id string | audio audio | instruction string | question string | answer string | script string | n_speakers string | data_source string |
|---|---|---|---|---|---|---|---|
task2_47 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who expressed fear about telling their parents the same one who reassured that the baby will be loved even if not married? | True | spk1: One pregnant woman at a time, please! I just want you to be okay.
spk2: So forcing her to marry Bobby is gonna make that happen?
spk1: Maybe! Well! Well
so-so uh, what kind of music does Numb NutsΒOh forget it! I canΒt!
spk3: Joey, I am scared to death about this.
spk3: But I really think I can do it, IΒm just g... | 4 | MELD | |
task2_18 | Read the conversation and generate an appropriate answer to the question. | Is the participant who expressed nervousness about their new year challenge the same one who is working on a website for their cakes? | True | <spk2>: Oh, but it's so wet already from
<spk4>: Do you have another towel?
<spk2>: Yeah. Oh, my gosh, why didn't I think of this?
<spk3>: Do you have another one of those?
<spk4>: Oh.
<spk3>: Or do you have, like
<spk3>: That's okay, I guess you can just let it dry.
<spk1>: Yeah, I can find something.
<spk3>: We can g... | 4 | CHIME-6 | |
task2_7 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned being in 'short supply' of money the same one who decided to terraform despite lacking power? | True | <spk3>: that one. Oh shit.
<spk4>: Okay. Um I get to pick.
<spk1>: Terraforming.
<spk2>: Terraform us.
<spk4>: Um.
<spk4>: Terraforming, hey?
<spk4>: Okay.
<spk4>: I dunno. Sure, I'll get this one.
<spk1>: Oh shoot. Oh.
<spk4>: Jeff?
<spk3>: Um, money? Cult.
<spk1>: Um.
<spk3>: Workers. I'll take money.
<spk3>: I'm in ... | 4 | CHIME-6 | |
task2_67 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned liking the Kin Da restaurant the same one who trusts Bill Murray's opinion on comedies? | False | <spk1>: Hello, do you like comedy?
<spk2>: I love comedy
<spk1>: Then we are alike, comedy has interesting roots all the way back to ancient Greece
<spk2>: Yeah that's cool. I didn't know that the word comedy derives from a Greek word
<spk1>: Yeah it refers to a discourse that is humorous or amusing that generates la... | 2 | MultiDialog | |
task2_89 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who discussed the importance of distinguishing between knowledge and opinion in group processes the same one who suggested segmenting meetings into interactive and mono-logic areas? | True | <spk2>: been looking
<spk3>: game
<spk2>: at the data . So that's
<spk3>: . He's been looking
<spk2>: that's
<spk3>: at the
<spk2>: a
<spk3>: data
<spk2>: step for
<spk3>: , okay
<spk2>: this
<spk1>: Mm-hmm
<spk2>: .
<spk3>: , s to
<spk1>: .
<spk3>: good . Um and then and then there are lots of different theories of ar... | 3 | AMI | |
task2_86 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who expressed uncertainty about the user interface design the same one who provided detailed instructions on saving the project file? | False | <spk1>: Um
<spk2>: Okay . Yeah
<spk1>: .
<spk2>: .
<spk2>: Right
<spk1>: That's it
<spk2>: . Um
<spk1>: .
<spk2>: , if we could hear from our Industrial engineer , or
<spk4>: Yeah
<spk2>: Designer
<spk4>: .
<spk2>: .
<spk4>: Uh , I was still working on stuff , I hadn't got it finished . Um , alright .
<spk4>: Click to ... | 4 | AMI | |
task2_62 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned Taylor Swift topping the Canadian iTunes charts with 8 seconds of white noise the same one who expressed amazement at the Ramones playing 2,263 concerts? | True | <spk1>: Are you into music? Did you know that the dark side of the moon was still one of the best selling albums of 2014 even though it was released 40 years ago.
<spk2>: I heard about it but I have never listened to it.
<spk1>: Interestingly 2 identical twins from japan released a rap album. funny thing was that the... | 2 | MultiDialog | |
task2_35 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who expressed a 'thing for the Days Of Our Life's people' the same one who sent others to a nighttime tour of a button factory? | True | spk1: Hi Joey it's Jane Rogers, can't wait for your party tonight.
spk1: Listen, I forgot your address, can you give me a call?
spk1: Thanks, bye.
spk2: Hey!
spk3: Hey!
spk2: What's happening?
spk3: Yeah, it's a real shame you can't make it to that one-woman show tonight.
spk2: Oh, I'd love to, but I gotta get up... | 4 | MELD | |
task2_30 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who felt like they had 10 drinks after meeting someone the same one who expressed concern about never having a first kiss again? | False | spk1: Oh, great!
