Formulaic expressions are commonly used in administrative texts, and may re昀氀ect standardisation of the decision-making process or its recording process. In this paper we investigate whether the use of formulas in the Resolutions of the Dutch States General (1576-1796) reveal an increase in standardisation. We use stylometric analysis and measures of textual repetition to identify shi昀琀s in the use of formulas, and study how the fraction of paragraphs that is covered by formulas changes over time to identify templates consisting of frequent combinations of formulas. Our 昀椀ndings are that there are stylistically clearly distinguishable periods, and that the use of formulas and templates increases between subsequent periods.
Formulaic expressions have been used in legal procedures and decision-making processes for
millennia 4[
In this paper, we investigate whether the development of formulaic expressions in the
resolutions (decisions) of the States General of the Dutch Republic in the period 1576-1796 re昀氀ects
changes in the decision-making process and its recording. The archive of resolutions contains
all decisions registered during daily meetings for a 220 year per3io2,d2[
The digitised archive contains almost 500,000 pages of handwritten and printed resolutions in over 600 volumes1. Textual representations of these pages were made through HTR, a昀琀er which the text has been segmented into paragraphs and individual decisio2ns.The archive consists of almost 1 million decisions and 150 million words.
The resolutions of the States General (SG) have a stable structure, consisting of two parts: a proposition and a decision. The proposition describes what was proposed or requested and by whom, and the decision paragraph details whether the proposition has been accepted, and if so, under what conditions and what actions are to be taken next. Earlier research has shown that the resolutions contain many text structuring formula2s9,[ 21], including formulas to introduce the proposition and decision paragraphs (Figu1r).e
So far, it is not clear whether these formulas were used throughout the entire period, and whether they changed over time, and, if so, how emergence, change and disappearance of formulas was related to changes in the meeting procedures, the way the decisions were recorded or who recorded them.
In this paper, we focus on a diachronic analysis of the resolutions and how changes in the use of formulas is potentially related to standardisation, both of phrasing and of procedure. 1See https://www.nationaalarchief.nl/onderzoeken/archief/1.01.02 2See https://republic.huygens.knaw.nl/index.php/en/republic-engli.sh/
Our analysis consists of three parts. We start with a stylometric analysis to investigate if volumes written closer to each other in time are also stylistically more similar, and if there are periods of stable language use. Next, we compare the periods in terms of vocabulary growth of individual words and word n-grams to analyse how textual repetition changed over time. Finally, we zoom in on resolutions that consist mostly of commonly co-occurring words to identify whether they reveal signals of standardisation of decision making or recording.
We report a number of 昀椀ndings. First, clusters of stylistically similar volumes form contiguous periods, which in the 17th century correspond with changes between the gre昀케ers who were in charge of the creation and organisation of the resolutions. Second, the di昀erent periods have a similarly sized vocabulary of frequently used individual words, but later periods have a larger vocabulary of frequently repeated multi-word phrases, suggesting an increase in formulaic expressions. Third, the resolutions show an increase in the use of template structures for speci昀椀c types of propositions and decisions, that consist almost entirely of formulas in which only names, locations and dates varied.
Our 昀椀ndings suggest that the resolutions show an increasing standardisation, of either the decision-making process, the recording process, or both. We end with a discussion of the implications of our analysis and suggestions on how to link speci昀椀c characteristics of the corpus to these two processes.
Our work builds on di昀erent strands of research: research related to the nature of administrative and legal discourse and the role of formulaic expressions; the study of the linguistic characteristics of formulaic expressions and textual repetition; and study of the decision-making and recording processes related to the creation of the archive of resolutions. The latter is discussed in Section 3.
According to Gotti, textual standardisation is particularly strong in legal genres, as texts are o昀琀en adaptations of earlier texts, and because their precise formulation determines their legal force. Frade noted that the operative clauses in contracts remain stable as part of an attempt to “unify and harmonise legal rules1”3[, pp.49-50].
Formulaic expressions are commonly used in administrative documents to signal important
aspects of a document [
Gustafsson estimated that legal texts have higher repetition of multi-word phrases than texts in other genres. This 昀椀nding was corroborated by Koolen and Hoekstra who found that administrative texts have more textual repetition of especially phrases of 3 or more words, than other genres.
The identi昀椀cation of formulas in text corpora is studied, a.o., in the 昀椀eld of computational
phraseology 3[
The resolutions are the written record of the decisions taken at the meetings of the assembly in The Hague on weekdays and sometimes even on Sundays from 1593 until 1796.
They have been studied extensively 3[
Knowledge about the procedures during and a昀琀er the meeting relating to the decision-making and its recording in the form of the written resolutions is relevant for our analysis (for more details see AppendixA).
Gre昀ier transitions Over the 220 year period, 13 di昀erent people held the position of gre昀케er.
