CMC Posts

While the concept of a post is not unique to computer-mediated communication (ask anyone who has posted a lost cat sign in the local market), this chapter concerns itself only with postings within a framework of a CMC system. Thus the element post is unique to the encoding of computer-mediated communication (CMC). Posts occur in a broad range of written CMC genres, including (but not limited to) messages in chats and WhatsApp dialogues, tweets in X (Twitter) timelines, comments on Facebook pages, posts in forum threads, and comments or contributions to discussions on Wikipedia talk pages or in the comment sections of weblogs.

Posts can be either written or spoken: written or multimodal posts: In the majority of CMC technologies posts are composed as stretches of text using a keyboard or speech-to-text conversion software in an entry form on the screen. In many cases the technology allows authors to include or embed graphics (emojis or images), video files, and hyperlinks into their posts. spoken (audio posts): A growing number of CMC technologies, e.g. messenger software such as WhatsApp or RocketChat, allow for an alternative, spoken production of posts by providing a recording function which allows users to record a stretch of spoken language and transmit the resulting audio file to the other parties.

The element post may co-occur with u, kinesic, incident, or other existing TEI elements within a div, or directly within the body, and may contain headings, paragraphs, openers, closers, or salutations.

The post element is a member of several TEI attribute classes, including att.ascribed, att.canonical, att.datable, att.global, att.timed, and att.typed, and as such may take a variety of attributes.

Attributes Specific to CMC post

Three attributes pertain specifically to post: The type of the content of a post (i.e., whether the content is text, an image, a video clip, etc.) is indicated by the child elements of the post. (E.g., a post might have a child p, or a child figure with a graphic, or a child figure with a media, or some combination thereof.) How that content was created—whether it was recorded speech or not—may be described with the modality attribute. Because spoken language differs significantly from written language, the suggested values only separate the spoken modality from the written modality, which covers all cases other than spoken natural language. The use of modality is recommended but not required.

screenshot of the google search for hairdresser "Pasha's Haare'm" with the average google rating (4,5 of 5 stars), the address, the phone number, and the opening hours.

The replyTo attribute is used to capture information drawn from the original metadata associated with a post that asserts to which previous post the current post is a response, or to which previous post it refers. This metadata is included by many, but not all, CMC environments, when the user executes a formal reply action (e.g., by clicking or tapping a reply button). This attribute should not be used to encode interpreted or inferred reply relations based on linguistic cues or discourse markers.

The replyTo attribute indicates the replied-to or referred-to posts by providing one or more pointers to them. In the following example, reply references in the source indicate that the first post is a reply to an initial post that is not part of the example, the second is a reply to the first, and the third is a reply to the second.

Es hat den Anschein, als wäre bei BER durchaus große Kompetenz am Bau, allerdings nicht in Form von Handwerkern….

http://www.zeit.de/2015/29/imtech-flughafen-berlin-ber-verzoegerung/komplettansicht

Nein Nein, an den Handwerkern kann es rein strukturel nicht gelegen haben. Niemand lässt seine Handwerker auf der Baustelle derart allein. Zudem gibt es höchstoffizielle “Abnahmen” von Bauabschnitten/phasen. Welcher Mangel auch bestanden hatte, er hätte Zeitnah auffallen müssen.

Uuups, für Imtek hab ich mal in einer Nachunternehmerfirma gearbeitet. Imtek is offenbar ein universeler Bauträger, der alles baut.

Stahlkunstruktionen dacht ich mal, was die bauen—oder bauen lassen.

Das ist schon ein übles Ding. Die Ausschreibungenund Angebote sind unauffällig, aber wenn Unregelmässigkeiten auftreten (im Bauverlauf) dann gibt es die saftigen Rechnungen. Da steht dann der Bauherr da und fragt sich, wie er denn so schnell einen fähigen Ersatz herbekommt. Und diese Frage erübrigt sich meist, weil der Markt der Baufirmen das nicht hergibt — weil tendenziel 100 % Auslastung. (und noch schlimmer: Absprachen) Was auch Folge des Marktdrucks gewesen war.

