Abstracts and materials

As is well-known, the popularization of the Internet has resulted in a tremendous growth of written communication, including conversational genres or traditions, thanks to the spread of social network sites. This presents linguists with the opportunity of accessing large amounts of conversational data in written format that were unthinkable some decades ago. Moreover, an important part of these data is public and can thus be easily collected. This workshop is divided in two sessions. During the theory session we will reflect on:
1) the characteristics of written digital conversations, comparing it with spoken conversations and non-conversational text: what kind of language are we observing when we look at digital conversations? What kind of insights can we get from the study of digital conversations? What possibilities offer the different social networks?
2) the methodological challenges this kind of data pose, regarding non-standard spelling, multimodality and speaker’s/user’s metadata.
3) the ethical issues the researcher must take into consideration when dealing with public data from social network sites, especially regarding the user’s expectations on privacy/publicity.
In the practical session we will discuss and try out some of the tools that are currently available to collect and analyze data from social networks. We will mainly focus on Twitter, which is one of the most frequently used source of digital text, but we will also address other social networks such as WhatsApp and Tumblr.

In this workshop we will examine the range of sociolinguistic methods that are often called surveys, questionnaires, or written questionnaires (WQs) for their benefits and limitations. The explicit inclusion of WQs in sociolinguistics and dialectology is a fairly recent development, which is owed to a revival of what used to be considered a second-rate method for most of the 20th century. In this workshop we will approach the method with an open mind as seek to explore if, when and in what ways WQs might be the best methodological choice, or at least a viable route of action.
Our views of data collection are limited by our research questions. If we prioritize minute phonetic detail, we are bound to overlook the considerable advantages of WQs. In the theory section we explore the epistemological reasons why WQs had to be argued back into the dialectologist’s methodological toolkit. We will trace their roots back to late 18th century Switzerland and show how, around 1950, the method should have been reinstated. Instead, it got sidelined even more.
In the practical part we will utilize a survey suite (such as Qualtrics) to work on question format, answer option delivery, the framing of WQs as well as research ethics. We will assess WQ items in order to mitigate both question skewing and, in the digital age, limit possible negative effects on our participants. We will see why every raw question needs to undergo testing before it can be put on a WQ.

This talk outlines a particular perspective on the type of language data that should be collected in the field. This perspective is that sentences and discourses are constituents of dialogue and therefore that all field research should include studies of dialogue and conversation, based in part on the linguist’s fluency in the language under study. I examine dialogue’s importance for linguistic theory, and the cooperative nature of theory construction. It is based not only on my own 40+ years of field research in South, Central, and North America, but also on a Peircean perspective of the linguistic enterprise that I develop in a forthcoming book, as well as many previous publications of mine on the importance of studying language and culture as a symbiotic relationship. I will draw on data from Banawa and Piraha, among other languages.

Linguistic data has played a vital role in research on culture and cognition. Language has been viewed as a window on thought since at least Boas, who claimed that “the purely linguistic inquiry is part and parcel of a thorough investigation of the psychology of the peoples of the world” (Boas, 1911: 63). Where correlations between language and thought have been observed, it has been hotly contested whether this correlation is causal, and if so, in which direction. Do our unconstrained patterns of thought cast language in their own mould? Or is it language that shapes thought in its own image? Drawing on a range of empirical studies, I argue that answering these questions is impeded by researchers’ biases (including binary bias, confirmation bias and desirability bias). We must embrace the complex interrelationships between language, culture, environment and cognition, as well as the complexity of each system in its own right, if we are to understand any of them. We begin with insights from a range of studies that tease apart the roles played by culture, language and the environment in shaping spatial cognition, as well as the influence of culture and topography in shaping spatial language. We will then move to consider how spatial language and cognition—along with some other factors—shape how people think and talk about time. Recent findings in these areas reveal a more nuanced relationship between language and thought than a polarised debate can do justice to. As this talk will show, differences in language, culture, local topography, non-linguistic representations and cognition may all—directly or indirectly—influence one another.

This workshop will serve as an introduction to language elicitation techniques for experimental research. It will cover techniques used with adults and children, as well as manipulations specific to gesture research. We will discuss the pros and cons of each method, as well as good documentation practice. The workshop will end with a practical section where one or more techniques (time dependent) will be practiced as a group. No previous training is required.

This workshop provides an introduction to the theory and methodologies that underly corpus-based approaches to syntactic typology, with a special focus on naturalistic data from lesser-used languages. In the first part of the course I will justify the rationale behind corpus-based approaches, and illustrate the arguments with case studies that address core topics from conventional syntactic typology: (i) referential density (anaphora and pro-drop); (ii) the interaction of transitivity and discourse structures; (iii) contact-induced word-order change.
In the second part we take a more hands-on look at the practicalities of corpus creation, first looking at more general issues of corpus design and sustainability, and principles of Open Science (data accountability, accessibility, and maximum utility). We then turn to the annotation of morpho-syntax, semantic categories, and information structure, drawing on examples from Multi-CAST (Multilingual Corpus of Annotated Spoken Texts, Haig & Schnell 2019). Finally, we consider a second approach to data markup, namely token-based, rather than comprehensive, annotation, illustrated using data from ongoing research in the Post-predicate Elements in Iranian and neighbouring languages project.