Each gre昀케er may have introduced his own preferences for spelling and vocabulary, and may have changed the existing formulas as well as introduced new formulas. It is plausible that formulas introduced by one gre昀케er were adopted by the next gre昀케er. If so, we would expect the amount of formulas used to gradually increase over time and not decrease.
Reduced spelling variation During the 17th and 18th centuries, Dutch spelling was not yet 椀昀xed resulting in spellingvariation (multiple spellings of the same word within a single volume) as well as spellincghange over time. However, the progressive tendency toward standardisation of spelling, also a昀ected standardisation of the surface form of formulaic phrasing. Furthermore, from 1705 the resolutions were printed, and have a lower character error rate in text recognition. The Character Error Rate (CER) for handwritten material is2 − 3% and for printed materia0l.5 − 1.0%. This a昀ects surface forms of expressions, so more variation is expected in the handwritten material due to recognition errors.
Increased time pressure Throughout the 17th century the daily number of decisions grew, resulting in increasing pressure to e昀케ciently minute and extend resolutions for approval the next day [37, Ch.4]. This may have lead the gre昀케er to introduce standard phrasings for the most repetitive elements of the resolutions. Procedural standardisation To improve the production, organisation and approval of resolutions, the SG introduced various changes to the procedur3e7,[pp. 436-437]: Introduction of resumption (1637) The gre昀케er read aloud all recorded resolutions for approval by the SG. From 1637, this was moved from the end of the meeting to the start of the next meeting, giving the gre昀케er more time to dra昀琀 the resolution [37, p.437].
Preparation by committees (1651) From 1651, committees of the SG increasingly prepared its administrative, organisational and ceremonial tasks to such an extent that discussing them in the meeting was reduced to a formality36[, p.162].
Extending by the commies (1680) Until 1680, the gre昀케er extended the minutes into full resolutions himself. A昀琀er 1680, this was taken over by the so-callceodmmies [32]. Strict protocols ensured that the resolutions re昀氀ected what was decided at the meeting. A modest manual comparison between minute and neat resolutions corroborates that there were indeed minimal deviations in word choice and word order, although there were individual di昀erences in spelling.
The corpus of resolutions that we used covers the period 1597-1796, minus the years 1651, 1703 and 1704, and consists of two parts. The handwritten resolutions of 1597-1702 come in the form of individual paragraphs (no resolutions boundaries have been detected yet). The printed resolutions of 1705-1796 are segmented at the level of resolutions (consisting of one or more paragraphs). For uniformity, we analyse both the handwritten and printed resolutions at the level of paragraphs.
A more extensive description of the selection of resolutions is given in AppenAdi.1x. The corpus for analysis consists of just over 1.5 million paragraphs and 125 million words.
One way to study the extent of language stability or change is through stylometric analy8s]is. [ Quantitative aspects of writing style that are indicative of authorship include high-frequency words and phrases 7[]. We use stylometry to identify sudden shi昀琀s in the style between two periods that could signal a change in spelling (from ’ae’ to ’aa’), in author (gre昀케er, commies or clerks), but also a change in recording the decisions (through introducing stable expressions for recurring aspects of decisions) or in procedural changes (the requirements of submitting proposals in written form in 1625, the introduction of resumptions in 1637, or the increased preparation by committees from 1651.
Burrows’ Delta 7[] is one of the most robust measures of textual distance and authorship
attribution [
We compute Cosine distances based on the top 1000 most frequent word 3-grams, for the resolutions of individual years. That is, we combined the text of all resolutions of one year, and compare the style of resolutions between years. The reason to use word 3-grams is that we want to identify transitions in the use of multi-word phrases. A heatmap of the cosine distances between all years is shown in Figur2e. The periods are ordered from 1597 to 1796 on the X- and Y-axis. Larger distances are darker red. The distances are lowest along the diagonal, which indicates that the resolutions written close to each other in time are more similar in style than resolutions written further apart in time.
Given that we do not know who the authors of the speci昀椀c resolutions are, we have no ground truth to evaluate how e昀ectively Burrows’ delta can cluster resolutions by author. What we can do is make hierarchical clusters and analyse: 1) how books of resolutions cluster temporally (e.g. do the books in the same cluster form a continuous period, or are books from di昀erent periods grouped together?), and 2) whether clusters overlap with periods of di昀erent gre昀케ers or periods of the same spelling preferences, or periods with the same formulas.
For the hierarchical clustering of the yearly resolutions we use SciPy’s hierarchy li3nkage
based on the Ward method 3[
Ideally, we would end up with clusters that group contiguous years, that is, the years in a cluster form one continuous period. For purposes of comparison, we want clusters that are neither very small (only 1 or 2 years) nor very big (grouping an entire century). We use SciPy’s fcluster function withdistance as metric and range over the cut thresho ldto 昀椀nd a suitable number of clusters.