spk1: We couldnΒt keep our eyes off each other all night and then every once and a while yΒknow, heΒd kinda lean over and stroke my hair and touch my neck.
spk2: Okay, stop it Phoebe, youΒre getting me all tingly.
spk1: All I could think of was yΒknow, "Is he gonna kiss me? Is he gonna kiss me?"
spk2: ... | 2 | MELD | |
task2_73 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned that Netflix spends 20 times more on postage than bandwidth the same one who noted that a fishing community in Indonesia keeps sharks as pets? | False | <spk1>: How's it going? Do you watch Netflix? They have 137 million total subscribers worldwide.
<spk2>: Yes, I am one of those subscribers.I love no commercials the most. I would not mind working for Netflix someday either.
<spk1>: That would be a good place to work. It's crazy from 9pm to 12am in North America 33% o... | 2 | MultiDialog | |
task2_1 | Read the conversation and generate an appropriate answer to the question. | Is the participant who expressed a preference for the 'tower version' of the game the same one who corrected the description of a 'balloon' to 'balloon animal'? | True | <spk3>: Panda.
<spk4>: What?
<spk1>: Bone.
<spk2>: Oh.
<spk1>: Sloth.
<spk3>: Donut.
<spk1>: Aw.
<spk2>: Aw.
<spk4>: Boots.
<spk1>: Dog.
<spk4>: Dog.
<spk2>: Dog.
<spk4>: Damn.
<spk2>: Oh my god.
<spk3>: Sunglasses.
<spk2>: Unicorn, oh go go ahead.
<spk4>: laughs
<spk4>: Sundae.
<spk1>: Aw that's what I was gonna say.
... | 4 | CHIME-6 | |
task2_42 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who mentioned the sidecar also the one who thought about the planetarium proposal while they were going out? | True | spk1: and then, we couldΒve gone from the ceremony to the reception with you in the sidecar!
spk2: Ross, it just wouldnΒt have been feasible.
spk1: But having a dove place the ring on your finger wouldΒve been no problem?
spk2: It was really fun being married to you tonight.
spk1: Yeah! And! And, it was the easiest 400... | 2 | MELD | |
task2_31 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who admitted to feeling 'all floopy' the same one who came to fill the ice cube trays? | False | spk1: IΒm sorry.
spk2: What are you doing here?!
spk1: I-I, came to fill your ice cube trays.
spk2: What?!
spk1: Umm, okay, okay, look.
spk1: I took this picture from your fridge.
spk1: Okay, because I know that this is my Father.
spk1: Yeah, this is Frank Buffay and you are standing right there next to him.
spk1: Now,... | 2 | MELD | |
task2_71 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned that baseball is really popular in Japan the same one who asked if catching a ball in a player's hat gives the other team three bases? | True | <spk1>: The Braves sure are a pupular team, they seem to be well rested for the playoffs
<spk2>: Is that good or bad. They have had no real good teams to play against.
<spk1>: I think its a bad thing, practice is everything and not playing enough games can result in lack of chemistry
<spk2>: Thursday at 8:07 am easter... | 2 | MultiDialog | |
task2_55 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned the Indonesian president releasing pop albums the same one who talked about the 'immortal' jellyfish? | True | <spk1>: Hi! Can you believe Trump isn't the wealthiest president of all time. In fact, he isn't even top three? That title belongs to Washington, Jefferson and JFK!
<spk2>: I guess it makes sense if you adjust for inflation. Woodrow Wilson was the only President with a PhD
<spk1>: Was he really? Of all the seemingly in... | 2 | MultiDialog | |
task2_57 | Read the conversation and generate an appropriate answer to the question. | Is the participant who expressed interest in seeing Apple's 80's clothing line the same one who mentioned owning tailored business attire and suits? | False | <spk1>: hey did you know that Pilgrims actually wore colorful cloths?
<spk2>: That's surprising! I always imagined them in the traditional black and white clothing they are usually depicted in.
<spk1>: I know right! that black outfit was actually what they wore only fore special occasions
<spk2>: That's mentally change... | 2 | MultiDialog | |
task2_94 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who suggested focusing on the menu as the best interface the same one who expressed concern about the product's performance for their criteria? | True | <spk2>: So ,
<spk2>: I would say that it would seem like the general opinion is we're gonna keep the L_C_ display 'cause it's about what really separates us
<spk3>: I
<spk2>: , despite
<spk3>: think so
<spk2>: the
<spk3>: .
