On social media, additional context is often present in the form of annotations and meta-data such as the post’s author, mentions, Hashtags, and hyperlinks. We refer to these annotations as Non-Textual Units (NTUs). We posit that NTUs provide social context beyond their textual semantics and leveraging these units can enrich social media text representations. In this work we construct an NTU-centric social heterogeneous network to co-embed NTUs. We then principally integrate these NTU embeddings into a large pretrained language model by fine-tuning with these additional units. This adds context to noisy short-text social media. Experiments show that utilizing NTU-augmented text representations significantly outperforms existing text-only baselines by 2-5% relative points on many downstream tasks highlighting the importance of context to social media NLP. We also highlight that including NTU context into the initial layers of language model alongside text is better than using it after the text embedding is generated. Our work leads to the generation of holistic general purpose social media content embedding.

NTULM: Enriching Social Media Text Representations with Non-Textual Units

A Standardized Speaking Valve Simulation to Assess Students Clinical Communication

**Author(s):** Morin, O., Mueller, T.F., Morisseau, T., Winters, J. **Author(s):** No **Author(s):** No The interpretation of referential messages is biased by communicative salience: some referents are easier to communicate because they come more readily to mind. We explored two aspects of communicative salience, in a referential communication task involving colors. We predicted that lexically focal referents would be communicatively salient, and that the codes used to indicate them would be more stable. We tested these hypotheses with an online gaming app, the Color Game, in which players had to use ambiguous black and white symbols to communicate specific color hues to other players (> 200,000 trials). Focal colors, centrally associated with basic color terms, are easier to communicate than less focal ones. Code stability is contingent on the type of colors conveyed, with participants being more likely to keep using the same symbols when referring to focal colors. Communicative salience impacts the evolution of semantic conventions. These findings bridge the pragmatics of referential communication and the evolution of lexical semantics.

Communicative salience shapes the emergence and stability of semantic conventions

Introduction: The observation that child-directed speech (CDS) benefits language learning has been taken to suggest that this register is part of natural pedagogy – a suite of evolved pro-social communicative behaviours that use ostension to teach culturally relevant knowledge. While prosodic features of CDS appear universal, phonetic, morphosyntactic and lexical features tend to vary cross-linguistically. I argue that (a) this variation in CDS reflects different solutions to trade-offs between affective bonding and teaching, based on specific linguistic affordances and cultural constraints, and that (b) the way didactic features are incorporated into CDS is itself acquired and transmitted. Evidence: Preliminary evidence comes from (a) comparisons of acted vs spontaneous CDS, where positive affect expression can render the latter suboptimal for language learning, (b) CDS of proficient L2-speakers which lacks some features of native CDS and (c) iterated language learning experiments in which learners who subsequently teach change transmission. Further evidence should come from tracing changes in teaching efficiency during cultural transmission under social vs asocial learning conditions. Discussion: CDS may be a special case of conventionalisation of didactic features which can themselves be learned thereby reducing the individual cost of prosocial behaviours like teaching.