In the CMC genre of wiki talk, users insert their contribution to a discussion by modifying the wiki page of the discussion—the talk page. Since there is no technical reply action available in wiki software, users apply textual indentation in the wiki code to indicate a reply to a previous message, and a threaded structure is formed by a series of such indentations. The attribute indentLevel records the level of indentation, that is the nesting depth of the current post in such a thread-like structure (as defined by its author and in relation to the standard level of non-indentation which should be encoded with an indentLevel of 0). It is used in wiki talk corpora but may also be used for other threaded genres, for example when HTML is used as a source.

The following is a sample encoding of a portion of a discussion among four different users on a Wikipedia talk page.

Attributes for General CMC Encoding

The attribute generatedBy is also unique to CMC encoding. But unlike modality, replyTo, and indentLevel, generatedBy is available not only on the post element, but on any of its descendants as well.

The generatedBy attribute may indicate, for post or any of its descendants, how the content transcribed in an element was generated in a CMC environment. That is, whether the source text being transcribed was created by a human user, created by the CMC system at the request of a human user (e.g., when the user activates a template that generates the content, such as in a signature), generated by the CMC system (e.g. a status message or a timestamp), or generated by an automated process external to the CMC system itself. This attribute is optional; when it is not specified on a post element its value is presumed to be unspecified; when it is unspecified on any descendant of post its value is inherited from the immediately enclosing element. In turn, if generatedBy is not specified on that element it inherits the value from its immediately enclosing element, and so on up the document hierarchy until a post is reached; the post either has a generatedBy attribute specified or its presumed value is unspecified.

A list of suggested values for generatedBy follows: human when the content of the respective element was naturally typed or spoken by a human user (cf. the chat posts in example haircut) template when the content of the respective element was generated after a human user activated a template for its insertion (often applicable to signed and time; e.g. see the signature in wiki talk in this example below) system when the content of the respective element was generated by the system, i.e. the CMC environment (see, e.g., the system message in an IRC chat in the this other example below) bot when the content of the respective element was generated by a bot, i.e. a non-human agent, typically one that is not part of the CMC environment itself unspecified when it is unspecified or unknown how the content of the respective element was generated (see, e.g. the retweet that forms the second post in this example below).

The following is a sample encoding of a chat post that contains an emoji. Although the post was written by a human, the emoji itself was marked in the source as having been generated by a template: Da kostet ein Haarschnitt 50 €

face screaming in fear U+1F631

In the following example, the user signature of a wiki talk post was inserted by activating a template, and is thus marked accordingly:

I'm not sure that this is a proper criterium, or even what this means. What if we set an explosion that breaks a comet into two pieces? What if we build a moon? Cheers,

Greenodd (talk) 01:00, 21 July 2011 (UTC)

In the following example, a tweet is specified as having been written by a human; however inside the tweet, the timestamp is marked as generated by the CMC system: 12:31 Heute mit super Unterstützung, wir grunzen, wenn die Zeit vorbei ist. #bcrn18 #wikidach PS: Die beiden brauchen noch Namen. Hinweise dazu am Empfang abgeben! @AndreLo79

12:43

Finally, in the following example of an IRC post, the status message that user Interseb has entered the room was generated by the system, i.e. the chat environment.

Interseb betritt den Raum.

Macrostructure of CMC Collections and Documents

When CMC documents are compiled into a collection, dataset, or corpus, we distinguish the following levels in the macrostructure of CMC in TEI: The corpus level

The level of a corpus or collection of CMC texts of a particular genre, generally obtained from a particular CMC platform, sometimes even from several platforms. This level may be represented by either a TEI element or a teiCorpus element. The teiHeader of the corpus (i.e., the teiHeader that is a child of the outermost TEI or teiCorpus) will contain metadata in its sourceDesc about the CMC platform(s). Metadata about the project responsible for collecting the data and building the corpus, if applicable, should be recorded as well.

The document level

A set of posts collected (or sampled) by a researcher for analysis. The posts of the document will often map directly to the set of posts grouped on an existing web page, thread, or document within a CMC environment. Within the CMC environment the document as such is often created by a particular user, thereby initiating the communication which other users may read, and to which some other users might contribute. This level will naturally be represented by the TEI element. The teiCorpus (or TEI) element that represents the corpus will contain one or more TEI elements as usual.

In the teiHeader of a document level TEI, the sourceDesc will contain metadata about the CMC document such as a title, its author or owner, its URL, the date of its creation, the date of the last change made to it, and other metadata that are available and to be recorded such as one or more categories associated with the document.