<spk2>: cost it's gonna incur . Um
<spk2>: are people maybe not happy with , but are willing to ... | 4 | AMI | |
task2_92 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who mentioned using DICE machines the same one who suggested cold-calling Marloes? | False | <spk1>: Yeah
<spk3>: builder
<spk1>: , yeah
<spk3>: to the actual
<spk1>: . screen shots . Yeah
<spk3>: yeah
<spk1>: , yeah .
<spk3>: . Um that nothing has I haven't heard anything since from either sort of the Netherlands or from you .
<spk2>: Okay
<spk1>: Oh
<spk2>: , so
<spk1>: .
<spk2>: I've got something on the tw... | 3 | AMI | |
task2_3 | Read the conversation and generate an appropriate answer to the question. | Is the participant who expressed skepticism about the sugar-free gummies also the one who gave advice on fitting into a wedding dress? | True | <spk1>: Would also be very expensive.
<spk2>: A close second as well, so.
<spk1>: Yeah, but this was this was a very pricey time to go.
<spk3>: I was like, what the heck is this going to be?
<spk4>: Six hundred something?
<spk3>: Seven hundred dollars a night.
<spk1>: Wow, oh, my God.
<spk3>: I was like, that's somebod... | 4 | CHIME-6 | |
task2_78 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who considered deleting their changes and copying from another's work the same one who proposed different result panes for topics, summaries, and transcripts? | True | <spk1>: Okay
<spk2>: and just
<spk1>: .
<spk2>: it's called start time and end time . And
<spk1>: Okay .
<spk2>: that will tell you the start times and end times
<spk1>: Okay .
<spk2>: . I think it returns it double .
<spk1>: Okay .
<spk1>: Yeah . So now I could just put it in some
<spk2>: Yeah , that's
<spk1>: Tuple
<... | 4 | AMI | |
task2_21 | Read the conversation and generate an appropriate answer to the question. | Is the participant who talked about needing to text Melissa the same one who expressed a favorite role in the game 'Bang'? | False | <spk4>: We still could, but we were, like, "No, just next time."
<spk1>: Next time.
<spk4>: Next time I'll do it.
<spk2>: Definitely, I'll do it. I'm down.
<spk2>: We need to clean that table a little bit.
<spk4>: Mhm.
<spk3>: Shan, wanna just, just put this in the oven or something just to get space?
<spk1>: Do I wann... | 4 | CHIME-6 | |
task2_32 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who is going to London to tell Ross they love him the same one who insists that Ross loves Emily? | False | spk1: Do you remember where the duck food is?
spk2: Yeah, itΒs in the guysΒ apartment under the sink. Why?
spk1: Because IΒm going to London.
spk2: What?! What do you mean youΒre going to London?
spk1: Yeah, I have to tell Ross that I love him.
spk1: Now honey, you take care, you donΒt have those babies until I get bac... | 2 | MELD | |
task2_15 | Read the conversation and generate an appropriate answer to the question. | Is the participant who questions owning clothes that can't be machine washed the same one who prefers to hang-dry eighty percent of their clothing? | False | <spk1>: cuz Kara will, will tackle them as they come up and so I feel like
<spk4>: Susie.
<spk1>: she does way more than I do because I don't think about it at the time.
<spk4>: On your pillow, go.
<spk4>: Go. Hey.
<spk2>: Yeah.
<spk1>: But when, when I do think of it, when the dishes do pile up to the point when I'm n... | 4 | CHIME-6 | |
task2_13 | Read the conversation and generate an appropriate answer to the question. | Is the participant who discussed the enzyme responsible for cilantro tasting like soap the same one who mentioned buying lactose-free sour cream? | True | <spk1>: Oh wait, we don't always have to be here, all right, I'm gonna walk I'm gonna walk outside.
<spk1>: Ooh.
<spk2>: Done?
<spk2>: Uh, I guess I'll use this knife.
<spk1>: Mmm, I love onions.
<spk1>: Is there onion song?
<spk2>: Yeah, by The Muppets.
<spk2>: It's the best song ever.
<spk2>: Yeah, it's so cheeky.
<s... | 4 | CHIME-6 | |
task2_12 | Read the conversation and generate an appropriate answer to the question. | Is the participant who explained the scoring system for dumplings the same one who asked about the rules for using chopsticks? | False | <spk3>: Like a duplicate.
<spk2>: Yeah.
<spk4>: Yeah.
<spk3>: So you can grab both at the same time.
<spk4>: Okay.
<spk1>: Come on! Gimme some good stuff!
<spk4>: Okay, okay, okay.
<spk1>: Gimme some good stuff!
<spk1>: Nigiri.
<spk2>: Chopsticks.
<spk1>: Ah! You took my dumpling!