When as a recent Ph.D. recipient in linguistics in the early 1990s I presented my first paper on language use in an internet discussion forum, the response of mainstream linguists was skeptical: “That’s .. interesting. But is it linguistics?” The field has come a long way since then in terms of appreciating the value of, and embracing the use of, digital language data. I contributed actively to this by developing an approach to analyzing online discourse based on theories and methods from linguistics, including structural linguistics, corpus linguistics, discourse analysis, critical discourse analysis, conversation analysis, pragmatics, and sociolinguistics (Herring, 2004, 2018). The approach – Computer-Mediated Discourse Analysis (CMDA) – includes a toolkit of methods tailored to address different kinds of language-focused questions. In addition, it includes best practices for data sampling, classification of digital data, and assessment measures to enhance the reliability and rigor of the analyses.
In this talk, I will outline the CMDA approach and illustrate its application to different kinds of digital data through worked examples. The data may include, for example, blogs, microblogs, Facebook or YouTube comment threads, Instagram live streaming, support forums, teen chat rooms, or university websites. While mainly focusing on textual data, the presentation will also consider spoken language and graphical elements such as emojis on multimodal platforms. Finally, although CMDA methods are mostly manual, the role of automation in the collection and analysis of digital discourse data will be addressed.

Herring, S. C. (2004). Computer-mediated discourse analysis: An approach to researching online behavior. In S. A. Barab, R. Kling, & J. H. Gray (Eds.), Designing for Virtual Communities in the Service of Learning (pp. 338-376). New York: Cambridge University Press.
 
Herring, S. C. (2018). The co-evolution of computer-mediated communication and computer-mediated discourse analysis. In P. Bou-Franch & P. Garcés-Conejos Blitvich (Eds.), Analysing digital discourse: New insights and future directions (pp. 25-67). London: Palgrave Macmillan.

Tools based on Natural Language Processing and Artificial Intelligence are by now part of our daily life, both in the personal realm as well as in the work-related arena. Recommender systems on social media, automatic translators, job application assessors are just some examples. We take many such tools for granted, and oftentimes we do not even realise they are playing a role, and which role, in our lives.
The Association for Computational Linguistics (ACL) recognises the need to address ethical aspects and consequences related to language processing research. Since 2016, the ACL holds an Ethics in NLP Workshop co-located with one of its major event exactly for such reasons: to highlight, discuss, and possibly fix some of the ethical problems connected with state of the art research in Natural Language Processing. Several papers published at major conferences include an Ethics Statement where the authors communicate what ethical issues might be involved in their research, and what consequences there could be.
In my talk I will discuss such issues. Some of them involve bias, and I will discuss how bias can be found in models, algorithms, data, society, and - importantly - in ourselves, too. We will see that the definition of bias is not straightforward, that what counts as bias isn't a universally accepted concept, as it heavily interacts with our own cultural and individual biases, and that it is not obvious what the best strategies are to deal with it from an NLP perspective. Biases that percolate from our society to our data and thus to our widely used language models and to the tools that are built on them are not the only problem, of course. The fact that contemporary language models, such as the GPT family, are able to generate text that can be indistinguishable from human-produced language yields potentially dangerous situations where, for example, large amounts of unaccountable machine-generated hateful comments are spread to social media, or fake news are beautifully concocted and massively circulated (at a volume and speed which are obviously unparalleled by human production.)
What should we do, then? How can we continue to pursue interesting, useful, and socially important research in language processing without causing harm? How should NLP practitioners pose themselves with respect to ethical issues intrinsic in, and raised by, their research? These are questions that I will critically address by means of examples and by interacting as much as possible with the students.

Since 1986 the Text Encoding Initiative represents a point of reference for the research in Digital Humanities as well as a model for engaging scholarly communities around the world dealing with texts and documents. This field is utterly fragmentary, with as many scholarly approaches as there are countries, languages, and traditions. However, the TEI not only has managed to make these heterogeneous experiences come together, but has indeed provided them with a tool for innovation and collaboration. Text encoding is challenging in many ways: because it demands the acquisition of special skills, because it forces to model texts in ways that are not natural, because it asks researchers and practitioners to change their heuristics and the way they "see" things. Yet, it provides with powerful representational tool and often offers a friendly way to enter in the much larger field of Digital Humanities.
The lecture will present the history of this standard, as well as some of the main features of the TEI encoding method, while the workshop will provide a first hands-on experience of the TEI encoding.

This workshop will cover the methodological principles underlying the collection of cross-linguistic spoken data, and provide practical examples of the issues that must be dealt with when field work and language documentation extend to the production of accessible and re-usable language data. Collecting, processing, and making fieldwork data accessible in a way that enables re-use beyond the fieldwork project itself is nowadays considered a responsibility to funders, to the speech community, and to the scientific community. It also has a number of benefits for the fieldworker, while the additional effort to optimize fieldwork data for re-use is relatively small. This workshop covers practical steps for optimization of data that are typically collected in the context of relatively small-scale fieldwork projects on small and under-described languages. The focus of the workshop is on the re-use of annotated texts, e.g. traditional narratives, for linguistic research, especially for cross-linguistic research in phonetics and prosody. It covers audio recording quality, contents in terms of genres, minimal annotation and metadata requirements, archiving, and open access to data.