Learning to Transmit: The Acquired Nature of Child-Directed Speech

**Objective** Medical journals often use visual abstracts (VAs),
“infographics” designed to graphically convey a study’s
research question, methods, findings, and conclusions, to
summarize and promote published research on social media.
Studies suggest that VAs increase an article’s reach and
engagement, mostly demonstrated for individual journals on
Twitter.1-3 The JAMA Network, a family of 2 general medical
and 11 specialty journals, began creating VAs for randomized
clinical trials (RCTs) in 2018-2020 and publishing them
online and in social media. This study examined the
association of VAs with social media–driven site traffic and
metrics. 
 **Design** This random-sequence simultaneous crossover trial
included all RCTs published in the 13 JAMA Network journals
with an accompanying VA between September 21, 2021, and
May 15, 2022. For each, tweets and Facebook posts
promoting the RCT were published including a standard text
summary with a link to the online article, accompanied by 3
image display types: (1) preview VA, a VA thumbnail that
linked to the article when clicked; (2) expandable VA, a VA
thumbnail that expanded to full screen when clicked; and (3)
article image, an article figure or table thumbnail that linked
to the article (Facebook) or expanded to full screen (Twitter)
when clicked (control display). Three tweets and 3 Facebook
posts, 1 with each image display type, were published to
Twitter and Facebook nearly simultaneously (1 minute apart)
in random sequence within an hour of article publication. The
primary outcome was the number of link clicks by image
display type at 1 week. Secondary outcomes were social media
Twitter impressions/Facebook reach (how many times people
saw the tweet or post) and engagement (sum total number of
comments/replies, shares by other users, and likes) by image
display type at 1 week. The study was designed to provide
80% power to detect a 50% increase in median link clicks
comparing either of the 2 VA formats vs article image display. 
 **Results** Among 205 JAMA Network RCTs with VAs, link
clicks to full-text articles were higher with preview VAs vs
article images; impressions/reach and engagement were
higher with expandable VAs vs article image displays
(**Table 24**). In a preplanned subgroup analysis, the higher
link clicks appeared attributable to preview VA display on
Twitter (preview VA median, 12 [IQR, 3-41] and expandable
VA median, 6 [IQR, 2-21] vs article image median, 6 [IQR,
1-21]; P < .001) rather than Facebook (medians, 1 [IQR, 0-14]
and 3 [IQR, 0-12] vs 1 [IQR, 0-13], respectively; P = .11). 
 ![](https://assets.underline.io/uploads/markdown_image/1/image/5a79dbae45c21d9e137c279642a638cf.png) 
 **Conclusions** In this random-sequence simultaneous
crossover trial, the use of visual abstracts in social media
posts was associated with higher social media–driven journal
site traffic when displayed as preview links, and higher social
media reach and engagement when displayed as expandable
images, compared with article images. 
 **References**
1. Oska S, Lerma E, Topf J. A picture is worth a thousand
views: a triple crossover trial of visual abstracts to examine
their impact on research dissemination. J Med Internet Res.
2020;22(12):e22327. doi:10.2196/22327 
 2. Ibrahim AM, Lillemoe KD, Klingensmith ME, Dimick JB.
Visual abstracts to disseminate research on social media: a
prospective, case-control crossover study. Ann Surg.
2017;266(6):e46-e48. doi:10.1097/SLA.0000000000002277 
 3. Chisari E, Gouda Z, Abdelaal M, et al. A crossover
randomized trial of visual abstracts versus plain-text tweets
for disseminating orthopedics research. J Arthroplasty.
2021;36(8):3010-3014. doi:10.1016/j.arth.2021.04.006 
 **Conflict of Interest Disclosures** N. Seth Trueger receives
salary support from the American Medical Association (AMA) for
his role as digital media editor, JAMA Network Open. Eman Aly
and Michael Berkwits are full-time employees of the AMA and the
JAMA Network. Sebastien Haneuse receives compensation from the
AMA for his role as statistical editor, JAMA Network Open. Michael
Berkwits is co-director of the Peer Review Congress but was not
involved in the review or decision for this abstract. 
 ![](https://assets.underline.io/uploads/markdown_image/1/image/d13a71ce0b4a73f13e287ffb906b44ed.png)

Association of Medical Research Visual Abstract Display With Social Media Driven Site Traffic

**Objective** To assess editors’ experience with and openness to
refining journal article abstracts by adding a results table and
a limitations header to improve abstract readability and
informational content.1-3

**Design** General medical journals were selected based on
journal impact factor rankings: all top 10 ranked journals and
5 among those ranked in the third quartile, published in
English, with multiple issues per year, and using a structured
abstract. Only one journal from the JAMA Network was
selected. Semistructured interviews were conducted with the editor recommended by the journal’s editor in chief. The
study ethics approval (exemption) was provided by
Dartmouth.