The clusters show strong temporal groupings at any level of the clustering. With 200 years of resolutions, the details of the dendrogram with 200 leaves are hard to read, but it clearly shows there are two main clusters. The le昀琀 cluster in Figur3e contains all the years with handwritten volumes (1597-1702), and the right cluster all the printed volumes (1705-1796) (see Appendix B for the full dendrogram). It seems plausible that the additional step of typesetting the resolutions had an in昀氀uence on spelling and style. The vertical dashed line shows the threshold = 71 at which we cut the hierarchical clustering and 昀氀atten the sub-clusters within each cluster at that level. All clusters except one are contiguous. At lower thresholds, the number of clusters with non-contiguous periods increases. The one non-contiguous cluster has a contiguous part from 1671 until 1692, and one extra year, 1702. In the dendrogram, this cluster is next to the contiguous cluster 1693-1701, so the discontinuity is still stylistically close to the intermediate period.
For easier analysis, we use the contiguous parts of the clusters and move the year 1702 to its neighbouring cluster 1693-1701, so that it forms one contiguous period as well. The resulting cluster boundaries are shown as dashed lines in Figu9re
There are 11 clusters, covering periods ranging between 9 and 32 years (Tab1)l.e Some clusters overlap with the periods of one or more gre昀케ers, e.g. the cluster 1597-1628 overlaps with the gre昀케ers Cornelis van Aerssen en Johan van Goch. The next two clusters, 1629-1637 and 1638-1651 overlap with gre昀케er Cornelis Musch (1628-1650), and the change in the resumption procedure (1637) as well as the requirement that propositions of anyone outside the SG were submitted in writing and signed by the proposer (16248)I.n the 18th century, the cluster boundaries seem more related to changes in spelling than to transitions of gre昀케ers. 3See https://docs.scipy.org/doc/scipy/reference/cluster.hierarchy.html#module-scipy.cluster.hierarchy 4This was in fact already instructed by the SG in 1623, but was reiterated in 1628 and again in 1646, when the then gre昀케er Cornelis Musch, known at the time for his corruption, turned out to still handle many requests in person [37, p.461].
With the 11 style-based periods, we investigate changes in the use of formulaic expressions. We start with analysing vocabulary growth and repetition2. 1][ used vocabulary repetition ratios to show that administrative texts, and especially the resolutions of the SG, contain more frequently repeated phrases than other document genres.
We start from the same premise that textual repetition is a signal for the amount of formulas being used, and analyse if the amount of textual repetition decreases or increases over time, and if so, whether these changes correspond in time with any of the changes in gre昀케ers, meeting procedures or spelling, as identi昀椀ed in Tabl1e.
Instead of using repetition ratios, we use vocabulary growth curv2e,s 3[]. These curves show how the size of the vocabular y grows as the number of term tokens increases. The vocabulary ( ) contains all terms in the analysed text, but growth curves can be computed for di昀erent threshold frequencies. E.g. 1( ) is the growth of the vocabulary of terms that occur at least once a昀琀er observing tokens, while 5( ) shows the growth for terms that occur at least 昀椀ve times.
To create the growth curves for a given period, we put the paragraphs of that period in a random order and count term frequencies. For a curve fo(r ) , a昀琀er every 10,000 terms, we record the number of terms with a frequen c(y ) ≥ .
From a frequency-based de昀椀nition of formulas, we expect a change in the use of formulas to be re昀氀ected by a change in vocabulary growth, with higher usage of formulas leading to faster growth of high-frequency terms.
Terms can be individual words, but also word n-grams. Formulaic expressions are multiword phrases, therefore, like21[], we expect di昀erences in the use of formulas to be especially re昀氀ected by di昀erences in growth curves of multi-word terms.
The curves for the 11 style-clustered periods are shown in Figu4rfoer up to 1 million words per period5. The top plot shows 1( ) for word 1-grams, while the middle and bottom plots show 25( ) for, respectively, word 1-grams and word 5-grams. In the top plot, it is clear that the di昀erent periods have di昀erent vocabulary sizes, with the 18th century periods having smaller vocabularies than 17th century periods. Among the 17th century periods, there is no linear relationship between period and vocabulary growth. For the earliest period resolutions (1597-1628), the vocabulary growths fastest — perhaps re昀氀ecting the highest amount of spelling variation — but for the following period (1638-1650) the vocabulary growths slower than for all other periods of handwritten resolutions.
The middle plot shows that among the more frequent words, the di昀erences between the periods are smaller. This shows that across all periods, there is a similarly sized vocabulary of common words.