The document sometimes contains, or is associated with, a prompt such as a video or a news item, either provided by the initiating user herself or located elsewhere and referenced at the beginning of the document. In such cases, the teiHeader of the document should also contain metadata about this prompt.

The post level

The level of the individual post is naturally represented by the post element; its encoding is further described in section . A TEI element will contain a number of post elements, which can be grouped or ordered in div elements representing sequences or threads (section ) if appropriate.

Sequences, Sections, Threads

As shown in Example above, nested threads of posts may be encoded sequentially, while the indentLevel attribute of post is used to keep track of the original nesting depth. This is especially meant for CMC text obtained from a wiki code or HTML source, where it is not always entirely clear whether the indentation information actually reflects a reply action from a user.

In genres where technical reply information is available for each post, reply links can be encoded using the replyTo attribute on post elements, as shown in the second example of . The network of all reply links will then also form a threaded structure, and visual indentations can be reconstructed from it and need not be explicitly encoded.

Threads may also be explicitly encoded as nested div elements as in the following skeleton:

Using this encoding strategy, this example from could be encoded as follows:

Multimodal CMC

As explained in section , the elements post, u, kinesic, and incident are available to encode textual transcriptions of written posts, spoken turns, bodily activities of avatars, and onscreen activity by users that occur in CMC data; and, as discussed in section , graphics or other media data within posts are encoded in a post with modality set to written. When two or more of these features occur in a CMC interaction, we can speak of multimodal CMC.

Some basic multimodality is available in many private chat systems such as WhatsApp, where spoken and written posts and media posts containing images or video clips can alternate in the sequence of posts. The following shows the suggested encoding of an extended part of the haircut chat example from above, including a spoken post, several written posts, and a post containing a graphic image (adapted from the MoCoDa2 corpus )

In Düsseldorf gibt's da so Abstufungen. Da gibt's einmal Oliver Schmidt, Oliver Schmidt's Hair Design, also dann, ist eher also, keine Ahnung, zum Beispiel ich war da bei dem etwas Günstigeren dann. Ich weiß nicht, ob's das in Essen auch gibt diese Abstufungen Ich schau mal :) Ich gehe immer nach Katernberg zu Pasha’s haarem Hahaha also die sind echt entspannt und gut und nicht teuer

screenshot of the google search for hairdresser "Pasha's Haare'm" with the average google rating (4,5 of 5 stars), the address, the phone number, and the opening hours.

Olivers hair und Oliver Schmidt gehören zusammen

In the graphical user interface (GUI) of a more complex multimodal CMC environment such as Second Life, a gaming and learning platform, interactions may consist of interleaved occurrences of posts (p), utterances (u) and nonverbal acts such as bodily activities (kinesic) or other on-screen activities (incident). In the following example a spoken utterance, an avatar's bodily activity, and a written post occur on the same level within the body element, representing parts of a multimodal chat in Second Life (adapted from the Archi21 corpus). ok hm for me this presentation was hm become too fast because it's always the same in our architecture school euh we have not time and hm too quickly sorry [...] eat(popcorn)

it went too quickly?

Note that the spoken utterance u represents a speaker turn that was transmitted via an audio channel of the application that is continuously open during a session, whereas a spoken post represents a spoken message that has been recorded in private and been posted to the CMC server as a whole. See section .

Documenting the Source of a Corpus of CMC data

The teiHeader of the corpus should contain metadata about the CMC platform(s), e.g. its name, information about its owner (often a company) including their address or location, the URL of the server where the CMC data were collected from, or the filename of a database dump that was used as a source. Metadata about the project responsible for collecting the data and building the corpus, if applicable, should be recorded as well.

The following example shows the sourceDesc of a X (Twitter) corpus. Twitter International Company

1 Cumberland Place Fenian Street Dublin 2 D02 AX07 Ireland

The following example shows how a Wikipedia database dump may be encoded as the source. Dump file in XML (compressed) Wikimedia Foundation, Inc. 01 Aug 19 dewiki-2019-08-01-pages-meta-current

Describing the Source of a CMC Document

A CMC document may be a chat logfile, a discussion page, or a thematical thread of posts as encoded within a TEI element. Among the metadata to be recorded in the sourceDesc of its teiHeader are, if available, its title, author or owner, its URL, the date of its creation and/or the date of its last change (i.e. the time when the last post was added to it).

