Bigram Analysis: Neville's Letters vs. Shakespeare

I created an XML file of Neville's letters from Winwood's Memorials of Affairs of State, Vol 1 and 2. That's 89 letters Henry Neville wrote, mostly as ambassador from 1599-1601.

Using Pervez Rizvi's database of early modern English plays, I did a comparison of bigrams (two word combos) to see which plays more closely match the letters. I evaluated 239 plays from 1590-1615. The results are stunning. Shakespeare's plays rank at the top consistently:

Rank	Year	Similarity	Title
1	1613	0.6126	Henry VIII [Full Play]
2	1609	0.6079	The Winter's Tale
3	1599	0.5897	Henry V
4	1613	0.5866	Henry VIII [Shakespeare Section]
5	1610	0.5843	Cymbeline
6	1603	0.5736	All's Well That Ends Well
7	1600	0.5687	Cynthia's Revels (Jonson)
8	1597	0.5659	Henry IV, Part 2
9	1608	0.5652	Coriolanus
10	1602	0.5645	The Royal King and the Loyal Subject
11	1607	0.5597	The Tragedy of Charles Duke of Byron
12	1595	0.5585	Love's Labor's Lost
13	1603	0.5583	Measure for Measure
14	1599	0.5528	1 Edward the Fourth
15	1599	0.5509	Every Man Out of His Humour (Jonson)
16	1597	0.5473	Henry IV, Part 1
17	1605	0.5448	Philotas
18	1601	0.5431	Hamlet
19	1599	0.5420	2 Edward the Fourth
20	1609	0.5414	Epicoene (Jonson)
21	1591	0.5404	Henry VI, Part 2
22	1607	0.5398	The Conspiracy of Charles Duke of Byron
23	1605	0.5392	King Lear
24	1614	0.5381	The Hector of Germany
25	1614	0.5356	Bartholomew Fair (Jonson)
26	1604	0.5333	Sejanus His Fall (Jonson)
27	1613	0.5326	Henry VIII [Fletcher Section]
28	1606	0.5324	The Isle of Gulls
29	1604	0.5310	The Widow's Tears
30	1605	0.5301	Volpone (Jonson)
31	1611	0.5299	Catiline His Conspiracy (Jonson)
32	1606	0.5289	Antony and Cleopatra
33	1610	0.5285	The Revenge of Bussy D'Ambois
34	1614	0.5280	The Staple of News
35	1598	0.5240	Every Man in His Humour (Jonson)
36	1592	0.5216	A Knack to Know a Knave
37	1590	0.5210	Jack Straw
38	1596	0.5205	The Merchant of Venice
39	1604	0.5187	When You See Me You Know Me
40	1595	0.5173	Richard II

This is not conclusive evidence that Henry Neville wrote the works of Shakespeare. But it is an objective and reproducible test that shows a clear affinity between the two-word phrases Henry Neville and Shakespeare used. 

This overlap is partly due to the topic of the plays aligning with the experiences Neville had as ambassador. This is not a defect in this study. Quite the opposite, the overlap is another piece of strong evidence.

This research was done with the help of Claude Code. 

I ran a similar test, with the help of ChatGPT Codex, that reduces reliance on topical words.  "The new test uses function‑word bigrams only (top 200 MFW), then compares length‑matched windows with bootstrapping and reports mean ± std. This reduces topical bias and makes comparisons fairer across different text lengths." Very similar results:

Rank	Year	Title	Mean_Sim
1	1613	Henry VIII [Shakespeare Sect]	0.7802
2	1613	Henry VIII	0.7548
3	1599	Henry V	0.7533
4	1607	Tragedy of Charles Duke of Byron	0.7435
5	1607	Conspiracy of Charles Duke of Byron	0.7386
6	1609	The Winter's Tale	0.7342
7	1605	Philotas	0.7341
8	1606	Macbeth	0.7309
9	1614	The Hector of Germany	0.7268
10	1595	Richard II	0.7219
11	1604	Sejanus His Fall	0.7207
12	1610	Cymbeline	0.7202
13	1608	Coriolanus	0.7191
14	1597	Henry IV, Part 2	0.7165
15	1590	The Reign of King Edward the Third	0.7160
16	1591	Locrine	0.7147
17	1606	The Rape of Lucrece	0.7130
18	1592	Summer's Last Will and Testament	0.7129
19	1610	The Revenge of Bussy D'Ambois	0.7117
20	1603	The Family of Love	0.7116
21	1596	King John	0.7108
22	1592	Henry VI, Part 1	0.7106
23	1606	Hymenaei	0.7103
24	1603	All's Well That Ends Well	0.7094
25	1595	Love's Labor's Lost	0.7091
26	1611	The Atheist's Tragedy	0.7089
27	1591	1 The Troublesome Reign of King John	0.7055
28	1613	Henry VIII [Fletcher Section]	0.7018
29	1604	The Widow's Tears	0.7017
30	1590	The Love of David and Fair Bathsheba	0.7012
31	1593	The Massacre at Paris	0.7005
32	1611	Catiline His Conspiracy	0.6993
33	1591	Henry VI, Part 2	0.6987
34	1591	2 The Troublesome Reign of King John	0.6970
35	1610	The Golden Age	0.6965
36	1606	The Isle of Gulls	0.6942
37	1614	The Staple of News	0.6939
38	1603	Measure for Measure	0.6935
39	1606	Antony and Cleopatra	0.6930
40	1590	Jack Straw	0.6928

I have placed all the necessary information to reproduce both tests here: https://nevilleresearch.com/bigram/ 

Neville wrote a confession on March 2, 1601 after the execution of the Earl of Essex.  You can read it here. If you take this confession and compare all of the rare bigrams (found in 5 or fewer plays), the ranking of 239 plays looks like this:

Rank	Year	Shared	Play
1	1610	10	Cymbeline (Shakespeare)
2	1613	10	Henry VIII (Shakespeare/Fletcher)
3	1613	10	Two Noble Kinsmen (Shakespeare/Fletcher)
4	1597	9	Henry IV, Part 2 (Shakespeare)
5	1600	9	Cynthia's Revels
6	1599	8	Clyomon and Clamydes
7	1604	8	The Widow's Tears
8	1611	8	Catiline His Conspiracy
9	1612	8	The White Devil
10	1604	7	Sejanus His Fall
...
15	1601	6	Hamlet (Shakespeare)

If just look at all bigrams, Hamlet is the second highest:

Rank	Year	Similarity	Play
1	1592	0.411	Edward the Second (Marlowe)
2	1601	0.401	Hamlet (Shakespeare)
3	1607	0.389	Tragedy of Charles Duke of Byron (Chapman)
4	1605	0.372	The Revenger's Tragedy
5	1599	0.369	1 Sir John Oldcastle
6	1602	0.360	The Gentleman Usher (Chapman)
7	1592	0.350	Richard III (Shakespeare)
8	1604	0.345	Bussy D'Ambois (Chapman)
9	1604	0.343	1 If You Know Not Me You Know Nobody
10	1593	0.343	The Massacre at Paris (Marlowe)

Please read the confession. 

Comparing Neville's "Case", another document written after his arrest for the Essex Rebellion, produce an even stronger result:

Here is another version of the same test, with extensive details, generated by ChatGPT Codex:

The Case (1601) vs. Early Modern Plays: Function-Word + POS Bootstrap Analysis

Overview

This report summarizes a function-word and part-of-speech (POS) n-gram analysis of Sir Henry Neville's The Case (1601), comparing it to 239 plays from 1590–1615. The method emphasizes syntactic habit over topical vocabulary, and uses bootstrap sampling to estimate the stability of similarity rankings.