Finally, the bottom plot shows how many di昀erent 5-word phrases occur at least 25 times. Here, the di昀erences are clear. The earliest period has very few repeated 5-grams, while the periods of the printed editions in the 18th century have well over 1750 5-grams a昀琀er observing 1 million words. We note that some of these 5-grams are partially overlapping, because they are part of a single, longer formulaic expression. It is possible that textual repetition increases because the number of formulaic expressions increases, but also because formulas are getting longer. There is a large gap between the earliest period 1597-1628 and the next period 16291637, possibly indicating that the earliest period has few frequent formulas. Throughout the 17th century, the number of frequent 5-grams increases. The last period of handwritten resolutions, 1693-1702, is close to the printed resolutions in terms of 5-gram repetition. It seems 5The periods di昀er in the number of resolutions and the amount of text, but all have at least a few million words. For each period, the 昀椀rst 1 million ngram tokens from randomly ordered paragraphs is thus semi-random sample. that from 1693, the number of frequent 5-grams per 1 million words stabilises.
The changes in vocabulary repetition can also be viewed by taking a single data point per period, e.g. the vocabulary size at 1 million words. This is shown in Fig5u,rwehere the vocabulary size 25(1, 000, 000) is shown for 1-5 word n-grams for the 11 periods. The vocabulary of word 1-grams and 2-grams does not change much over 200 years, indicating that, despite di昀erences in the amount of spelling variation, each period has a similarly large vocabulary that is used in frequent expressions. For larger n-grams, the di昀erences in vocabulary size become more pronounced. Between 1600 and 1700, the amount of repetition steadily grows, and remains stable in the 18th century. These results make it unlikely that the reduction in spelling variation is the only cause for the increase in the use of long repetitive phrases.
Spelling stabilised throughout the 17th and 18th century, but there were also consistent changes, e.g. from ‘ae’ to ‘aa’ for long a-vowels as ivnerstaen vs. verstaan (EN: ‘understand’) and waerop and waarop (EN: ‘on which’). This is visible in the yearly relative term frequencies of variants of these terms, shown in Figur6e. The variants ofverstaan are shown at the top, and those ofwaarop at the bottom.
The 昀椀rst notable change in both plots in Figure6 is around 1628, when Cornelis Musch becomes gre昀케er, until 1650. Both verstaen and waerop are suddenly used much more frequently. This corresponds to a growth in the amount of textual repetition during this period (see Figure4). These words are part of the central decision formuwlaaerop gedelibereert sijnde is goetgevonden ende verstaen dat (EN: ‘which, a昀琀er deliberation, it has been accepted and understood that’, see Figure1), that is introduced in the resolutions around the time when Musch takes over the role of gre昀케er. The variant verstaen is more or less the only variant until brie昀氀y in 16701672, when there are two transitions of gre昀케ers, once in 1670 and again in 1672. With Gaspar Fagel as gre昀케er in 1670-1672, both verstaen and verstaan were used. With the next gre昀케er,
Hendrik Fagel de Oudste,verstaen is used exclusively. Then, in 1690, when François Fagel takes over, verstaan is used more frequently thanverstaen. Their order swaps again around 1700 until 1717, when the use ofae is discarded andverstaan is used exclusively. For variants of waarop, swaps in their relatively frequency occur at the same moments in time, suggesting that these spelling changes have the same underlying causes (perhaps change of gre昀케er or commies). Forwaarop there is another source of variation, which is the segmentation of function words. The termwaarop can also be spelled as two wordwsaar op. For these variants, there is one period between 1690 and 1705, where all four variants occur. Again, from 1717, only a single variant is used. For many key terms in the resolutions, the distributions show similar shi昀琀s from almost exclusive use of one variant to almost exclusive use of another. This suggests that spelling variation of key terms was minimal, which is further evidence that the increasing repetition of long phrases is not only due to increasing spelling standardisation.
If the recording of resolutions standardised over time, we would expect to see resolutions to increasingly consist of frequent formulas or phrases. To idenctiafnydidate formulas, we used a frequency threshold to extract sequences of commonly co-occurring wor2d1]s.[We use these sequences to analyse what fraction of a resolution consists of succahndidate formulas.
The eleven periods identi昀椀ed in Section4 have di昀erent amounts of text. To make a fair comparison across these periods, we use frequency thresholds that are relative to the amount of text in each period. For the minimum co-occurrence frequency of word pairs, we use a threshold of once per 100,000 word tokens. That is, a continuous sequence of words is considered acandidate formula if each pair of words in the sequence co-occurs at least once per 100,000 words. This threshold was chosen on practical grounds. Per period, this corresponds to a minimum term frequency between 45 (for period 1629-1637 with 4.5 millions words) and 172 (for 1736-1764 with 17.2 million words). This means all low-frequency terms (including most terms with recognition errors) are removed, while retaining a large enough vocabulary to identify formulas containing mid-frequency terms, if they mostly occur in the same context.
