": [ "zero", "one", "two", "three", "four", "five", "six", "seven", "eight" ], "stands": [ [ "Heavily tilted towards her", "The story gives almost all the room to the women; the men get little to say, want or decide.", "Heavily tilted", "towards her" ], [ "Clearly tilted towards her", "The story clearly gives the women more room than the men.", "Clearly tilted", "towards her" ], [ "Slightly tilted towards her", "Both get room, but the seesaw leans towards the women.", "Slightly tilted", "towards her" ], [ "Level", "Women and men get about the same room.", "Level", "" ], [ "Slightly tilted towards him", "Both get room, but the seesaw leans towards the men.", "Slightly tilted", "towards him" ], [ "Clearly tilted towards him", "The story clearly gives the men more room than the women.", "Clearly tilted", "towards him" ], [ "Heavily tilted towards him", "The story gives almost all the room to the men; the women get little to say, want or decide.", "Heavily tilted", "towards him" ] ], "filmHard": { "pw": "He decides. She gets rescued. The money is his.", "ti": "She tells it. She chooses. But work is for the men." }, "tree": [ [ "1929", "Virginia Woolf", "A Room of One's Own: women in books almost only exist in relation to men.", "https://en.wikipedia.org/wiki/A_Room_of_One%27s_Own" ], [ "1975", "Laura Mulvey", "The male gaze: the woman as image, the man as the one who looks.", "https://en.wikipedia.org/wiki/Visual_Pleasure_and_Narrative_Cinema" ], [ "1976", "George Gerbner, Larry Gross", "Being absent from the screen is symbolic annihilation.", "https://en.wikipedia.org/wiki/Symbolic_annihilation" ], [ "1985", "Alison Bechdel, Liz Wallace", "The Bechdel test: two women talk to each other about something other than a man.", "https://en.wikipedia.org/wiki/Bechdel_test" ], [ "1991", "Katha Pollitt", "The Smurfette principle: one woman in a group of men.", "https://en.wikipedia.org/wiki/Smurfette_principle" ], [ "1999", "Gail Simone", "Women in refrigerators: hurt to motivate a man.", "https://en.wikipedia.org/wiki/Women_in_Refrigerators" ], [ "2013", "Mako Mori-test", "Does she have a storyline of her own?", "https://en.wikipedia.org/wiki/Mako_Mori_test" ], [ "2016", "Manohla Dargis", "The DuVernay test: do characters of colour have lives of their own?", "https://en.wikipedia.org/wiki/Bechdel_test#People_of_color" ], [ "2017", "Sap et al.", "Power and agency in the verbs of a screenplay.", "https://aclanthology.org/D17-1247/" ], [ "2002–", "Martha Lauzen", "It's a Man's (Celluloid) World: every year, who speaks and who leads in the top US films.", "https://womenintvfilm.sdsu.edu/" ], [ "2004–", "Geena Davis Institute", "Research into who is on screen and for how long, with the GD-IQ.", "https://geenadavisinstitute.org/" ], [ "2007–", "USC Annenberg Inclusion Initiative", "Counts of who speaks in the most popular films.", "https://aii.annenberg.usc.edu/" ], [ "2025", "Willemien Sanders, Sandra Becker", "'Gewoon' een Nederlandse vrouw: how women are portrayed in Dutch films and series (Utrecht University, Vrouwen in Beeld).", "https://www.vrouweninbeeld.nl/onderzoek/" ] ], "dims": { "presence": { "name": "PRESENCE", "q": "Who is there, and who gets the words?" }, "story": { "name": "STORY", "q": "Whose story is it?" }, "power": { "name": "POWER", "q": "Who decides and who has the money?" }, "talk": { "name": "VOICE", "q": "Who talks with whom, and about what?" }, "identity": { "name": "IDENTITY", "q": "Who is someone in their own right?" }, "work": { "name": "WORK", "q": "Whose work matters?" }, "age": { "name": "AGE", "q": "Who may grow older?" }, "fate": { "name": "FATE", "q": "How does the story treat them?" } }, "questions": { "P2": { "q": "What share of the words do women speak?", "kort": "words", "insp": "Anderson and Daniels (2016)" }, "P3": { "q": "Who is central in who talks with whom?", "kort": "central", "insp": "Kagan, Chesney and Fire (2020)" }, "S1": { "q": "Who comes first: in the credits and in the plot?", "kort": "top billing", "insp": "Lauzen, protagonists" }, "S2": { "q": "Who has a goal of their own?", "kort": "own goal", "insp": "Mako Mori test; Lauzen, goals" }, "S3": { "q": "Who is essential to the story?", "kort": "essential", "insp": "Sexy Lamp test (DeConnick)" }, "M1": { "q": "Who makes the decisions at the turning points?", "kort": "turning points", "insp": "Lauzen, protagonists; Sap et al. (2017), power" }, "A5": { "q": "Who has the money, and who depends on it?", "kort": "money" }, "St2": { "q": "Do women talk with each other as often as men do?", "kort": "with each other", "insp": "Bechdel test, made continuous" }, "L1": { "q": "Who talks about the other more often?", "kort": "talking about the other", "insp": "the Bechdel-Wallace test (1985)" }, "I2": { "q": "Are people called by their name, or by a role or pet name (\"honey\", \"girl\", \"son\", \"sir\")?", "kort": "called by name" }, "I3": { "q": "Whose relationship status comes up more often?", "kort": "relationship status", "insp": "Lauzen, It's a Man's (Celluloid) World" }, "W1": { "q": "Whose work matters to the story?", "kort": "work", "insp": "Lauzen, occupations" }, "J2": { "q": "Who talks about work more often?", "kort": "talking about work" }, "G1": { "q": "Who is older in the couple? More than 5 years apart gets a red flag. (The actors' ages at release.)", "kort": "age gap in the couple", "insp": "Stephen Follows on age gaps" }, "G3": { "q": "How old are the women and men who speak the most?", "kort": "age of the leads", "insp": "Lauzen, age" }, "F1": { "q": "Whose suffering is used for the other's story?", "kort": "suffering used for the other's story", "insp": "women in refrigerators, Simone (1999)" } }, "types": { "count": "count", "claude": "language-model judgment", "cast": "cast data", "plot": "Judgement from the plot summary (Wikipedia)" }, "status": { "hand": "Scored by hand · automatic measurement to follow", "cast": "From Wikidata", "counted": "Counted in the subtitles", "claude": "Judged from the dialogue, with quotes", "pudding": "Actors' ages from the Pudding dataset", "plot": "From the plot summary, judged twice and separately by Claude", "plot1": "From the plot summary, judged once by Claude; the second judgement is still to come" }, "tags": { "level": "balanced", "him": [ "slightly tilted towards the men", "tilted towards the men", "strongly tilted towards the men" ], "her": [ "slightly tilted towards the women", "tilted towards the women", "strongly tilted towards the women" ] } }