**Results** Eleven of the 15 invited journals participated (9
from the top 10 and 2 of 5 from the third quartile by impact
factor). Interviews were conducted with 4 editors in chief, 3
executive editors, and 4 other editor types by S.W. on a web
conferencing platform (1 editor responded in writing) from
February 4 to March 4, 2022, and lasted 15 to 20 minutes.
Calls were recorded and autotranscribed. All study authors
reviewed the full interview transcripts, R.J.W. summarized
key themes from transcripts, and all authors reached
consensus on abstract results tables key themes (**Table 56**).
One journal had experience publishing abstract results tables.
There was strong interest in a limitations header and few
concerns about it having any potential harms.

![](https://assets.underline.io/uploads/markdown_image/1/image/061ccdeb247a9e97036918916ab0570d.png)

**Conclusions** These findings provide preliminary support for
a trial evaluating the addition of results tables and limitations
to abstracts. Limitations of this study are that it may not be
representative of all journals and interviews did not include
abstract readers or authors.

**References**

1. Bauchner H, Henry R, Golub RM. The restructuring of
structured abstracts: adding a table in the results section.
JAMA. 2013;309(5):491-492. doi:10.1001/jama.2013.76
2. Ertl N, Gazette K. A new way of documenting scientific data
from medical publications. Karger Gaz. 1969;20(27):1-3.
3. Haynes RB, Mulrow CD, Huth EJ, Altman DG, Gardner
MJ. More informative abstracts revisited: a progress report.
Ann Intern Med. 1990;113(1):69-76. doi:10.7326/0003-4819-
113-1-69

**Conflict of Interest Disclosures** Steven Woloshin serves on the
Cochrane Collaboration and the JAMA Internal Medicine editorial
boards. Lisa Bero is Senior Research Integrity Editor, Cochrane,
and serves on the Cochrane Editorial Board; an academic editor,
Meta-Research, PLoS Biology; and a member of the Peer Review
Congress Advisory Board but was not involved in the review or
decision for this abstract. No other disclosures were reported.

**Acknowledgment** This paper is dedicated to the memory of Lisa
Schwartz, MD, MS, our remarkable partner, colleague, and friend
whose work inspired this project. The authors thank the journal
editors and staff who participated in interviews.

![](https://assets.underline.io/uploads/markdown_image/1/image/3ecec4b2b458a89d469a50971969554a.png)



Editors' Perspectives on Adding a Results Table and Limitations Section to Medical Journal Abstracts: A Qualitative Study