Next, we compute the fraction of words in each paragraph that are partcoafndidate formulas. For a paragraph of 100 words with 昀椀ve candidate formulas of ten words each, thcoeverage of a paragraph by candidate formulas is 50% of the text. With this de昀椀nition ocfoverage, we look at its distribution for the eleven periods (see Figu7r)e, for both the handwritten and printed resolutions (top and bottom plot respectively). There are clear di昀erences between periods. For the earliest period, the bulk of the paragraphs have a coverage below 0.4. For later periods, the bulk is shi昀琀ed to the right, with the exception of 1638-1650 and 1652-1670. These two periods have a high number of paragraphs with zero coverage by candidate formulas. These are mostly very short paragraphs of a few words due to incorrectly inserted paragraph boundaries. In the printed resolutions, the paragraph coverage by candidate formulas is higher, with the bulk having a coverage between 0.6 and 0.9, and a peak around 0.9. Throughout the 18th century, the peak at 0.9 gets higher, with the exception of the last period 1787-1796. Note that the last period of handwritten resolutions (1693-1702) has a distribution similar in shape to those of the printed resolutions, but with a less pronounced peak.
These results suggest that over time, an increasing part of paragraphs became formulaic. This prompts several questions. For instance, what is the nature of paragraphs that consist almost exclusively of candidate formulas? And does the gradual increase of candidate formulas correspond to increasingly standardised resolutions?
We zoom in on the paragraphs with highcoverage, for four of the eleven periods, to see how the nature of these paragraphs changes.
First, we analyse whether high-coverage paragraphs are di昀erent from other paragraphs by
identifying key phrases. We turn each candidate formula into word 5-grams and count their
frequencies. Then we compare their frequencies in high-coverage paragraphs against their
frequencies in all paragraphs in a given period. Fkoeryness we compute the log likelihood ratio
[10] and percentage di昀erence [
To 昀椀nd out what changes, we zoom in further, and select paragraphs where thkeey phrases cover at least 30% of paragraphs. We then make TF-IDF representations of these paragraphs and cluster them using K-means, wit h= 20 . The reason to cluster paragraphs is to 昀椀nd out if there are speci昀椀c types of paragraphs with speci昀椀c functions and consisting of relatively 昀椀xed sets of formulas. TheElbow and Silhouette measures both indicate tha t = 3 gives the optimal clustering, but we 昀椀nd that higher values of K result in a better separation of di昀erent resolution templates. But they also lead to multiple clusters with the same formulas and functions (which are merged with = 3 ).
The clustering was done for each of four periods 1597-1628, 1671-1692, 1717-1735 and 17651786. Based on manual inspection of the results, we created sets of regular expressions for each identi昀椀ed template and classi昀椀ed the resolutions in each of the clusters. Most clusters are highly homogeneous and consist of paragraphs with a single stable template, while some clusters are mixtures of multiple templates and of paragraphs that do not seem to follow a template structure.
With = 20 , many of the clusters contain paragraphs with small variations of the same template (which are grouped into a single cluster when we cho os=e 3 ), that has the following main form:
Ontfangen een Missive van <PERSON>, geschreven te <LOCATION> den <DATE>, houdende advertentie, Waar op geen resolutie is gevallen. (EN: Received a Missive of <PERSON>, written in <LOCATION> on <DATE>, containing intelligence, on which no resolution was made.)
Another large and stable group consists of so-called resumption paragraphs, which appear at the start of each session to indicate that the written versions of the resolutions of the previous day have been read and approved:
DE Relolutien gisteren genomen, zyn geleesen en geresumeert, gelijk ook geresumeert en gearresteert zyn de Depeches daar uit resulteerende. (EN: The resolutions made yesterday are read and summarised, as well as the depeches that follow from them.
With this paper, we aim to shed light on the use of formulaic expressions and templates to standardise the decision-making process of the States General of the Dutch Republic, and of the recording of the decisions resulting from it.
The 200 years of recorded decisions of daily meetings form an invaluable resource to study these processes, but come with many challenges as well. Over this long period, many aspects of both processes changed, and it is di昀케cult to establish how these changes are re昀氀ected in or have in昀氀uenced the creation of the written resolutions.
Our stylometric analysis suggests that the resolutions went through several stylistic changes that align in the17ℎ century with transitions between gre昀케ers who were in charge of recording and organising the resolutions, and in th1e8ℎ century by changes in spelling conventions.
The analysis of vocabulary growth of individual words and word 5-grams shows that over time, the amount of textual repetition of multi-word phrases increases, suggesting an increasing use of formulaic expressions. 6See Appendix C for examples of each type.
Finally, our analysis of paragraphs that consist almost entirely of frequent multi-word phrases reveal a set of template paragraphs, and show that both the number of template types and the frequency of their usage increased over time.
Our 昀椀ndings do not directly identify to what extent this standardisation is a re昀氀ection of the decision-making process, or of its recording, or of both. It is possible that changes in one of the processes in昀氀uenced the other. It is also possible that, with what written record has survived/been handed down (resolutions or otherwise), it is impossible to separate these two and identify more precise causal factors.