The following example is the sourceDesc of a TEI encoding of a YouTube page that contained a video and user comments on the video (which are encoded in the body of the text as posts). The metadata contain a URL reference to the video and the YouTube channel that posted the video in relatedItem elements. The date when the page was created is not known. The example is adapted from the NottDeuYTSch corpus (), where the video itself is not contained in the corpus. DieAussenseiter posted video, created page YouTube

The following example is the sourceDesc of a Wikipedia talk page. Note that a relatedItem element is used to record a reference to the Wikipedia article that the transcribed discussion is about. OnkelSchuppig, et al. Wikimedia Foundation, Inc. 7632113 FKM-Richtlinie

Documenting the Sampling of CMC data

The documentation of how the data were collected, e.g. how it was scraped or sampled from the web, or downloaded from a server, should be recorded in the samplingDecl. Like other metadata, information about sampling should be recorded at the highest level applicable. That is, if the information applies to an entire corpus, the samplingDecl should appear in the teiHeader of the corpus level; if the information is different for each document, it should appear in the teiHeader of the document level texts.

The sampling information typically considered of interest consists of at least the following four components: interface: The API that was used for the download, possibly encoded as a name type="API"; client: The client or other tool that was used for the download, possibly encoded as a name type="client"; query: The query or command used for the download, possibly encoded with a ptr type="query" when it is a URI, or a code when it is a command; date: The date of the download. For example, in the case of an X (Twitter) corpus a sampling declaration might look like the following:

Sampled using the Twitter Filtered stream v2-API (see ) Filtered for the German language and the following countries: Germany, Austria, Belgium, Switzerland, Denmark, and Luxembourg. Downloaded on Mon 12 Dec 22 using the command requests.get("https://api.twitter.com/2/tweets/search/stream", headers=headers, params=params, stream=True,) in the python script collectFilteredTwitterStream.py.

The samplingDecl of a Usenet Newsgroup corpus:

Downloaded from the news.individual.de server on 2016-01-15 using nntp client in Python

Participants

A listPerson may be used to maintain an inventory of users and bots taking part in a CMC interaction, along with information about them. As with other such contextual information, it may be kept in the teiHeader (where it would occur in a particDesc within a profileDesc) or in a separate document completely. In either case, an encoded post may then be linked to its author by use of the who attribute.

In the following example, a list of participants is maintained in a teiHeader. M /wiki/User:M mike@mydomain.com CH P /wiki/User:P pat@super.net ES PKP /wiki/User:Pi

In the following version of the body portion of the same example, the list of interactants is stored in a separate file (in this case the file userList.xml in the same directory).

Alternatively, a prefixDef may be used to declare a prefix which can be used in the value of who to generate a complete URI, thus making the values of who shorter, less error-prone, and easier to maintain. For example, the prefix uL: could be used to map the value uL:06 to file:/userList.xml#cmc_user_06. See for more information on establishing prefix definitions.

This indirection—using a listPerson, particularly one in a separate file, to store information about the users involved in a CMC interaction—is particularly useful when there is both a need to keep such information locally, and to remove it (e.g., to anonymize the data) when the data are published or shared with other researchers.

Timeline

From most CMC environments, user posts come provided with a timestamp marking the time (often down to the second) when the post arrived and was registered at the CMC server. In the display of chat interactions, for instance, the time is automatically added by the system and usually precedes or follows the actual content of the post. In Wikipedia talk, a timestamp is automatically added when the user inserts his or her signature. A timestamp in the text body may be transcribed using a date or time element, in which case the when attribute may be used to record a normalized version of the date, time, or date and time if this information is available or reconstructible. 21:52 das ist auf jedenfall krankheit

Those haven't happened. If they do, we can revisit the concern.

P 01:35, 8 April 2014 (UTC) Alternatively the timestamp may be recorded using the when attribute of post. In this case, if the details of how the timestamp appeared in the original are considered unimportant, the actual transcription may be omitted.

Those haven't happened. If they do, we can revisit the concern.

