In this work, we revisit the classic Les Misérables benchmark network from the perspective of social communication. Using the 1998 film adaptation of the novel, we construct the publicly available Les Misérables Dialogue temporal text network, which explicitly distinguishes speakers, addressees, and other recipients of each utterance, along with its content and timing. Using the ten social dimensions model, we also derive multilayer network representations of the network, in which layers correspond to distinct relational dimensions. Beyond the network dataset, this work offers a general methodology for building similarly rich communication datasets from similar narrative sources.
Figures & tables
Component
Majority (0-1 flags)
Unanimous (0 flags)
Speaker
99.23%
97.81%
Addressee(s)
98.75%
94.42%
Recipient(s)
99.35%
90.63%
Table 1: Agreement during the independent validation of the utterance annotations.
Figure 1: Flattened communication network LMD-flat a . The 21 characters also present in the canonical Les Misérables network are highlighted and labelled by name.
Network
Nodes
Edges
⟨k⟩
σk
r
c
S
Les Misérables
77
254
6.60
6.04
-0.17
0.50
1.00
LMD-flat a
106
172
3.25
6.18
-0.40
0.09
0.91
LMD-flat r
116
345
5.95
8.39
-0.29
0.29
1.00
Table 2: Structural properties of the original Les Misérables co-occurrence network and the two flattened communication networks derived from our dataset. We report the number of non-isolate nodes and edges, mean degree ⟨k⟩ , standard deviation of the degree distribution σk , degree assortativity r , global clustering coefficient c , and relative size of the largest connected component S .
Layer
Edges
⟨k⟩
σk
r
c
S
d
Power
12 ( 19 )
0.21 ( 0.33 )
0.74 ( 1.12 )
-0.57 ( -0.41 )
0.00 ( 0.05 )
0.03 ( 0.03 )
2 ( 2 )
Similarity
1 ( 1 )
0.02 ( 0.02 )
0.13 ( 0.13 )
– ( – )
0.00 ( 0.00 )
– ( – )
– ( – )
Conflict
27 ( 62 )
0.47 ( 1.07 )
1.50 ( 3.05 )
-0.36 ( -0.36 )
0.10 ( 0.08 )
0.04 ( 0.06 )
3 ( 4 )
Identity
0 ( 0 )
0.00 ( 0.00 )
0.00 ( 0.00 )
– ( – )
0.00 ( 0.00 )
– ( – )
– ( – )
Support
28 ( 54 )
0.48 ( 0.93 )
1.98 ( 3.41 )
-0.62 ( -0.53 )
0.00 ( 0.01 )
0.02 ( 0.03 )
1 ( 3 )
Knowledge
9 ( 13 )
0.16 ( 0.22 )
0.79 ( 0.98 )
-0.74 ( -0.49 )
0.00 ( 0.06 )
0.03 ( 0.03 )
2 ( 2 )
Table 3: Structural properties of the multiplex network derivations, LMD-ML a0.8 and LMD-ML r0.8 (in parentheses). We report number of nodes and edges, mean degree ⟨k⟩ , standard deviation of the degree distribution σk , degree assortativity r , clustering coefficient c , and relative size of the largest strongly connected component S .
Figure 2: Distribution of classification scores for the ten dimensions across all messages.
Figure 3: Cumulative message frequency by social dimension over time, for the three most common character pairs. A message counts toward a social dimension if its score for that dimension exceeds 0.8, binned into 5-minute intervals. Color indicates the running total of messages within each bin. Each row shows one dyad, with the two columns giving the two directions of communication (sender → receiver).
We present "Testimole-conversational" a massive collection of discussion boards messages in the Italian language. The large size of the corpus, more than 30B word-tokens (1996-2024), renders it an ideal dataset for native Italian Large Language Models'pre-training. Furthermore, discussion boards' messages are a relevant resource for linguistic as well as sociological analysis. The corpus captures a rich variety of computer-mediated communication, offering insights into informal written Italian, discourse dynamics, and online social interaction in wide time span. Beyond its relevance for NLP applications such as language modelling, domain adaptation, and conversational analysis, it also support investigations of language variation and social phenomena in digital communication. The resource will be made freely available to the research community.
Matteo Rinaldi, Rossella Varvara, Viviana Patti
*Dipartimento di Informatica, University of Turin, Italy Corso Svizzera 185, 10149 Torino, Italy
Reading a social situation often depends on behavior, not words alone. We introduce FriendBench, a benchmark for inferring whether two people are already familiar or are meeting as strangers, from a 20-second clip of a dyadic ice-breaker conversation. Every pair answers the same type of prompt, so only the manner of interaction can reveal the answer. Across text, audio, and video, we compare 26 models from seven companies against matched human panels over 96 balanced dyads. The best model and the human crowd are statistically indistinguishable on accuracy in every modality, but reach it differently: humans stay balanced across the two answers, while the strongest models favor ``stranger.'' This is a difference in effective prior, not in discrimination. Richer channels help both unequally, and only humans gain from visible behavior on top of speech. We release the stimuli, human ratings, and model predictions.
Jeffrey M. Girard, Jason Z. Zheng, Jacqueline R. Vertino +2
Training Multimodal Large Language Models for audio-visual social understanding is a crucial step toward embodied social intelligence. Chain-of-thought (CoT) reasoning has become the dominant approach, with HumanOmniV2 and its IntentBench benchmark as a prominent reference point. In this context, we report three findings. First, IntentBench is highly noisy: ∼7% of questions are broken and ∼23% are trivially answerable without the video input. We remove the affected questions and release Intentbench-Prime. Second, current reasoning approaches are expensive and surprisingly ineffective. A simple Vanilla SFT baseline matches or outperforms existing reasoning methods across three benchmarks at a fraction of the cost, establishing it as an essential baseline for evaluating novel fine-tuning techniques. Third, our analysis reveals that substantial priors can be learned solely from the text modality and that using a textual caption instead of the video yields performance on par with Vanilla SFT. These surprising findings reveal the limitations of current MLLMs when it comes to social understanding. IntentBench-Prime, Vanilla SFT model, and code are publicly available.
Koen P. de Vries, Xavier Alameda-Pineda, Estefanía Talavera +1
Inria at Univ. Grenoble Alpes, CNRS, LJK, France · University of Twente, The Netherlands