Our 昀椀ndings prompt further questions in this direction. For instance, to what extent can the introduction of speci昀椀c formulas or templates be attributed to individual gre昀케ers? Can we distinguish the in昀氀uences of the gre昀케er and other people/clerks involved by comparing in more detail the minutes and neat resolutions, and the di昀erent series of copies that have been archived?
Our 昀椀ndings only partially align with earlier analysis by Riemsdijk and Thomassen,
Thomassen, who disagree with each other on whether the commies took over the extending
of minutes to full resolutions from the gre昀케er in 16803[
This research is funded by the Dutch Research Council (NWO) through the NWO Groot project REPUBLIC (an acronym for REsolutions PUBlished In a Computational Environment) 20192024.
We thank the volunteers for their invaluable contributions to this project.
H. De Boor.Actum et datum: eine Untersuchung zur Formelsprache der deutschen Urkunden im 13. Jahrhundert. 4. Verlag der Bayerischen Akademie der Wissenscha昀琀en, 1975.
Although representatives of the regions that made up the Netherlands had been meeting since 1464 to provide advice and money to the sovereign, the States General only really became a political factor of importance in the context of the Dutch Revolt against the Spanish king Philip II. The States General developed into a political body in which the rebellious provinces–昀椀rst including the southern, but from 1585 only the northern regions that became the Republic of the Seven United Netherlands–negotiated and decided upon their common interests, such as foreign policy, defence, and religious matters. The resolution registers are the written record of the decisions taken at the meetings of the assembly, which resided in The Hague from 1588 onwards, and met practically on all weekdays and sometimes even on Sundays from 1593 until 1796.
In the course of the period under study, various administrative customs developed which are important for the analysis of the resolutions. The gre昀케er attended the meetings and had to record the decisions taken. He 昀椀rst noted the conclusion and ‘extended’ it, until 1637 still during the meeting, into a resolution by preceding it with the proposition. He was not allowed to include the intermediate discussion unless a con昀氀ict ran high and one of the provinces explicitly wished their position to be recorded. The gre昀케er read the resolutions, or just the ‘disposition’ (the decision), thus dra昀琀ed for approval by the assembly. A昀琀er 1637, the reading and approval (the ’resumption’) was moved to the beginning of the next meeting, usually the following day. As a result, more attention was paid to the dra昀琀ing of resolutions, leading to more standardisation. The procedure of dra昀琀ing resolutions involved other o昀케cials, who gradually took on a more important role as the administrative burden increased. The extension of some resolutions was done by a clerk a昀琀er 1637, under the supervision of the gre昀케er. In 1637, the States General appointed acommies, in some periods calledsecond gre昀케er , who replaced the gre昀케er in his absence. From 1680 the commies/second gre昀케er took over the extension of the ordinary resolutions entirely from the gre昀케er, sometimes with the help of a clerk. A昀琀er 1744, the extension was le昀琀 entirely to a clerk. Of importance for the analysis of the ordinary resolutions is also that in our project we do not use the minutes of the resolutions, but the written neat copies until 1703, and therea昀琀er the printed copies. The reason for using the printed copes where available, is that automatic text recognition gives the better results for printed than for handwritten text. The neat copies were written by clerks. A small-scale manual comparison between the minute and neat resolutions revealed that the latter are a very faithful copy of the minutes, but the spelling of individual words sometimes di昀ers.
Obviously, the involvement of di昀erent o昀케cials in the dra昀琀ing of the resolutions, and the fact that the neat copies were written by clerks, has implications for our analysis. A昀琀er all, the clerks form an additional layer between what the gre昀케er and commiezen dra昀琀ed and the neat copies we used. Nevertheless, our analysis does show a correlation between variations in the resolutions and the periods when di昀erent clerks were in o昀케ce when it comes to word usage and spelling. Future research could also include the minute-resolutions in the analysis and possibly distinguish between the di昀erent handwritings of gre昀케ers, commiezen and clerks.
In addition to the ordinary resolutions, there was also a series of secret resolutions in which the gre昀케er noted down resolutions that were too sensitive to be included in the ordinaris register. Unlike the ordinary resolutions, the gre昀케er continued to extend these himself throughout the period, obviously keeping them as secret as possible. Nor were these resolutions printed like the ordinaris resolutions from 1703 onwards. As we do not currently have a proper paragraph separation for the secret resolutions, these could not yet be included in the analysis.
The full corpus of resolutions consists of 220 years of decisions recorded in over 600 volumes of handwritten and printed text. As with any large archive that was created over centuries, its creation process was complex and went through many changes.
Initially, there was only one series of minute resolutions, which consisted of the dra昀琀 resolutions with additions or amendments ordered by the assembly. From 1578 onwards, these ‘messy’ minutes were copied in a neat register, creating a second series of ordinary resolutions, or the 昀椀rst series of neat resolutions. The introduction of the resumption in 1637 created a third series. The gre昀케er’s dra昀琀 resolutions served as the basis of the resumption. Once this dra昀琀-with amendments-was approved by the assembly, it was copied as the second series of neat resolutions. In the mean time, the custom of having another neat copy produced remained. From 1671, there was even a fourth series produced. For our analysis, we use the 昀椀rst set of neat resolutions up to 1637, and from there the second set of neat resolutions.