P

Instead of transcribing timestamps or recording the timestamp directly on an attribute of post, all timestamps of a set of posts can be collected in when elements in a timeline element in the teiHeader, most suitably in the interaction element (itself in the textDesc in the profileDesc). In which case, similar to the encoding of transcripts of spoken utterances (for which see ), each individual post can be linked to its timestamp via the synch attribute as in the following alternative encoding of the Wikipedia talk example above. M /wiki/User:M P /wiki/User:P PKP /wiki/User:Pi Note that the synch attribute is provided by the module described in chapter .

Removing timestamps from the text body can help meet requirements of text anonymization. For instance, if the particDesc and the timeline are stored in a separate file, the rest of the corpus can be distributed without this separate file. Thus the recipient of the corpus may know in what order posts were made (if the values of the synch are sequential), and will be able to group posts made by the same user, but will not have exact timestamps or actual user names, thus providing a significant degree of anonymization.

Those haven't happened. If they do, we can revisit the concern.

[_DELETED-SIGNATURE_] [_DELETED-TIMESTAMP_] As demonstrated above, the synch attribute can be used on date or time (or indeed any other element) rather than on the post itself.

Emojis and Emoticons

Emojis are iconic or symbolic, invariant graphic units which the users of social media applications such as WhatsApp, Instagram, and X (Twitter) can select from a menu or emoji keyboard and embed into their written posts. Examples are 😁, 😷, 🌈, 😱, and 🙈. An emoji is encoded by one or more Unicode characters which are intended to be mapped directly to a pictorial symbol.

Emoticons predate emojis and are created as combinations of ASCII punctuation and other characters using the keyboard. Examples are :-), ;-), :-(, :-x, \O/, and Oo. They first occurred on a computer bulletin board system at Carnegie Mellon University (Fahlman, 2021) and then became frequent in chat communications during the mid-1980s. An emoticon typically consists of several Unicode characters (from the ASCII subset) in a row, each of which has an intended use other than as part of an emoticon.

Both emoticons and emojis may be simply transcribed as a sequence of characters. As with any other characters, they may be entered as numeric character entities if this is more convenient. (E.g., ❤ might be transcribed as in any XML document, including a TEI document; see .)

When the text of a post is being tokenized, e.g. for linguistic analysis, it may be useful to encode the emoticon or emoji as a separate token. In such cases elements such as w or c may be used for tokenization, and the pos attribute may be used to indicate that the encoded string is an emoji or an emoticon. (See .)

For example, the source post da bin ich nicht so empfindlich ;) (English:. I am not so touchy with that ;)) ends with an emoticon, and might be encoded as follows: da bin ich nicht so empfindlich ;)

Similarly, the source post Klar 😁 (Sure 😁 in English) might be encoded as follows: Klar 😁

The values of pos in the above examples are from the STTS_IBK Tagset for German (see ), which includes tags for CMC-specific elements such as EMOASC for an ASCII-based emoticon and EMOIMG for an icon-based emoji.

Alternatively, e.g. when w is not regularly used to encode tokens in the TEI document, c may be used to mark an emoji. For example, the source post Da kostet ein Haarschnitt 50 € 😱 (from the corpus , in English A haircut there costs 50 € 😱) might be encoded as follows: Da kostet ein Haarschnitt 50 € 😱

Sometimes, e.g. when the source of the TEI document was a web page in HTML, the emojis may occur only as an icon graphic in the source. In such a case, they may be encoded using figure. The corresponding Unicode character can then be recorded in the desc element by the encoder if desired.

For example, the source text: ... ich überlege noch 🙈 (English: ... I'm still thinking 🙈) might be encoded as follows: ... ich überlege noch

see no evil monkey U+1F648

Posts with Graphics

A post in a CMC interaction may contain a graphic in addition to some text or even contain only a graphic (without any text). As explained in , the modality of such a post should be considered as written. To encode the graphic information, the figure element may be used at the appropriate place.

In the following example a private chat post that contained only a screenshot of a google search result for a hairdresser is encoded as a post with a child figure. A link to the graphic file itself is not included presumably because this is a text-only corpus that does not include images.

screenshot of the google search for hairdresser "Pasha's Haare'm" with the average google rating (4,5 of 5 stars), the address, the phone number, and the opening hours.