This syntactic result aligns with earlier findings from word-bigram and rare-bigram analyses, which also placed Hamlet and other Shakespeare plays in the top tier. The key difference is that this test is largely insulated from content vocabulary, suggesting that the observed affinity persists even when topical overlap is minimized. That convergence across methods strengthens the case for a genuine stylistic proximity rather than a coincidence of subject matter.

Method in brief

1. Lemmas and POS tags were extracted from the XML text of The Case.
2. Each play was tokenized by lemma from the Early Modern Plays database, then POS-tagged with spaCy.
3. Two feature sets were built for each text: function-word n-grams and POS-tag n-grams (n = 2--4).
4. Cosine similarity was computed between The Case and each play for both feature sets.
5. The final score is the mean of the function-word and POS similarities.
6. Bootstrap stability: 500 random 400-token windows were drawn from The Case, producing 500 rankings; we report the mean similarity and a 95% interval for each play.

Parameters

• Plays: 239 texts dated 1590--1615
• Function words: 200+ Early Modern function words (Burrows-style list with EM additions)
• POS tagger: spaCy en_core_web_sm
• N-gram sizes: 2--4
• Window size: 400 tokens
• Bootstrap samples: 500

Top 20 plays by combined mean similarity

Rank	Play	Year	Combined Mean	95% Interval
1	Hamlet	1601	0.1337	[0.1093, 0.1742]
2	Richard III	1592	0.1329	[0.1138, 0.1712]
3	2 Edward the Fourth	1599	0.1306	[0.1124, 0.1587]
4	Henry VI, Part 2	1591	0.1297	[0.1086, 0.1719]
5	Cymbeline	1610	0.1291	[0.1066, 0.1670]
6	The True Chronicle of King Leir	1590	0.1290	[0.1110, 0.1551]
7	Othello	1604	0.1281	[0.1071, 0.1513]
8	The Winter's Tale	1609	0.1275	[0.1067, 0.1629]
9	Edward the Second	1592	0.1269	[0.1078, 0.1583]
10	Alphonsus, Emperor of Germany	1594	0.1269	[0.1072, 0.1621]
11	Antony and Cleopatra	1606	0.1267	[0.1066, 0.1647]
12	The Queen's Arcadia	1605	0.1266	[0.1135, 0.1540]
13	The Tragedy of Charles Duke of Byron	1607	0.1266	[0.1056, 0.1753]
14	Clyomon and Clamydes	1599	0.1264	[0.1060, 0.1467]
15	Henry IV, Part 2	1597	0.1264	[0.1017, 0.1702]
16	Fair Em	1590	0.1263	[0.1094, 0.1531]
17	Measure for Measure	1603	0.1262	[0.1023, 0.1562]
18	As You Like It	1599	0.1253	[0.1052, 0.1536]
19	Volpone	1605	0.1253	[0.1060, 0.1558]
20	Two Lamentable Tragedies	1594	0.1248	[0.1062, 0.1564]

Interpretation

The highest-ranked plays are dominated by Shakespeare’s late and middle-period works, with Hamlet at the top of the list. This is notable because the method suppresses content vocabulary and instead emphasizes grammatical habit (function-word sequences and POS patterns). That means the observed affinity is less likely to be driven by shared topics and more likely to reflect structural linguistic tendencies.

Caveats

• The Case is short (1,422 tokens), so even bootstrap windows are drawn from a limited pool.
• POS tags are produced by a modern tagger; Early Modern syntax may be partially misclassified.
• Similarity does not prove authorship; it indicates stylistic proximity under a specific metric.