We were able to obtain the digital textual representations, transformed from lines of text on physical pages into running paragraphs where word-breaks have been resolved, for the 200 year period 1597-1796, with the exclusion of 1651, 1703 and 1704. For these years, the HTR process generated no usable output (yet).
The individual resolutions have been identi昀椀ed in the printed volumes–covering the period 1705-17967–by [22]. This corpus has 286,340 resolutions. The handwritten volumes we obtained, have not been segmented into individual resolutions, but 昀椀rst into individual sessi8o,ns and into individual paragraphs. To have a more uniform corpus for our analysis, we removed 7There are printed copies of earlier volumes of resolutions, by they are archived separately, and have not gone through the OCR pipeline yet. 8The resolutions were record chronologically, with the resolutions of each day preceded by the date and a list of attendants of the meeting. the date and attendance lists of the handwritten resolutions and cut t1h8eℎ century resolutions into individual paragraphs as well. The total corpus has just over 125 million words.
We attempted algorithmic spelling normalisation, using a combination of FastText word
embeddings [
We experimented with a combination of techniques to normalise spelling, for two reasons. First, to study how spelling variation a昀ects the frequency of multi-word phrases, and second, to bring the di昀erent periods closer together in spelling, or at least in amount of spelling variation.
We use FastText word embeddings 4[] to 昀椀nd spelling variants that occur in similar contexts. Some spelling changes and variations are sudden and happen with whole groups of words. We use a list of character ngram pairs and a list of context-sensitive rewrite rules to 昀椀nd spelling variants. We compute word 2-gram frequencies per 5-year period and check if the merger of each word 2-gram is more frequent in di昀erent periods. This results in a list of unigram and bigram pairs that can be algorithmically normalised (e.g. always use the unigram or bigram, or always use the most frequent of the two. All these term pair lists were manually curated to remove errors.
Although this worked well for the later periods, we found that it captured only a small fraction of the variation for the earliest periods. Given these 昀椀ndings, we perform all the analyses in this paper without spelling normalisation.
The threshold value for cutting the hierarchical clustering and 昀氀atten the sub-clusters was set to = 71 for two reasons. First, at that level, the clusters are relatively close to each other in size in number of years per cluster (between 9 and 30 years) and the number of clusters (11) is small enough to make manual comparison feasible.
Using more conventional criteria like silhouett3e4][ would have resulted in two clusters, one for the handwritten resolutions of 1597-1702 and one for the printed resolutions of 17051796. This is by far the strongest division. The thresho ldwhere a single cluster splits into two clusters is = 323 , and the threshold between a two cluster and a three cluster spl it=is213 .
The second reason is that we want the number of non-contiguous clusters to be minimal. At = 71 , all but one of the clusters represent contiguous periods, and there is only a single year (1702) that is grouped with a contiguous period (1671-1692). The closest cluster to that one in terms of cosine distance is the intermediate period (1693-1701). All lower (integevra)lues result in multiple temporal discontinuities in the clusters.
Below are example paragraphs for each of the template types. The formula combinations are underlined. Variable slots that require a named entity of a speci昀椀c type are indicated with <VAR>.
Template type: No resolution.
ONtfangen een Missive van <PERSON>, geschreeven <LOCATION> den <DATE>, houdendeadvertentie. WAAR op geen resolutie is gevallen.
ONtfangen een Missive van <PERSON>, geschreeven te <LOCATION> den <DATE>, geaddresseert aan den Gri昀케er Fagel, houdende advertentie. WAAR op geen resolutie is gevallan.
ONtfangen een Missive van <PERSON>, geschreeven te <LOCATION> den <DATE>, houdendeadvertentie. WAAR op geen resolutie is gevallen.
ONtfangen een Missive van <PERSON>, geschreeven te <LOCATION> den <DATE>, houdendeadvertentie. WAAR op geen resolutie is gevallen.
Template type: The resolutions taken yesterday have been read and approved.
DE Relolutien gisteren genomen, zyn geleesen en geresumeert, gelijk ook geresumeerten gearresteertzyn de Depeches daaruit resulteerende.
DE Resolutlengisteren genoomen, zyn geleesen en geresumeert, gelijkook geresumeerten gearresteertzyn de Depeches daaruit resulteerende.
DE Resolutien gisteren ge. nomen, zyn geleesen en geresumeert, gelijkook geresumeerten gearresteertzyn de Depeches daaruit resulteerende.
DE Resolustiengisteren genoomen, zyn geleesenen geresumeert,gelijkook geresumeerten gearresteertzyn de Depeches daaruit resulteerende.