The following is an example of the encoding of a tweet which contains both text (including hashtags and mentions) and a graphic. The graphic element retains the URL of the graphic on the web just as in the source. 16:24 Bro Tri-Engel...so hab ich mir das vorgestellt!!! @AndreLo79 #bcrn18 #wikidach @Heiko komm' mal Twitter! #Engel

Circulation

The following recommendations on how to encode features of the circulation of posts, such as IDs, re-posts (retweets), hashtags, and mentions use X (Twitter) posts (tweets) as an example; this phenomenon is not in any way unique to X (Twitter), however.

In the following example, the type of post (in this case, a tweet) is recorded using the type attribute of post. If it were useful to record a particular sub-categorization of tweet, the subtype attribute could also be used. Furthermore, the original unique identifer of the tweet as supplied by X (Twitter) is recorded as part of the value of the xml:id attribute of the post.

Also in the following example a retweet and its corresponding retweeted tweet are encoded as two separate posts each with its own set of attributes. The post representing the retweet itself does not contain or duplicate the content of the retweeted tweet. Instead it refers to the ID of the retweeted tweet via a ptr in the post content. All original content of the retweet goes in the content of the post element as well. In addition, the hashtags found in the body of the source tweets have been encoded using ref elements (with a type of hashtag), as they are links like any other hyperlink.

Ich mich auch? #dynamicduo #wirkümmernunsauchumIhrenEmpfang #bcrn18 #wikidach 14:35 Immer wieder gerne. Kann ich mich schon für nächstes Jahr als Empfangs- #Engel für das nächste BarCamp bewerben 🤪 #bcrn18

Note that in the above example CoMeRe style (cf. ) encoding is used to represent the number of favorites. It would also be reasonable to use a TEI measure element instead of the fs.

Linguistic Annotation

For encoding linguistic analyses of CMC text, we may use the dedicated elements and attributes from the analysis module, which is described in . For example, the tokenization (segmentation into word-like units) of a CMC text should be encoded using the w element.

Let us take, for example a posting that contains the content 00:22 Bin soooooo im stress gewesen ich Armer lol (in English: I was soooooo stressed out poor me lol). This may be encoded as follows. 00:22 Bin soooooo im stress gewesen ich Armer lol

In many CMC genres, especially in private chat, informal writing abounds including irregular spellings imitating spoken language, omitted word boundaries, and spurious boundaries leading to tokens separated in parts. For encoding these writing phenomena typical of CMC, the TEI attributes norm and join may be used.

For example, the normalized spelling of an irregularly spelled word may be recorded using the norm attribute (from att.linguistic): 00:22 Bin soooooo im stress gewesen ich Armer lol

When the boundaries between w elements are generally thought of as denoting word boundaries, we can keep track of boundaries not present in the source by using the join attribute, also from att.linguistic. For example, for an original post that has nothing more than the token Inmyoffice, the following encoding demonstrates an interpretation that the single token represents the three words In my office: In my office

Alternatively, and especially when the normalization information pertains to more than one token, we can apply the notation using the elements reg and orig, related by a choice element as described in . hastes hast du es

Other analysis attributes like lemma and pos (for part of speech) may be used as with traditional text. It is a matter of the tagset used to cater for POS categories that are appropriate for CMC. In the example below, for instance, the tag AKW stands for Aktionswort (action word, see ). 00:22 Bin soooooo im stress gewesen ich Armer lol

Named Entities and Anonymization

Named entities (NEs) may be marked up using name or the elements encoding different subcategories of names as described in such as surname or geogName, or rs for a general referencing string. In the following chat example (adapted from ), nicknames are linked to a person entry as shown in section via the ref attribute. Konstanze versucht nasenloch den wunsch zu erfüllen

In the following version of the same chat snippet, the text strings with the nicknames have been replaced by category label strings for the purpose of anonymization. [_FEMALE-PARTICIPANT-A04_] versucht [_PARTICIPANT-A04_] den wunsch zu erfüllen

In the preceding example, pairs of a gap and a supplied element encode the fact that some substring has been removed and replaced with another string for anonymization purposes. Note that in this example, the name and the w elements and their attributes also provide some categorical information about what has been removed. Using gap and supplied to record the anonymization is especially recommendable when the original name or referencing string has been pseudonymized, i.e. replaced by a different referencing string of the same ontological category (such as replacing the female name Konstanze by the female name Kornelia.). In that case, the markup would be the only place where it can be seen that a pseudonymization has been carried out, as in the following version of the example.

Kornelia versucht