[Please note, if you take Hamlet and run the same bigram test against ALL the plays from 1590-1615, Shakespeare plays pop up almost exclusively, showing the value of the test:

  1. Hamlet (503) 1.0000

  2. All’s Well That Ends Well (496) 0.7479

  3. King Lear (507) 0.7459

  4. Othello (512) 0.7456

  5. The Winter’s Tale (525) 0.7448

  6. Henry IV, Part 2 (491) 0.7442

  7. Richard III (515) 0.7400

  8. Measure for Measure (509) 0.7285

  9. Troilus and Cressida (523) 0.7227

  10. 1 Sir John Oldcastle (357) 0.7221

  11. The White Devil (52) 0.7215

  12. Henry IV, Part 1 (489) 0.7190

  13. Cymbeline (499) 0.7181

  14. Henry VIII (502) 0.7122

  15. The Gentleman Usher (452) 0.7058

  16. The Woman Hater (413) 0.7021

  17. Antony and Cleopatra (495) 0.6979

  18. 2 Edward the Fourth (340) 0.6925

  19. Much Ado About Nothing (494) 0.6900

  20. Sir Giles Goosecap (388) 0.6895]

Comparing Neville's 1613 "Advice" to King James, we get these amazing results: 

 Top 20 (lemma bigrams, 1590–1615):

  - 1 | 1599 | 0.3974 | Henry V

  - 2 | 1604 | 0.3802 | Sejanus His Fall

  - 3 | 1609 | 0.3779 | The Winter's Tale

  - 4 | 1613 | 0.3771 | Henry VIII

  - 5 | 1600 | 0.3738 | Cynthia's Revels

  - 6 | 1595 | 0.3697 | Love's Labor's Lost

  - 7 | 1605 | 0.3666 | Philotas

  - 8 | 1608 | 0.3617 | Coriolanus

  - 9 | 1590 | 0.3606 | Jack Straw

  - 10 | 1610 | 0.3581 | Cymbeline

  - 11 | 1607 | 0.3570 | The Conspiracy of Charles Duke of Byron

  - 12 | 1597 | 0.3562 | Henry IV, Part 2

  - 13 | 1601 | 0.3542 | Hamlet

  - 14 | 1610 | 0.3537 | The Revenge of Bussy D'Ambois

  - 15 | 1597 | 0.3527 | Henry IV, Part 1

  - 16 | 1611 | 0.3526 | Catiline His Conspiracy

  - 17 | 1604 | 0.3517 | Arches of Triumph

  - 18 | 1607 | 0.3504 | The Tragedy of Charles Duke of Byron

  - 19 | 1599 | 0.3471 | 1 Edward the Fourth

  - 20 | 1603 | 0.3471 | All's Well That Ends Well

Trigrams provide an equally strong result:

 Top 20 (lemma trigrams, 1590–1615):

  - 1 | 1609 | 0.0263 | The Captain

  - 2 | 1613 | 0.0263 | Henry VIII

  - 3 | 1603 | 0.0253 | All’s Well That Ends Well

  - 4 | 1605 | 0.0251 | The Noble Gentleman

  - 5 | 1609 | 0.0249 | The Winter’s Tale

  - 6 | 1599 | 0.0247 | 2 Edward the Fourth

  - 7 | 1598 | 0.0236 | Much Ado About Nothing

  - 8 | 1597 | 0.0235 | Henry IV, Part 2

  - 9 | 1599 | 0.0235 | Henry V

  - 10 | 1611 | 0.0233 | A King and No King

  - 11 | 1599 | 0.0232 | 1 Edward the Fourth

  - 12 | 1610 | 0.0231 | The Maid’s Tragedy

  - 13 | 1601 | 0.0220 | Hamlet

  - 14 | 1595 | 0.0220 | Love’s Labor’s Lost

  - 15 | 1599 | 0.0220 | As You Like It

  - 16 | 1592 | 0.0218 | Edward the Second

  - 17 | 1607 | 0.0217 | Cupid’s Revenge

  - 18 | 1599 | 0.0216 | Julius Caesar

  - 19 | 1597 | 0.0208 | Henry IV, Part 1

  - 20 | 1597 | 0.0208 | An Humorous Day’s Mirth