Template type: requesting the SG to send a bill to the Generaliteits Reekenkamer (EN: Chamber of Finance):
IS ter Vergaderinge geleesen de Requeste van <PERSON>, presenteerende daar nevens aan haar Hoog Mogende sijne Declaratievan Daggelden en Verschotten <PERIOD>; versoekende dat haar Hoog Mog. deselve ten 昀椀ne van examinatie en liquidatie gelieven te senden aan den Raad van Staate en de Generaliteits Reekenkamer.
IS ter Vergaderinge geleesen de Requeste van <PERSON>, presenteerende daar nevens aan haar Ho g Mo. sijne Deelaratievan Tractementen en Verschotten, welke aan hem in voorschreeve qualiteit nog waren competeerende, <PERIOD>; versoekende derhalven, dat haar Hoog Mog. deselve ten 昀椀ne van examinatie en liquidatie gelieven te senden aan den Raad van Staate en de Generaliteits Reekenkamer. IS ter Vergaderinge geleesen de Requeste van <PERSON>, presenteerende daar nevens aan haarHoogMog. sijne Declaratievan Taactementen, Daggelden en Verschotten, <PERIOD>; versoekende dat haarHoog Mog. deselveten 昀椀ne van examinatie en liquidatie, gelievente senden aan den Raad van Staate en de Generaliteis Reekenkamer.
Template type: acceptance to send a bill to the Generaliteits Reekenkamer (ENC:hamber of Finance).
WAAR op gedelibereertzynde, is goedgevonden en verstaan, dat de Declaratie nevens de voorsz <DOCUMENT> gevoegt,gesonden sal worden aan den Raad van Staate en de GeneraliteitsReekenkamer, om te visiteeren, examineeren en liquideeren,volgens en in conformiteit van de ordresvan het Land.
WAAR op gedelibereertzynde, is goelgevonden en verstaan, dat de Declaratie, neevens de voorschreeve <DOCUMENT> gevoegt,gesonden sal worden aan den Raad van Staate en de GeneraliteitsReekenkamer, om te visiteeren, examineeren en liquideeren,volgens en in conformiteit van de ordresvan het Land.
WAAR op gedelibereertzynde, is goedgevonden en verstaan, dat de Declaratie nevens de voorsz <DOCUMENT> gevoegt,gesonden sal worden aan den Raad van Staate en de GeneraliteitsReekenkamer, om te visiteeren, examineeren en liquideeren,volgens en in conformiteit van de ordresvan het Land.
WAAR op gedelibereertzynde, is goedgevonden en verstaan, mits deesen te versoeken de Heeren Gedeputeerden van de Provincien, welkesig op het voorszsubject nog niet hebben verklart, hun daartoe meede te willenbekwamen.
WAAR op gedelibereertzynde, is goedgevonden en verstaan, mits deesen te versoeken de Heeren Gedeputeerden van de Provincie van <PROVINCE>, welke sig op het voorschreevesubject nog niet hebben verklaart,hun daar toe meede te willenbequamen.
DE Heeren Gedeputeerden van de Provincie van <PROVINCE>, hebben ter Vergaderinge ingebragt en laaten leesen de Resolutie van de Heeren Staaten van hooggemeldeProvincie hunne Principaalen, waar by hoogstdeselve de gemelde
DE Heeren Gedeputeerden van de Provincie van <PROVINCE>, hebben ter Vergaderinge ingebragt en laaten leesen de Resolutie van de Heeren Staaten van hooggemeldeProvincie hunne Principaalen, waar by hoogstdeselve consenteeren in de door het Collegie ter Admiraliteit in Westvriesland en het Noorder Quartier versogte tweede Negociatie van 6ococo guldens ter voldoening hunner agterstallen;volgende de voorszResolutie hierna geinsereert.
DE Heeren Gedeputeerden van de Provincie van <PROVINCE>, hebben ter Vergaderinge ingebragt en laaten leesen de Resolutie van de Heeren Staaten van hooggemeldeProvincie hunne Principaalen, waar by hoogstdeselve alsnog deelareeren, het Corys van den Rhyngrave van Salm met den 7 Mey laatstleeden, te houden voor afgedankt, soo nogtans, dat aan het selve nog zullen worden betaald twee maanden Soldye, ingaande met den 7 Mey en 18 Junvv;olgende de voorschreeveResolutie hierna geinsereert.
DE Heeren Gedeputeerden van de Provincien van <PROVINCE> en van <PROVINCE>, hebben ter Vergaderingeingebragt en laaten leesen de Resolutien van de Heeren Staaten van hooggemeldeProvincien hunne Principaalen, waar bv hoogstdeselve sig conformeeten met het Rapport den 24 Augusty laatstleeden alhier ter Vergaderinge uitgebragt, over het getal der Leeden, dewelke souden dienen te worden benoemt in de bekende saak der niet geexccuteerde Expeditie na Brest; volgende de voorschreeveResolutie hierna geinsereert.