Since my last post on research using with the Catalogue of Political and Personal Satires in the British Museum (hereafter BMSat) as a dataset I’ve been working on various threads, including Named Entity Extraction (I hope to write about this in my next blog), extracting the words used when figures in the prints are ‘speaking’ (with variable success), and wrangling my initial thoughts, reflections, and experiments into something that has started to resemble publishable research (or at least something I’m happy to talk about in a research paper setting).

One thread I’ve also been pursuing with some success has been pulling together data on what people are doing in the prints (or least what BMSat as a proxy for the prints themselves says they are doing – and before you worry, yes I am working on how the results from analysis of this proxy maps to the original sources). When we look at the fifty most common lemma words for all the satires dated 1770-1830 (covering all of the entries made by Mary Dorothy George) the doing words that stand out are:

Rank    Lemma    Count (total 1123450)
5    stand    9847
7    wear    8096
10    say    6991
11    hold    5722
12    holding    4984
15    look    4564
18    dress    4388
20    sit    3698
22    saying    3211
44    seat    2234
48    lie    2119
49    point    2080

Now these aren’t perfect: a strong lemma for lie is ‘lying’ (count: 209) which – of course – has meanings both in sense of lying down on the floor and not telling the truth about stuff. Nevertheless, what is striking here is that aside from standing about and wearing things (which both animate and inanimate objects can do) there is lots of saying going on in the descriptions in BMSat. Indeed the combined total occurrences for the lemma say (said, say, says) and saying (saying, sayings) is 10202, the second highest of any action lemma (that is if we also combine hold and holding, which I do for the remainder of this post).

If we break down these occurrences of both the say and saying lemmas by decade, convert them from raw counts into percentage occurrence across the total words for the each decade (so `(word total / total words) x 100) we note that uses of the say lemma in BMSat peak in the 1780s, fall away in the 1790s and 1800s, and regain strength in the 1810s. We also note that the trajectory of the say lemma tracks two other lemmas: first 1770-1789 ‘inscribe’ and thereafter ‘wear’. Now as these percentages are by decade the data is very smoothed and should only be used to draw very preliminary conclusions (including, though, the conclusion I need to spend some time working on year by year percentages for the action words! – research as ever begets research…) *nevertheless* this data makes some sense: the initial tracking of ‘say’ and ‘inscribe’ map to what we know about the change in style of satirical prints in the 1770s-1780s from prints with legends to prints with speech bubbles; and it isn’t surprising that Dorothy George describes in BMSat about what characters are wearing and saying in relatively equal measure: these are, after all, the distinguishing characteristics of figures in Georgian satirical prints.

To test whether spending more time creating fine-grained action data to better understand the say lemma is going to be worth it, I’ve also spent some time looking at the most common five words either side of ‘said’, ‘say’, ‘says’, and ‘saying’ (hereafter just ‘say’) in BMSat; otherwise described as 5L5R collocates. Again, the results make a lot of sense. Below I’ve flagged with ** the characters (both specific and non-specific) that appear to be speaking and/or spoken about based on this data:

Rank    Total    5L    5R    Word
1    529    447    82    right
2    451    411    40    left
3    404    25    379    ll
**4    377    268    109    man**
5    292    12    280    oh
6    291    243    48    hand
7    259    21    238    shall
8    254    13    241    come
9    255    245    10    looks
10    257    255    2    inscribed
**11    245    197    48    fox**
12    239    13    226    dear
13    236    183    53    behind
14    226    194    32    head
15    205    193    12    holding
**16    198    77    121    lord**
17    199    188    11    holds
18    192    173    19    arm
**19    191    135    56    john**
20    188    174    14    hands
21    190    184    6    hat
22    176    159    17    stands
23    162    152    10    shoulder
**24    164    138    26    king**
25    154    3    151    ah
26    152    9    143    let
**27    154    72    82    bull**
**28    151    27    124    sir**
**29    151    121    30    pitt**
30    148    18    130    say
31    146    27    119    good
**32    145    20    125    mr**
33    140    19    121    dont
34    138    44    94    old
35    132    5    127    ha
36    135    104    31    looking
37    125    4    121    aye
38    132    107    25    says
39    122    115    7    arms
40    118    30    88    little
**41    119    101    18    woman**
**42    112    5    107    gentlemen**
43    118    117    1    profile
44    110    15    95    poor
45    107    6    101    ye
46    105    19    86    like
47    103    30    73    away
48    103    8    95    make
49    106    70    36    long
**50    103    91    12    napoleon**
**51    100    19    81    master**
52    103    74    29    cf
53    90    83    7    glass
54    88    85    3    turns
55    87    5    82    think
56    86    68    18    north
57    85    23    62    look
**58    85    48    37    devil**
**59    83    23    60    friend**
60    86    77    9    face
61    82    81    1    points
62    84    78    6    round
63    81    64    17    raised
64    86    79    7    large
65    82    75    7    paper
**66    80    45    35    lady**
**67    78    64    14    duke**
68    75    48    27    pointing
69    74    1    73    wish
70    74    37    37    eyes
**71    78    56    22    men**
**72    74    31    43    boy**
73    74    67    7    pocket
**74    72    12    60    brother**
75    72    39    33    smile
**76    73    54    19    mrs**
77    71    16    55    sic
**78    71    53    18    prince**
79    72    59    13    wig
80    71    70    1    sword
81    69    8    61    know
**82    69    20    49    jack**
83    71    66    5    extreme
84    71    34    37    stand
**85    67    17    50    fellow**
86    66    1    65    got
87    65    3    62    pray
88    64    10    54    mind
89    63    33    30    expression
90    64    59    5    bag
91    62    5    57    pretty
92    62    54    8    nose
93    61    3    58    ve
94    62    15    47    great
95    66    64    2    wearing
96    59    45    14    whip
97    58    48    10    extended
98    59    48    11    breeches
99    59    17    42    st
**100    57    53    4    sheridan**

Presenting this in the context of other 5L5R collocates of ‘say’ demonstrates how important characters are to speech acts recorded in BMSat and that there is value in distinguishing between 5L and 5R collocates: words from speech (‘oh’, ‘ye’, ‘ha’, and ‘aye’) and words that might open speech acts (‘dear’, ‘good’, ‘come’) are strong 5R; words associated with how something in being said such as ‘turns’ (“turns and says:…”) and ‘points’ (“points and says:…”) or describing the speaker and his or her possessions (‘sword’, ‘hat’, ‘pocket’) are strong 5L . And when we look at this more closely and start breaking the data down by decade we start seeing some interesting patterns. ‘John’ (presumably in most cases ‘Bull’), ‘Napoleon’, and to a less extent ‘King’ appear 5L of ‘say’ often in the first decade of the nineteenth century (ranked 2, 7 and 18 respectively): they have, presumably, a lot to say. Around the same time ‘Lord’ flips from being the speaker to the subject of things that are said (accounting for, one imagines, a steady presence of speech acts that begin “My lord, …”. And looking at ‘man’ and ‘woman’ – the subject of a previous post – we find the former appears within five words either side of ‘say’ more than twice as often as the latter. Hardly surprising. And yet for ‘man’ 71.1% of those occurrences appear 5L of ‘say’ and for ‘woman’ 84.9%, suggesting that when they appear in prints generic women (for of course this count takes no account of those speaking men and women described in BMSat by name or some other nomenclature) tend to speak.

There are other stories to tell from this data: the strength in the 1780s of ‘Fox’ as both a subject and instigator of speech and the steep decline of the word thereafter; the consistent words such as ‘dear’, ‘let’, and ‘oh’; and the sharp decline in ‘profile’ as a collocate of ‘say’ (over half as few occurrences in 1800-1830 as in the 1790s: are there, perhaps, fewer figures after 1800 ‘standing in profile saying “My lord, …”). Anyway, these are thoughts for another time.

If you are interested in how the lemmatisation and collocation work was done, my tool of preference at the moment is AntConc.