**Objective** To compare media attention, Twitter engagement,
and citations among COVID-19–related clinical trial preprints
and corresponding peer-reviewed publications. 
 **Design** Preprints of clinical trials were included in this
cross-sectional study if they reported results related to the
treatment or prevention of COVID-19 and were uploaded to
an open access preprint server indexed by the National
Institutes of Health iSearch COVID-19 portfolio in 2020. Two
investigators, including a medical librarian, independently
searched Medline, Google Scholar, and Embase to identify
peer-reviewed publications corresponding to the included
preprints. Altmetric data were used to quantify media
mentions and Twitter interactions within 3 months after
publication for both preprints and peer-reviewed
manuscripts. Citation counts were also recorded using Web of
Science. Descriptive statistics were reported for the included
trials, and linear regression was used to assess associations
between study characteristics and media mentions, Twitter
interactions, and citation counts. 
 **Results** Of 22,615 preprints screened for eligibility, 145 were
included. MedRxiv was the source for most eligible preprints
(n =100; 69%). Peer reviewed publications matching 118 of
145 preprints (81%) were found. Sixty-eight preprints (47%)
received media citations within 3 months of publication
(median [IQR] number of mentions per preprint, 0 [0-9];
maximum mentions, 568), and 118 (81%) had Twitter
interactions (median [IQR] tweets per preprint, 28 [4-202];
maximum mentions, 18,177). One hundred preprints (69%)
were cited in published literature (median [IQR] citations, 2
[0-12]). Among 118 preprints with matching peer-reviewed
publications, Altmetric and citation data were unavailable for
5 (4%). Preprints received more media mentions than the
corresponding peer-reviewed publications in 33 cases (28%),
equal mentions in 38 cases (32%), and fewer media mentions
in 42 cases (36%). Twitter mentions were greater for
preprints than peer-reviewed publications in 54 cases (46%),
equal in 2 (2%), and fewer for 62 (53%). In 26 cases (22%),
preprints received more citations than peer-reviewed
publications; citation counts were equal in 6 cases (5%) and
in 85 cases (72%), the peer-reviewed version received more
citations. Study size, government funding, failure to
prospectively register, and positive study results were most
often associated with increased media mentions, tweets, and
citations (**Table 20**). 
 ![](https://assets.underline.io/uploads/markdown_image/1/image/59664a9a5f5c8dc866bb6ab1f830560e.png) 
 **Conclusions** Although peer-reviewed publications had more
media, Twitter, and citation activity than corresponding
preprints in most cases, it was not uncommon for preprints to
receive more attention than peer-reviewed publications.
Measures of trial reliability or quality were generally not
associated with increases in these metrics. 
 **Conflict of Interest Disclosures** Christopher W. Jones reported
receiving grants from AstraZeneca, Abbott, Vapotherm, and
Ophirex outside the submitted work. Timothy F. Platts-Mills is an
employee of Ophirex. No other disclosures were reported. 
 **Funding/Support** This work was supported by the US
Department of Health and Human Services (HHS) Office of
Research Integrity (grant ORIIR200066-01-00). 
 **Role of the Funder/Sponsor** The sponsor had no role in
the design and conduct of the study; collection, management,
analysis, and interpretation of the data; preparation, review, or
approval of the manuscript; and decision to submit the abstract for
presentation. 
 **Additional Information** The contents of this manuscript are
those of the authors and do not represent the official views of, nor
an endorsement by, the Office of the Assistant Secretary for Health,
the HHS, or the US government. 
 ![](https://assets.underline.io/uploads/markdown_image/1/image/3af6ee0204355a3795cd4012c2ef7123.png)

Media Attention, Twitter Engagement, and Citations of COVID-19 Clinical Trial Preprints and Their Corresponding Peer-Reviewed Publications

Online data collection has become a prominent option due to the COVID-19 pandemic. It is crucial to understand to what extent online studies can be compared with face-to-face studies, particularly in multimodal language research on which the modes of communication have a crucial effect. This study investigated multimodal communication across face-to-face and videoconferencing settings, focusing on gesture production and speech disfluency in a daily routine description task (N=64). Results suggested that overall disfluency rate was higher for those who communicated via videoconferencing than those who communicated face-to-face. The use of specific disfluency types also differed across the two settings, signaling an interplay between cognitive and communicative strategies. Overall gesture frequency and iconic gesture use were comparable across the two settings. Iconic gesture use negatively predicted the overall disfluency rate, regardless of the setting. Using different contexts is required to understand whether multimodal language differs between face-to-face and online communication.

Multimodal Communication in Virtual and Face-to-Face Settings: Gesture Production and Speech Disfluency

Myriad linguistic cues, subtle and otherwise, have been shown to shape thinking and reasoning. Such linguistic framing effects are attributed variously to irrational biases, affective processes, and analogy. Across two studies, we explored pragmatic reasoning (PR) as a potential unifying mechanism underlying disparate framing effects. In Study 1, we adapted a measure of PR used in attribute framing research, assessing the ability to infer information subtly communicated by a speaker’s choice of frame. Our measure captured individual differences in PR distinct from cognitive reflection and mentalizing ability. In Study 2, PR predicted the effect of identity labels on attitudes towards illegal immigration, but not other framing effects. Two such effects were driven, however, by participants’ explicit recognition of the influence of the framing language. These results suggest that framing effects may indeed be mediated by PR, but that our measure of PR does not fully capture its mediating role.