<spk2>: Mhm.
<spk4>: Yeah?
<spk2>: Oh,... | 4 | CHIME-6 | |
task2_69 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned that the Library of Alexandria saved a copy of the internet the same one who talked about the hamburger button on websites? | True | <spk1>: Do you like to watch TV?
<spk2>: Hi! I love TV! In fact, The Simpson's is my favorite show. However, there's an episode where Bart never makes an appearance nor is even mentioned that I still want to see. I bet it's weird never seeing Bart! What about you?
<spk1>: Wow, I wonder why. I like Simpsons as well. Whe... | 2 | MultiDialog | |
task2_11 | Read the conversation and generate an appropriate answer to the question. | Is the participant who mentioned being skeptical about 'New Girl' due to being difficult to please with comedy the same one who described 'Trial and Error' as brilliant? | True | <spk2>: You got a bad name this, man.
<spk2>: How? With pigeon?
<spk1>: Carrier pigeon.
<spk1>: The fireworks.
<spk2>: You almost need to be connected for anything. But I'm just saying, like, some people are more addicted, some people are more addicted to it, and some people are less addicted to it. I'm not addicted to... | 4 | CHIME-6 | |
task2_90 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who suggested writing help files or integrating tool tips the same one who proposed waiting until the code is ready before putting it back in the group directory? | True | <spk4>: love
<spk3>: think
<spk4>: that
<spk3>: so
<spk4>: kind of
<spk3>: too
<spk4>: stuff
<spk3>: .
<spk2>: We should probably
<spk4>: . Just we'll just write , you know , little help files or something
<spk2>: inquire
<spk4>: .
<spk2>: about that or like integrate tool tips that would be visible .
<spk4>: Yeah .
<s... | 5 | AMI | |
task2_91 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who expressed uncertainty about the bottom-up approach the same one who suggested using the beefy scratch space for storage? | True | <spk4>: Yeah , I'm also not sure how we can go from from bottom-up . I have always thought it's like more that oh , whatever , I'm a can't think of it at the moment .
<spk4>: Probably this is all too complicated worrying about that at that moment anyway .
<spk1>: Mm is there anything else we should discuss ?
<spk4>: No... | 4 | AMI | |
task2_99 | Read the conversation and generate a complex, relational question with a short noun phrase answer. | Is the participant who discussed the ease of recording TV programs with a single button press the same one who proposed offering a customer service similar to Carphone Warehouse for older users? | False | <spk2>: just need
<spk1>: .
<spk2>: one 'cause you've already got all the numbers
<spk1>: You've already
<spk2>: there
<spk1>: got
<spk2>: anyway
<spk1>: the numbers
<spk2>: , yeah
<spk1>: for typing
<spk2>: .
<spk3>: Right
<spk1>: in anyway .
<spk3>: , I've not come across that function but it sounds wonderful
<spk1>:... | 3 | AMI |
π§ M3-SLU Task 2 β Sample Dataset
π£οΈ Multi-Speaker, Multi-Turn, Multi-Modal Spoken Language Understanding
π Overview
The M3-SLU (Task 2 Sample) dataset is part of the M3-SLU Benchmark designed to evaluate speaker-attributed reasoning in multi-speaker, multi-turn conversations.
It pairs long-form audio, transcripts, and contextual metadata, enabling fair comparison between cascade (SD + ASR + LLM) and end-to-end MLLMs.
π This sample includes 100 instances across 4 open corpora β CHiME-6, MELD, MultiDialog, and AMI β for public preview and benchmarking.
π Full release ( β 12 K validated samples ) will follow after π LREC 2026 camera-ready submission.
𧩠Data Structure
| Field | Type | Description |
|---|---|---|
id |
string |
Unique sample identifier |
audio |
audio |
Long-form dialogue segment (1 β 4 min) |
instruction |
string |
Prompt for speaker-grounded comprehension |
question |
string |
Speaker-attributed QA question |
answer |
string |
Short noun-phrase answer |
script |
string |
Transcribed utterances with speaker tags |
n_speakers |
string |
Number of speakers in audio |
data_source |
string |
Source corpus (AMI, CHiME-6, MELD, MultiDialog) |
π― Tasks
π¨οΈ Task 2 β Speaker Attribution Utterance Match (T/F)
Evaluates reasoning about speaker identity by checking if two utterances or actions were made by the same person.*
Each instance requires models to reason over speaker identity + content to choose the correct noun-phrase answer.
π Usage
from datasets import load_dataset
dataset = load_dataset("benchcheck/M3-SLU-Task2-sample", "task2-sample")
print(dataset["sample"][0])
- Downloads last month
- 5