For corpus query and analysis the RSC is encoded in CQP format (cf. IMS Open Corpus Workbench (CWB)). The CWB requires a simple XML as an input format. In the so-called vrt-format (vertical text format) annotations on the token level (positional attributes, e.g. word, pos, lemma) are represented in a one-word-per-line with TAB deliminated columns for each positional attribute. Annotations beyond token level (structural attributes, e.g. texts, sentences, pages) are represented as XML-tags with possible attribute-value pairs. Metadata, e.g., are encoded as attributes of the <text>
-element. In the following we give a detailled overview of annotations on token level (positional attributes) and on structural level (structural attributes).
Attributes are listed in the order of the columns in the VRT file. Attribute names refer to the positional attributes encoded in the online corpus.
word pos lemma orig srp srp_avg srp_rnd srp_avg_rnd doc doc_avg doc_rnd doc_avg_rnd s50 s50_avg s50_rnd s50_avg_rnd s10 s10_avg s10_rnd s10_avg_rnd
of IN of of 0.11 1.621 0 2 0.04 0.367 0 0 0.06 2.023 0 2 0.10 2.214 0 2
some DT some some 3.49 6.976 3 7 2.11 0.861 2 1 2.66 6.306 3 6 2.03 6.387 2 6
Books NPS Books Books 3.48 8.743 3 9 0.65 1.500 1 2 0.90 6.109 1 6 0.88 5.479 1 5
Positional Attribute | Description |
---|---|
word |
Normalized word form (VARD) |
pos |
Part-of-speech tag (Penn Treebank Tagset) |
lemma |
Lemma, according to TreeTagger |
orig |
Original word form |
srp |
Surprisal |
srp_avg |
Average surprisal |
srp_rnd |
Surprisal (rounded) |
srp_avg_rnd |
Average surprisal (rounded) |
doc |
Document surprisal |
doc_avg |
Average document surprisal |
doc_rnd |
Document surprisal (rounded) |
doc_avg_rnd |
Average document surprisal (rounded) |
s50 |
Surprisal on 50-year periods |
s50_avg |
Average surprisal on 50-year periods |
s50_rnd |
Surprisal on 50-year periods (rounded) |
s50_avg_rnd |
Average surprisal on 50-year periods (rounded) |
s10 |
Surprisal on decades |
s10_avg |
Average surprisal on decades |
s10_rnd |
Surprisal on decades (rounded) |
s10_avg_rnd |
Average surprisal on decades (rounded) |
Structural attributes are given in two different representations:
Metadata are encoded on the text level.
<text id="" issn="" title="" fpage="" lpage="" year="" volume="" journal="" author="" type="" corpusBuild="" doiLink="" language="" jrnl="" decade="" period="" century="" pages="" sentences="" tokens="" visualizationLink="" doi="" jstorLink="" isAbstractOf="" hasAbstract="" primaryTopic="" primaryTopicPercentage="" secondaryTopic="" secondaryTopicPercentage="">
Example:
<text id=“100997” issn=“03702316” title=“An Extract of a Letter Written by Dr. Edward Brown from Vienna in Austria March 3. 1669. Concerning Two Parhelia's or Mocksuns, Lately Seen in Hungary” fpage=“953” lpage=“953” year=“1669” volume=“4” journal=“Philosophical Transactions (1665-1678)” author=“Edward Brown” type=“fla” corpusBuild=“5.2” doiLink=“http://dx.doi.org/10.1098/rstl.1669.0015” language="" jrnl=“transactions” decade=“1660” period=“1650” century=“1600” pages=“1” sentences=“10” tokens=“245” visualizationLink=“http://corpora.clarin-d.uni-saarland.de/surprisal/6.0.3/?id=100997” doi=“10.1098/rstl.1669.0015” jstorLink=“http://www.jstor.org/stable/100997” isAbstractOf="" hasAbstract="" primaryTopic="Reporting" primaryTopicPercentage="47.2589599153001" secondaryTopic="Astronomy" secondaryTopicPercentage="34.1265051578955">
Structural Attribute | Description |
---|---|
<text_author> |
Author of the article |
<text_century> |
Century of publication |
<text_corpusBuild> |
Internal version number |
<text_decade> |
Decade of publication |
<text_doi> |
DOI of article |
<text_doiLink> |
Link to DOI resolver |
<text_fpage> |
First page of the article |
<text_hasAbstract> |
ID of the corresponding abstract |
<text_id> |
JSTOR ID |
<text_isAbstractOf> |
ID of the corresponding article |
<text_issn> |
ISSN of the journal |
<text_journal> |
Journal in which the article was published |
<text_jrnl> |
Journal abbreviation |
<text_jstorLink> |
Link to the source text on JSTOR |
<text_language> |
Language of article |
<text_lpage> |
Last page of the article |
<text_pages> |
Number of pages in text |
<text_period> |
50-year period of publication |
<text_primaryTopic> |
The most prominent topic according to our topic model |
<text_primaryTopicPercentage> |
Percentage of the most prominent topic in the text |
<text_secondaryTopic> |
The second most prominent topic according to our topic model |
<text_secondaryTopicPercentage> |
Percentage of the second most prominent topic in the text |
<text_sentences> |
Number of sentences in text |
<text_title> |
Title of the article |
<text_tokens> |
Number of tokens in text |
<text_type> |
Text type |
<text_visualizationLink> |
Link to visualization |
<text_volume> |
Volume of the article |
<text_year> |
Year of publication |
<page id="" no="" tokens="">
Structural Attribute | Description |
---|---|
<page> |
Page (attribute from JSTOR) |
<page_id> |
Absolute page number |
<page_no> |
Relative page number |
<page_tokens> |
Number of tokens in page |
Texts are split into sentences based on the output of the TreeTagger.
<s srp="" doc="" s50="" s10="" no="" tokens="">
Structural Attribute | Description |
---|---|
<s> |
Sentence boundary (based on SENT tags of TreeTagger) |
<s_srp> |
Average surprisal of sentence based on srp |
<s_doc> |
Average surprisal of sentence based on doc |
<s_s50> |
Average surprisal of sentence based on s50 |
<s_s10> |
Average surprisal of sentence based on s10 |
<s_no> |
Relative sentence number (within a text) |
<s_tokens> |
Number of tokens in sentence |
Normalised words are represented on the token level and on the structural level to account for one-to-many relations (e.g. ’tis \(\rightarrow\) this is, my self \(\rightarrow\) myself) on the one hand and to allow for an easy access on the other hand.
<normalised orig="" auto="">
Structural Attribute | Description |
---|---|
<normalised> |
Normalised token(s) |
<normalised_auto> |
Always “true” as all normalisations are automatic |
<normalised_orig> |
Original token(s) |
The element <inferred>
is part of the JSTOR distribution. It refers to illegible text which was recovered from the context.
For more information on annotation and metadata, please consult our paper in RiCL:
Katrin Menzel, Jörg Knappe, and Elke Teich (2021): "Generating linguistically relevant metadata for the Royal Society Corpus", Research in Corpus Linguistics 9(1):1-18, DOI: 10.32714/ricl.09.01.02