Exploring the Role of Pragmatic Reasoning in Linguistic Framing Effects

Gestures facilitate speech production by helping speakers reduce cognitive load. Studies on gesture-speech interaction mostly examined the effect of representational gestures on spatial contexts. However, abstract deictics (e.g., pointing at objects that are not visually present) might also have a role in facilitating cognitive processes. The present study investigated the effect of gestures on disfluency rates by presenting a narrative task in three conditions: spontaneous, restricted, and encouraged gesture use. We found that disfluency rates across three conditions did not significantly differ. The use of abstract deictics in the spontaneous gesture use condition was a significant predictor of disfluency rates in the gesture restricted condition. Results indicate that gestures’ facilitative roles might be manifested differently depending on the context. Abstract deictics might also benefit speakers, especially in a narrative context. Studying abstract deictics can provide new insights on gesture and speech production interaction.

Gesture and Speech Disfluency in Narrative Context: Disfluency Rates in Spontaneous, Restricted, and Encouraged Gesture Conditions

Accessibility plays a major role in speech production. Here we investigate and measure four factors that influence speakers to produce one word over another more optimal word form. Three experiments asked participants to label images of insects and instruments. Participants were incentivized to produce an accurate specific label (e.g., bee), over a more general label (e.g., insect), so that specific labels were more optimal. Each of three experiments manipulated a different factor that could influence accessibility – word frequency, priming, and interference – and all experiments additionally varied whether labels had to be produced under time pressure or not. Results showed that each variable significantly influenced the accessibility of labels: participants produced more specific labels when those labels were higher frequency, when they were primed, when a visually-similar label had not been primed, and when participants were unconstrained by time pressure. These findings demonstrate that multiple factors influence the accessibility of familiar words during production, regularly leading participants to rely on “good-enough” rather than optimal options to convey their message.

Accessibility factors that lead to good-enough language production

There is growing evidence that mobile text message exchanges between patients and therapists can augment traditional cognitive behavioral therapy. The automatic characterization of patient thinking patterns in this asynchronous text communication may guide treatment and assist in therapist training. In this work, we automatically identify distorted thinking in text-based patient-therapist exchanges, investigating the role of conversation history (context) in distortion prediction. We identify six unique types of cognitive distortions and utilize BERT-based architectures to represent text messages within the context of the conversation. We propose two approaches for leveraging dynamic conversation context in model training. By representing the text messages within the context of the broader patient-therapist conversation, the models better emulate the therapist's task of recognizing distorted thoughts. This multi-turn classification approach also leverages the clustering of distorted thinking in the conversation timeline. We demonstrate that including conversation context, including the proposed dynamic context methods, improves distortion prediction performance. The proposed architectures and conversation encoding approaches achieve performance comparable to inter-rater agreement. The presence of any distorted thinking is identified with relatively high performance at 0.73 F1, significantly outperforming the best context-agnostic models (0.68 F1).

Underline digital video library

NTULM: Enriching Social Media Text Representations with Non-Textual Units

A Standardized Speaking Valve Simulation to Assess Students Clinical Communication

Communicative salience shapes the emergence and stability of semantic conventions

Learning to Transmit: The Acquired Nature of Child-Directed Speech

Association of Medical Research Visual Abstract Display With Social Media Driven Site Traffic

Editors' Perspectives on Adding a Results Table and Limitations Section to Medical Journal Abstracts: A Qualitative Study

Media Attention, Twitter Engagement, and Citations of COVID-19 Clinical Trial Preprints and Their Corresponding Peer-Reviewed Publications

Multimodal Communication in Virtual and Face-to-Face Settings: Gesture Production and Speech Disfluency

Exploring the Role of Pragmatic Reasoning in Linguistic Framing Effects

Gesture and Speech Disfluency in Narrative Context: Disfluency Rates in Spontaneous, Restricted, and Encouraged Gesture Conditions

Accessibility factors that lead to good-enough language production

Identifying Distorted Thinking in Patient-Therapist Text Message Exchanges by Leveraging Dynamic Multi-Turn Context

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES