Uniloc 2017 LLC

Patent Trials and Appeals Board

Oct 7, 2021

IPR2020-00755 (P.T.A.B. Oct. 7, 2021)

Trials@uspto.gov Paper No. 28
571-272-7822 Date: October 7, 2021

UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

GOOGLE LLC,
Petitioner,

v.

UNILOC 2017 LLC,
Patent Owner.
____________

IPR2020-00755
Patent 6,366,908 B1
____________

Before SALLY C. MEDLEY, KRISTEN L. DROESCH, and
SHEILA F. McSHANE, Administrative Patent Judges.

MEDLEY, Administrative Patent Judge.

JUDGMENT
Final Written Decision
Determining All Challenged Claims Unpatentable
35 U.S.C. § 318(a)
IPR2020-00755
Patent 6,366,908 B1

2
I. INTRODUCTION
Google LLC (“Petitioner”) filed a Petition for inter partes review of
claims 6–12 of U.S. Patent No. 6,366,908 B1 (Ex. 1001, “the ’908 patent”).
Paper 1 (“Pet.”). Uniloc 2017 LLC (“Patent Owner”) filed a Preliminary
Response. Paper 8 (“Prelim. Resp.”). In accordance with Board
authorization, Petitioner filed a Reply to the Preliminary Response (Paper 9)
and Patent Owner filed a Sur-reply (Paper 11). Upon review of the papers,
we instituted inter partes review, pursuant to 35 U.S.C. § 314, as to
claims 6–12 based on the challenges set forth in the Petition. Paper 14
(“Decision to Institute” or “Dec.”).
Subsequent to institution, Patent Owner filed a Patent Owner
Response (Paper 17, “PO Resp.”), Petitioner filed a Reply to Patent Owner’s
Response (Paper 19, “Pet. Reply”), and Patent Owner filed a Sur-reply
(Paper 20, “Sur-reply”). On July 15, 2021, we held an oral hearing. A
transcript of the hearing is of record. Paper 27 (“Tr.”).
For the reasons that follow, we conclude that Petitioner has proven by
a preponderance of the evidence that claims 6–12 of the ’908 patent are
unpatentable.
A. Related Matters
Petitioner and Patent Owner indicate that the ’908 patent is the subject
of court proceeding, Uniloc 2017 LLC v. Google LLC, No. 2:18-cv-00553
(E.D. Tex.). Pet. 93; Prelim. Resp. 3. Petitioner’s motion to transfer the
district court case to the Northern District of California was granted.
Ex. 1046. According to Petitioner, “the Northern District of California
found that at least Fortress Credit Co. LLC held sufficient rights in the
challenged patent that Uniloc lacked standing to sue. The court then
IPR2020-00755
Patent 6,366,908 B1

3
dismissed the litigation for lack of subject matter jurisdiction.” Paper 18, 1
(citing Uniloc 2017 LLC v. Google LLC, No. 4:20-cv-05345-YGR (N.D.
Cal. Entered Dec. 22, 2020 (single order addressing eleven cases)).
Petitioner states “Uniloc filed a notice of appeal on December 31, 2020.” Id.
B. The ’908 Patent1
The ’908 patent describes a keyfact-based text retrieval method and a
keyfact-based text index method. Ex. 1001, 1:6–7. The ’908 patent states
that a keyfact is “an important fact contained in sentences which constitute a
document,” where the keyfact is “represented by an object and property
information through syntactic analysis of a sentence.” Id. at 1:15–18. As
described in the ’908 patent, a keyword-based text retrieval method was the
mainstream in conventional text retrieval methods, but the precision of a
keyword-based text retrieval method was less than ideal for several reasons.
Id. at 1:19–22. First, the meaning of the document was not precisely
represented because the document was represented by keywords, which are
nouns. Id. at 1:22–25. Second, when a query included a natural language
phrase, sentence, or keywords, the intention of the user’s query was not
reflected precisely in a keyword-based text retrieval method. Id. at 1:27–32.
Therefore, the keyword-based text retrieval method had a fundamental
limitation in retrieval precision because it performed document retrieval by
keywords. Id. at 1:32–34. Phrase-based text retrieval methods perform
more precise text retrieval than the keyword-based text retrieval method, but
performs less precise text retrieval than a concept-based text retrieval
method, which expresses text by concept units. Id. at 1:45–49.

1 The ’908 patent appears to be expired.
IPR2020-00755
Patent 6,366,908 B1

4
The ’908 patent further describes a new approach to keyfact-based
text retrieval that overcomes the shortcomings of the keyword-based text
retrieval method and generalized phrase-based text retrieval methods. Id.
at 1:50–53. In a keyfact-based text retrieval method, parts of text that
represent the same meaning are described as a keyfact, where the phrases or
words having the same meaning are indexed as the same indexing terms. Id.
at 1:53–55, 1:60–62. According to the ’908 patent, since the keyfact-based
retrieval method is a concept-based retrieval method, indexing and retrieval
of the keyfact-based retrieval method are performed with the unit of the
keyfact, and precision of the retrieval is greatly improved. Id. at 1:55–59.
A block diagram of a keyfact-based text retrieval system is illustrated
in Figure 1, reproduced below.

Figure 1 is a block diagram illustrating a keyfact-based text retrieval
system. Id. at 4:22–23. The keyfact-based text retrieval system comprises
IPR2020-00755
Patent 6,366,908 B1

5
keyfact extraction device 11, keyfact index device 12, and keyfact retrieval
device 13. Id. at 4:23–26. The ’908 patent describes that once a document
collection 14 or a query 15 is given, the keyfact extraction device 11 extracts
words without ambiguity by performing morphological analysis and tagging.
Id. at 4:42–45. The keyfact generation rule is applied to the words and then
the keyfacts are extracted. Id. at 4:45–46. The keyfact index device 12
indexes the document collection 14 or the query 15 with the unit of keyfact
and calculates the frequencies of the keyfacts. Id. at 4:47–49. The
frequencies of the keyfacts are stored into the index structure 16 with the
document ID information. Id. at 4:49–51. The keyfact retrieval device 13
orders documents using a similarity calculation method and shows retrieval
results. Id. at 4:51–53.
A block diagram of a keyfact extraction device of a keyfact-based text
retrieval system is illustrated in Figure 3, reproduced below.
IPR2020-00755
Patent 6,366,908 B1

6

Figure 3 illustrates a block diagram of keyfact extraction device 11
that analyzes a document and generates keyfacts through the processes of
morphological analysis, part-of-speech tagging, keyfact pattern extraction,
and keyfact generation. Id. at 5:15–18. The ’908 patent describes that a
document is supplied at stage 31 and morphological analysis is performed at
stage 32. Id. at 5:19–20. More specifically, a sentence in the document is
divided into words, the morphological analysis is performed with
dictionaries 36 at stage 32, and various results are obtained. Id. at 5:20–22,
5:59–60. At stage 33, part-of-speech tagging is performed, where a precise
sequence of tags is chosen among the various results of the morphological
analysis. Id. at 5:61–63. Once the final sequence of tags is obtained, the
IPR2020-00755
Patent 6,366,908 B1

7
stage of keyfact pattern extraction 34 searches the keyfact pattern rule 37
and extracts meaningful keyfact patterns necessary for keyfact generation.
Id. at 6:5–8. “Keyfact terms that have forms of [object, property] are
generated as to the input keyfact pattern at the stage of the keyfact
generation 35 by searching the keyfact generation rule 38,” resulting in a
keyfact list 39. Id. at 6:38–41, 51.
A block diagram of a keyfact index device of a keyfact-based text
retrieval system is illustrated in Figure 4, reproduced below.

IPR2020-00755
Patent 6,366,908 B1

8
Figure 4 illustrates a block diagram of keyfact index device 12 that
calculates statistical frequencies of keyfacts in a document obtained from the
keyfact extraction device 11, and that forms an index structure. Id. at 6:58–
60. The ’908 patent describes that, for each document, a keyfact frequency
and document frequency of a keyfact 42 are calculated at 43 in order to
obtain the frequency information of the keyfacts. Id. at 6:65–67. Next,
supplementary tables, such as a document index table, a document table, and
a keyfact index table, are generated to form an efficient index structure 44.
Id. at 7:1–3. Subsequently, an index structure is formed in a unit of the
keyfact at 45, and an index file is stored at 46. Id. at 7:10–11.
A block diagram of keyfact retrieval device of a keyfact-based text
retrieval system is illustrated in Figure 5, reproduced below.

IPR2020-00755
Patent 6,366,908 B1

9
Figure 5 illustrates a block diagram of keyfact retrieval device 13 that,
at 54, forms a document vector and query vector with a keyfact supplied
from a keyfact extraction device 53 and an index file 52 generated by a
keyfact index device 51. Id. at 7:35–39. At 55, the keyfact weight constants
(CKfType#), which are fit for the attribute of a document collection, are
determined. Id. at 7:40–41. Subsequently, at 56, the keyfact weights are
calculated from the document and query vector. Id. at 7:42–43. Further,
at 57, the similarity of each document appropriate for the query is calculated
by employing a keyfact retrieval model 58, where the result of a similarity
calculation determines an order of appropriate documents. Id. at 8:20–24.
C. Illustrative Claim
Petitioner challenges claims 6–12 of the ’908 patent. Pet. 1. Claim 6
is an independent claim, and claims 7–12 depend therefrom. Claim 6 is
reproduced below.
6. A keyfact-based text retrieving method comprising:
keyfact extracting step for analyzing a document collection and
a user query, and extracting keywords without part-of-speech
ambiguity from said document collection and said user query,
and respectively extracting keyfacts of said document
collection and said user query from said keywords;
keyfact indexing step for calculating the frequency of said
keyfacts of said document collection and generating a keyfact
list of said document collection for a keyfact index structure;
and
keyfact retrieving step for receiving said keyfact of said user
query and said keyfacts of said document collection and
defining a keyfact retrieval model in consideration of weigh
factors according to a keyfact pattern and generating a
retrieval result.
Ex. 1001, 9:54–10:7.
IPR2020-00755
Patent 6,366,908 B1

10
D. Instituted Grounds of Unpatentability
We instituted inter partes review based on the following grounds of
unpatentability under 35 U.S.C. § 103(a)2 as follows (Dec. 9–10, 38):
Claim(s) Challenged 35 U.S.C. § Reference(s)/Basis
6–12 103(a) Braden-Harder3, Grossman4
6–12 103(a) Braden-Harder, Heidorn
5,
Messerly6, Grossman
7–9 103(a) Braden-Harder, Grossman, Kucera7
7–9 103(a) Braden-Harder, Heidorn, Messerly, Grossman, Kucera
9 103(a) Braden-Harder, Grossman, Miller8
9 103(a) Braden-Harder, Heidorn, Messerly, Grossman, Miller
9 103(a) Braden-Harder, Grossman, Kucera, Miller

2 The Leahy-Smith America Invents Act, Pub. L. No. 112-29, 125 Stat. 284
(2011) (“AIA”), amended several provisions of 35 U.S.C., including § 103.
Because the ’908 patent has an effective filing date before the effective date
of the applicable AIA amendments, we refer to the pre-AIA version of
35 U.S.C. § 103. Petitioner asserts, and Patent Owner does not dispute, that
each relied upon reference is prior art under the pre-AIA version. Pet. 2, 7,
9, 12, 15–16; see generally PO Resp.
3 US 5,933,822, issued Aug. 3, 1999 (Ex. 1020, “Braden-Harder”).
4 David A. Grossman and Ophir Frieder, Information Retrieval: Algorithms
and Heuristics, Kluwer International Series in Engineering and Computer
Science (Kluwer Academic Publishers, Springer Science + Business Media
New York prtg.) (1st. ed. 1998) (Ex. 1010, “Grossman”).
5 US 5,966,686, issued Oct. 12, 1999 (Ex. 1022, “Heidorn”).
6 US 6,076,051, issued June 13, 2000 (Ex. 1025, “Messerly”).
7 US 4,868,750, issued Sept. 19, 1989 (Ex. 1011, “Kucera”).
8 George A. Miller et al., Introduction to WordNet: An Online Lexical
Database, International Journal of Lexicography, Vol. 3, No. 4, 235–244
(1990) (Ex. 1027, “Miller”).
IPR2020-00755
Patent 6,366,908 B1

11
Claim(s) Challenged 35 U.S.C. § Reference(s)/Basis
9 103(a)
Braden-Harder, Heidorn,
Messerly, Grossman, Kucera,
Miller
II. DISCUSSION
A. Principles of Law
To prevail in its challenges to Patent Owner’s claims, Petitioner must
demonstrate by a preponderance of the evidence that the claims are
unpatentable. 35 U.S.C. § 316(e); 37 C.F.R. § 42.1(d) (2019). A patent
claim is unpatentable under 35 U.S.C. § 103(a) if the differences between
the claimed subject matter and the prior art are such that the subject matter,
as a whole, would have been obvious at the time the invention was made to a
person having ordinary skill in the art to which said subject matter pertains.
KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 406 (2007). The question of
obviousness is resolved on the basis of underlying factual determinations
including (1) the scope and content of the prior art; (2) any differences
between the claimed subject matter and the prior art; (3) the level of ordinary
skill in the art; and (4) when in evidence, objective evidence of
nonobviousness.9 Graham v. John Deere Co., 383 U.S. 1, 17–18 (1966).
B. Level of Ordinary Skill
In determining the level of ordinary skill in the art, various factors
may be considered, including the “type of problems encountered in the art;
prior art solutions to those problems; rapidity with which innovations are
made; sophistication of the technology; and educational level of active

9 Patent Owner does not present any objective evidence of nonobviousness
as to the challenged claims.
IPR2020-00755
Patent 6,366,908 B1

12
workers in the field.” In re GPAC Inc., 57 F.3d 1573, 1579 (Fed. Cir. 1995)
(citation omitted). Petitioner relies on the declaration testimony of
Dr. Bernard J. Jansen, who testifies that a person having ordinary skill in the
art “would have had a Bachelor’s degree in computer science, information
technology, or the equivalent, and one to two years of experience working
with at least one of information retrieval systems, the structure and
organization of databases, and natural language processing.” Pet. 22 (citing
Ex. 1003 ¶ 142). According to Dr. Jansen, “significant work experience in
any of these areas could substitute for formal education, and vice versa.”
Ex. 1003 ¶ 142.
Patent Owner indicates Petitioner’s proposed assessment is improper
because it does not define thresholds for “significant work experience” or
“formal education.” PO Resp. 4. Patent Owner, however, does not offer its
own assessment of the level of skill in the art, or indicate that the resolution
of any issue depends on a specific type of work experience or a specific
level of education. See id. Patent Owner additionally states that even if
Petitioner’s proposed level of qualifications were applied, the Petition does
not establish unpatentability. Id. at 5.
We adopt the assessment offered by Petitioner as it is consistent with
the ’908 patent and the asserted prior art. We further note that the prior art
of record in the instant proceeding reflects the appropriate level of ordinary
skill in the art. Cf. Okajima v. Bourdeau, 261 F.3d 1350, 1354–55 (Fed. Cir.
2001) (holding the Board may omit specific findings as to the level of
ordinary skill in the art “where the prior art itself reflects an appropriate
level and a need for testimony is not shown”).
IPR2020-00755
Patent 6,366,908 B1

13
C. Claim Construction
In an inter partes review for a petition filed on or after November 13,
2018, “[claims] of a patent . . . shall be construed using the same claim
construction standard that would be used to construe the [claims] in a civil
action under 35 U.S.C. § 282(b), including construing the [claims] in
accordance with the ordinary and customary meaning of such claims as
understood by one of ordinary skill in the art and the prosecution history
pertaining to the patent.” See 37 C.F.R. § 42.100(b) (2019); see also
Phillips v. AWH Corp., 415 F.3d 1303, 1312–14 (Fed. Cir. 2005) (en banc).
“step for” limitations
Petitioner argues that the claim 6 “keyfact extracting step for,”
“keyfact indexing step for,” and “keyfact retrieving step for” (collectively,
“keyfact steps”) should each be construed as “step-for” limitations in
accordance with 35 U.S.C. § 112, ¶ 6. Pet. 25–31. For the same reasons
presented in its Preliminary Response, Patent Owner argues that these claim
6 terms should not be construed in accordance with 35 U.S.C. § 112, ¶ 6.
Compare PO Resp. 10–14, with Prelim. Resp. 15–19. Although we
preliminarily agreed that the claims should not be construed in accordance
with 35 U.S.C. § 112, ¶ 6, we also determined that whether the claim 6 terms
invoke 35 U.S.C. § 112, ¶ 6 “is not material to resolving the issues currently
before us, because Petitioner has shown a reasonable likelihood of prevailing
under either construction.” Dec. 17 (explaining how Petitioner’s proposed
construction is narrow, whereas Patent Owner’s proposed construction is
broad, and that Petitioner contends that the prior art renders the claims
obvious “regardless of whether § 112, ¶ 6 applies”). Petitioner agrees with
IPR2020-00755
Patent 6,366,908 B1

14
our initial determination that we need not “address the issue because the
claims are unpatentable even under Uniloc’s argument.” Pet. Reply 5.
Patent Owner, however, presents numerous arguments why the claims
should not be construed in accordance with 35 U.S.C. § 112, ¶ 6 without
clearly explaining why it is necessary to address § 112, ¶ 6 in order to
resolve the controversies on the merits that are before us. PO Resp. 9–19;
Sur-reply 7–9. As discussed below, beside the arguments directed to the
sufficiency of the prior art teaching the “keyfact steps” limitations based on
Patent Owner’s construction of the term “keyfact,” Patent Owner fails to
explain sufficiently what is lacking from Petitioner’s challenges and related
analyses provided that assert that claims 6–12 would have been obvious
based on the cited prior art under either construction. For example, Patent
Owner argues that “Petitioner fails to recognize that the claim language
recites various acts” and that Petitioner construes each recited act as a
function. PO Resp. 9. In applying a 112, ¶ 6 analysis to claim 6, however,
Petitioner looks to the ’908 patent for the corresponding acts associated with
the recited function. See, e.g., Pet. 26–31. Thus, Patent Owner’s arguments
appear to us to be a matter of form over substance because Petitioner
addresses the corresponding acts by looking to the Specification of the ’908
patent. Tr. 8:12–25. Patent Owner does not persuade us otherwise. For
example, Patent Owner argues that the claim 6 “keyfact extracting step”
requires “a respective set of acts” such as “‘analyzing …,’ ‘extracting …,’
and ‘respectively extracting … .’” PO Resp. 12–13 (alterations in original).
Patent Owner argues that “Petitioner fails to comprehend the logical
interrelationship between groups of acts and the underlying respective result,
expressed as a ‘step,’ that each group of acts operated to collectively
achieve.” Id. at 14. Patent Owner, however, fails to identify or explain how
IPR2020-00755
Patent 6,366,908 B1

15
Petitioner’s analysis of the claim language and prior art fails to address the
claimed acts. As explained in detail below, we determine that Petitioner
shows, by a preponderance of the evidence, how Braden-Harder and
Grossman teach or suggest each of the claimed steps, such as the “keyfact
extracting step.” Pet. 33–47. As further explained below, Patent Owner
does not present meaningful arguments to the contrary.
Patent Owner further argues that Petitioner’s proposed constructions
“violate[] both the doctrine of claim differentiation and the ‘all elements’
rule.” PO Resp. 14–19. In particular, Patent Owner argues that dependent
claim 7 recites additional acts required for the “keyfact extracting step” of
claim 6 that are distinct from what claim 6 recites. Id. at 15–16. Patent
Owner argues that “the ‘analyzing’ act of claim 6 (expressly directed to both
‘a document collection and a user query’ together) must require something
more than the distinct ‘analyzing’ of dependent claim 7 (which is directed,
instead, to ‘morphology of an input sentence’).” Id. at 15. Patent Owner
contends that Petitioner erroneously conflates the acts of claim 6 with the
acts of the dependent claims “thereby failing to give meaningful effect to
those acts affirmatively recited in claim 6.” Id. at 16–19.
Petitioner argues that the dependent claims “all identically recite
‘wherein said step of … comprises the steps of …’ language.” Pet. Reply 5
(alterations in original). As such, Petitioner argues that the dependent claims
should be “construed as describing details required as part of performing the
previously introduced steps” and that the dependent claims do “not add new
‘distinct and additional’ steps—contrary to Uniloc’s contentions.” Id. at 5–9
(citing Realtime Data, LLC v. Iancu, 912 F.3d 1368, 1375 (Fed. Cir. 2019)),
24–25 (arguing that dependent claims 7, 10, and 11 do not add new distinct
and additional steps, but “give details of the more general steps previously
IPR2020-00755
Patent 6,366,908 B1

16
claimed”). As such, Petitioner argues that “Uniloc’s arguments fall with its
own erroneous interpretation of the dependent claims.” Id. at 24–25.
Petitioner further argues that whether 112, ¶ 6 applies “[t]he challenged
claims are unpatentable either way.” Id. at 8–9. Petitioner contends that the
“all elements” rule is not applicable here, because that “rule relates to the
doctrine of equivalents for infringement.” Id. at 9 (citing Warner-Jenkinson
Co. v. Hilton Davis Chem. Co., 520 U.S. 17, 29 (1997)).
We agree with Petitioner that the dependent claims do not add new
distinct steps or acts, but “give details of the more general steps previously
claimed.” Id. at 5–9, 24–25. That is, we agree with Petitioner that claim 7
recites “wherein said step of keyfact extracting comprises the steps of,”
which should be construed as describing details required as part of
performing the previously introduced step. Id. at 5. On the other hand,
Patent Owner fails to explain sufficiently why we should construe the
dependent claims as reciting additional acts to the steps recited in claim 6.
Accordingly, we agree with Petitioner that Patent Owner’s arguments
regarding claim differentiation necessarily fail because the arguments are
based on an erroneous interpretation of the dependent claims to include
additional steps or acts untethered from the steps recited in claim 6.
Additionally, we agree with Petitioner that whether § 112, ¶ 6 applies “[t]he
challenged claims are unpatentable either way.” Id. at 8–9.10

10 Patent Owner’s arguments that Petitioner fails to address or rebut our
preliminary determination that “claim 6 does not invoke 35 U.S.C. § 112,
¶ 6” (Sur-reply 7) are misplaced because, as discussed, whether the claim 6
terms invoke 35 U.S.C. § 112, ¶ 6 is not material to resolving the issues
here.
IPR2020-00755
Patent 6,366,908 B1

17
Accordingly, consistent with the Decision to Institute, we determine
that it is not necessary to construe the claim 6 “step for” terms because
whether the claim 6 keyfact steps invoke 35 U.S.C. § 112, ¶ 6 is not material
to resolving the issues currently before us because Petitioner has shown that
the challenged claims are unpatentable under either construction. See Vivid
Techs., Inc. v. Am. Sci. & Eng’g, Inc., 200 F.3d 795, 803 (Fed. Cir. 1999)
(holding that “only those terms need be construed that are in controversy,
and only to the extent necessary to resolve the controversy”); see also Nidec
Motor Corp. v. Zhongshan Broad Ocean Motor Co. Matal, 868 F.3d 1013,
1017 (Fed. Cir. 2017) (citing Vivid Techs. in the context of an inter partes
review).
“keyfact”
Claim 6 recites the term “keyfact” several times. For example,
claim 6 recites, “extracting keyfacts of said document collection and said
user query from said keywords.” Patent Owner argues that the term
“keyfact” means “a factual extraction of a sentence which expresses
semantic relation between words in the sentence in the form of [object,
property].” PO Resp. 9 (citing Ex. 1001, 4:58–60). Patent Owner further
argues that “[t]his form represents a paired relationship between an ‘object’
that is the head and a ‘property’ that is the modifier.” Id. at 7 (citing
Ex. 1001, 1:8–10). Petitioner argues that we need not construe the term
“keyfact,” because the challenged claims are unpatentable over the asserted
prior art even under Patent Owner’s construction. Pet. Reply 1–2.
For the reasons provided in this Decision discussed below, we agree
with Petitioner that Braden-Harder’s “extracted logical forms are each ‘a
factual extraction of a sentence which expresses semantic relation between
words in the sentence in the form of [object, property].’” See id. at 10
IPR2020-00755
Patent 6,366,908 B1

18
(citing Pet. 47; Ex. 1003 ¶¶ 221–222). As explained below, we disagree,
however, that a “keyfact” must be in the exact paired form “[object,
property]” and nothing else. For purposes of this Decision, we need not
otherwise expressly construe any claim terms.
D. Asserted Obviousness of Claims 6–12 over Braden-Harder, Heidorn,
Messerly, and Grossman
1. Braden-Harder
Braden-Harder describes an information retrieval system that utilizes
natural language processing to process results retrieved by an information
retrieval search engine that ultimately yields a set of retrieved documents.
Ex. 1020, code (57). Each document is subjected to natural language
processing to produce a set of logical forms, where each logical form
encodes, in a word-relation-word manner, semantic relationships between
words in a phrase. Id. A user-supplied query is analyzed in the same
manner to yield a set of corresponding logical forms, where documents are
ranked as a predefined function of the logical forms from the documents and
the query. Id.
A high-level block diagram of an information retrieval system is
illustrated in Figure 1, reproduced below.
IPR2020-00755
Patent 6,366,908 B1

19

Figure 1 is a high-level block diagram of an information retrieval
system 5. Id. at 7:25–27. In operation, a user supplies a search query to
system 5. Id. at 7:35–36. System 5 applies this query both to information
retrieval engine 20 and natural language processor 30. Id. at 7:40–41. In
response to the query, engine 20 searches through a dataset 10 of stored
documents to yield a set of retrieved documents. Id. at 7:41–43. The set of
documents is then applied, as symbolized by line 25, as an input to
processor 30. Id. at 7:43–46. Within processor 30, each of the documents in
the set is subjected to natural language processing, specifically
morphological, syntactic and logical form, to produce logical forms for each
sentence in that document. Id. at 7:46–50. Each such logical form encodes
semantic relationships between words in a linguistic phrase in that sentence.
Id. at 7:50–53. Processor 30 analyzes the query in an identical fashion to
yield a set of corresponding logical forms. Id. at 7:53–55. Processor 30 then
compares the set of forms for the query against the sets of logical forms
associated with each of the documents in the set in order to ascertain any
match between logical forms in the query set and logical forms for each
document. Id. at 7:55–59. Documents that produce no matches are
IPR2020-00755
Patent 6,366,908 B1

20
eliminated from further consideration, where each remaining document that
contains at least one match is retained and heuristically scored by
processor 30. Id. at 7:59–63. Finally, processor 30 presents the retained
documents to the user rank-ordered based on their respective scores. Id.
at 8:2–4.
2. Heidorn
Heidorn describes a method and system for performing semantic
analysis of an input sentence within a natural language processing (“NLP”)
system. Ex. 1022, 6:28–30. A semantic analysis subsystem receives a
syntax parse tree generated by morphological and syntactic subsystems. Id.
at 6:30–32. The semantic analysis subsystem applies two sets of semantic
rules to make adjustments to the received syntax parse tree. Id. at 6:32–34.
The semantic analysis subsystem then applies a third set of semantic rules to
create a skeletal logical form graph from the syntax parse tree. Id. at 6:34–
36. The semantic analysis subsystem finally applies two additional sets of
semantic rules to the skeletal logical form graph to provide semantically
meaningful labels for the links of the logical form graph, to create additional
logical form graph nodes for missing nodes, and to unify redundant logical
form graph nodes. Id. at 6:36–42. The final logical form graph generated by
the semantic analysis subsystem represents the complete semantic analysis
of an input sentence. Id. at 6:42–44.
IPR2020-00755
Patent 6,366,908 B1

21
An example semantic analysis subsystem is illustrated in Figure 43,
reproduced below.

Figure 43 is a block diagram of a computer system for a semantic
analysis subsystem. Id. at 13:49–50. The computer 4300 contains a
memory with semantic rules 4304–4308 and rule application engine 4303.
Id. at 13:50–51. The rule application engine, under control of a central
processing unit, applies the five sets of rules to a syntax parse tree 4301 to
generate a corresponding logical form graph 4302. Id. at 13:51–54. The
syntax parse tree is generated by the morphological and syntactic
subsystems, which are not shown. Id. at 13:55–56.
3. Messerly
Messerly describes performing information retrieval using semantic
representation of text. Ex. 1025, code (57). A tokenizer generates, from an
input string information retrieval, tokens that characterize the semantic
relationship expressed in the input string. Id. The tokenizer first creates,
IPR2020-00755
Patent 6,366,908 B1

22
from the input string, a primary logical form characterizing a semantic
relationship between the selected words in the input string. Id. The
tokenizer then identifies hypernyms that each have an “is a” relationship
with one of the selected works in the input string. Id. The tokenizer then
constructs one or more alternative logical forms from the primary logical
form. Id. For each of one or more of the selected words in the input string,
the tokenizer constructs each alternative logical form by replacing the
selected word in the primary logical form with an identified hypernym of the
selected word. Id. Finally, the tokenizer generates tokens representing both
the primary logical form and the alternative logical forms. Id.
An example overview flow diagram showing steps performed in order
to construct and access an index semantically representing target documents
is illustrated in Figure 3, reproduced below.
IPR2020-00755
Patent 6,366,908 B1

23

Figure 3 is an overview flow diagram showing the steps performed by
a facility in order to construct and access an index semantically representing
target documents. Id. at 5:31–34. As represented in steps 301–304, the
facility: converts each sentence of each target document into a number of
tokens representing an expanded logical form portraying the relationship
between the important words in the sentence (including hypernyms having
similar meanings); and stores the semantic tokens in an index along with a
location in the target documents where the sentence occurs. Id. at 5:34–41,
IPR2020-00755
Patent 6,366,908 B1

24
5:56–59. After all of the target documents have been indexed, the facility is
able to process information retrieval queries against the index. Id. at 5:41–
44. As represented in steps 305–308, for each received query, the facility:
tokenizes the text of the query (i.e., converts the sentence into semantic
tokens together representing an expanded logical form for the query text);
compares these semantic tokens to the semantic tokens stored in the index to
identify locations in the target documents for which these semantic tokens
have been stored; and ranks the target documents containing these semantic
tokens in the order of their relevance to the query. Id. at 5:44–53.
4. Grossman11
Grossman describes the retrieval of information (e.g., finding relevant
documents) in response to user queries. Ex. 1010, 1. More specifically,
Grossman describes building an inverted index as part of an information
retrieval system to find quickly terms in a document collection. Id. at 134.
As described in Grossman, the inverted index includes a set of files,
including: an index file that contains an actual posting list for each distinct
term in the collection; a document file that contains information about each
distinct document; and a weight file that contains the weight for each
document. Id. at 136–137.
Grossman describes a preprocessor that reads an input file, outputs
separate files, and bulk-loads the output files into a relational database. Id.

11 Petitioner contends that Grossman (Exhibit 1010) is a textbook published
in 1998, and that a copy of the textbook was cataloged and available at the
UCB Library by November 17, 1998. Pet. 12 (citing Ex. 1009). Patent
Owner does not challenge Petitioner’s contentions as to the prior art status of
Grossman. See generally PO Resp. For the reasons provided in the Petition,
we agree with Petitioner that Grossman is “prior art at least under §102(a).”
Id.
IPR2020-00755
Patent 6,366,908 B1

25
at 169–170. Each output file corresponds to a relation. Id. at 170. The first
relation, “DOC,” contains information about each document. Id. The
second relation, “INDEX,” models the inverted index and indicates which
term appears in which document and how often the term has appeared. Id.
A third relation, “TERM,” tracks statistics for each term based on its number
of occurrences across the document collection. Id.
Example DOC, INDEX, and TERM relations are illustrated in
Tables 5.5, 5.6, and 5.8, reproduced below.

IPR2020-00755
Patent 6,366,908 B1

26

Table 5.5 illustrates a DOC relation which models the documents in
the document collection; Table 5.6 illustrates an INDEX relation which
models an inverted index by storing the occurrences of a term in a
document; and Table 5.8 illustrates a TERM relation which contains an
inverse document frequency (“idf”) of a term, where idf = log (number of
documents / number of documents which contain the term). Id. at 16, 172–
175.
5. Discussion
Petitioner contends that claims 6–12 are unpatentable under 35 U.S.C.
§ 103(a) as obvious over Braden-Harder and Grossman, or Braden-Harder,
Heidorn, Messerly, and Grossman. Pet. 33–84. For the Braden-Harder and
Grossman challenge, Petitioner contends that Braden-Harder “refers to and
incorporates two co-pending U.S. patent applications assigned to Microsoft,”
which are the Heidorn and Messerly patent applications. Id. at 5–6 (citing
Ex. 1020, 14:53–61). Petitioner further explains that, because Braden-
IPR2020-00755
Patent 6,366,908 B1

27
Harder identifies the Heidorn and Messerly patent applications and where
they can be found, “such broad and unequivocal language is plainly
sufficient to incorporate both patent applications in their entirety.” Id. at 6
(citing Paice LLC v. Ford Motor Co., 881 F.3d 894, 907 (Fed. Cir. 2018)).
Petitioner further contends that “[e]verything disclosed in the Heidorn and
Messerly patent applications is ‘effectively part of’ Braden-Harder as if
these disclosures were ‘explicitly contained therein.’” Id. Petitioner
explains that the Petition cites to the Messerly and Heidorn patents instead
of the respective patent applications because the respective patents “disclose
substantively the same subject matter and contain substantively identical
disclosures.” Id. at 7 n.2, 9 n.3.
For the Braden-Harder, Heidorn, Messerly, and Grossman challenge,
Petitioner argues that to the extent we decide that the disclosures in Heidorn
and Messerly were not incorporated into Braden-Harder, it would have been
obvious to combine these references. Pet. 33 (citing Commonwealth Sci. &
Indus. Research Org. v. Buffalo Tech. (USA), Inc., 542 F.3d 1363, 1372
(Fed. Cir. 2008)).
Patent Owner does not dispute Petitioner’s assertions that the Heidorn
and Messerly applications are incorporated by reference into Braden-Harder,
or that the respective Heidorn and Messerly patents “disclose substantively
the same subject matter and contain substantively identical disclosures” as
the incorporated Heidorn and Messerly applications. See generally PO
Resp. Nor does Patent Owner contest Petitioner’s alternative theory, that it
would have been obvious to combine Braden-Harder, Heidorn, Messerly,
and Grossman. Id. For purposes of this Decision, we agree with Petitioner,
for the reasons provided in the Petition, that Heidorn and Messerly were
properly incorporated into Braden-Harder at the time of the Braden-Harder
IPR2020-00755
Patent 6,366,908 B1

28
invention and that the respective Heidorn and Messerly patents disclose
substantively the same subject matter and contain substantively identical
disclosures.
Accordingly, we focus on Petitioner’s challenge that claims 6–12 are
unpatentable under 35 U.S.C. § 103(a) as obvious over Braden-Harder
(incorporating Heidorn and Messerly) in view of Grossman.12 In support of
its showing, Petitioner relies upon the declaration of Dr. Bernard J. Jansen.
Id. (citing Ex. 1003). Patent Owner contends that the combined teachings of
Braden-Harder and Grossman do not teach the following limitations:
(1) “keyfact”; (2) the acts for the “keyfact extracting step”; and (3) the
“keyfact retrieving step.” PO Resp. 10–28; Sur-reply 10–22. Patent Owner
does not introduce testimony from a declarant in support of its positions.
For the reasons that follow, we conclude that Petitioner has met its
burden of proving by a preponderance of the evidence that each of the
challenged claims would have been obvious in view of the asserted prior art.
a. Claim 6: “a keyfact-based text retrieving method
comprising:” (preamble)13
For the above limitation, Petitioner contends, and we agree, that
Braden-Harder’s information retrieval (IR) method describes subjecting
documents and queries to natural language processing (NLP) to produce
logical form triples, with each of the triples being of the form “word-
relation-word.” Pet. 33–34 (citing Ex. 1020, 5:22–24, 5:29–35, 11:44–55).
Petitioner further contends, and we agree, that the logical-form triples

12 We refer to this challenge as Braden-Harder in view of Grossman or
Braden-Harder (incorporating Heidorn and Messerly) in view of Grossman.
13 We need not determine whether the preamble is limiting because
Petitioner shows that Braden-Harder meets the preamble.
IPR2020-00755
Patent 6,366,908 B1

29
express semantic relationships between important words in an input string or
phrase. Id. at 34 (citing Ex. 1020, 5:18–20, 11:44–46; Ex. 1001, 1:53–59).
Petitioner further contends that Braden-Harder’s logical-form triples are
equivalent to the claimed keyfacts and provide a concept-based retrieval. Id.
(citing Ex. 1003 ¶ 176). Although Patent Owner argues that Petitioner fails
to show that Braden-Harder in combination with Grossman meets the
“keyfact extracting step,” as explained below, Patent Owner does not dispute
Petitioner’s showing with respect to the preamble. See generally PO Resp.
b. Claim 6: “keyfact extracting step for analyzing a document
collection and a user query, and extracting keywords
without part-of-speech ambiguity from said document
collection and said user query, and respectively extracting
keyfacts of said document collection and said user query
from said keywords”
For the “analyzing a document collection and a user query” portion of
the above limitation, Petitioner contends that Braden-Harder subjects
documents and user queries to morphological analysis to parse text into
constituent words using a stored lexicon. Pet. 37–38 (citing Ex. 1020, 5:33–
35, 11:62–12:1, 12:12–16, 12:30–34; Ex. 1003 ¶¶ 184–187; Ex. 1022, 1:22–
24; Ex. 1029, 112). Petitioner further contends, and we agree, that a
POSITA14 would have understood that Braden-Harder’s lexicon is a
machine-readable dictionary. Id. at 38 (citing Ex. 1003 ¶¶ 184–187;
Ex. 1029, 112). As pointed out by Petitioner, Braden-Harder’s lexicon
“includes various classes of words, such as e.g., prepositions, conjunctions,
verbs, nouns, operators and quantifiers that define syntactic and semantic
properties inherent in the words in an input string.” Ex. 1020, 12:12–16.
Petitioner further contends that using its lexicon, Braden-Harder’s

14 A person of ordinary skill in the art.
IPR2020-00755
Patent 6,366,908 B1

30
morphological analysis normalizes differing word forms to a common
morphological form for use by a parser. Id. at 38 (citing Ex. 1020, 12:34–
37).
Petitioner contends that to accomplish the function of “extracting
keywords without part-of-speech ambiguity from said document collection
and said user query,” Braden-Harder teaches applying parts-of-speech tags
to each extracted word. Id. at 39. Specifically, Petitioner asserts that
Braden-Harder’s morphological analysis attributes each word, or phrase
(noun phrase) in an input string, with a category of its corresponding
grammatical function (part-of-speech tag for that word). Id. (citing
Ex. 1020, 12:40–45). Petitioner contends that a noun extracted from a noun
phrase in Braden-Harder is a keyword used to retrieve documents. Id.
(citing Ex. 1003 ¶ 189; Ex. 1022, 3:36–40).
Petitioner further contends that Braden-Harder’s system syntactically
analyzes stemmed words using grammatical rules and attributes to yield a
syntactic parse tree with at least three levels (Table 1): top (root) node (e.g.,
declarative sentence), intermediate level nodes (e.g., verb, noun, or
prepositional phrases), and a lower-level (leaf) nodes which represent the
initial set of attribute/value records. Id. at 39–40 (citing Ex. 1020, 12:37–40,
Table 1; Ex. 1022, 3:54–59). Petitioner asserts that Table 1 of Braden-
Harder shows that the intermediate-level nodes and lower-level nodes
represent part-of-speech tag sequences for the input string. Id. at 40 (citing
Ex. 1020, Table 3). Petitioner asserts, and we agree, that in Braden-Harder’s
syntactic-parse tree, the extracted words and phrases have no “part-of-
speech ambiguity.” Id.
Petitioner contends that Heidorn describes the specific details of
syntactic processing (extracting keywords without part-of-speech
IPR2020-00755
Patent 6,366,908 B1

31
ambiguity). Id. at 40–42 (citing Ex. 1022, 4:2–3, 4:44–51, Figs. 6, 22; Ex.
1003 ¶ 199). Petitioner argues that, to the extent the claim requires
replacing a sequence of tags rather than selecting a final sequence of tags,
like the ’908 patent’s “KEY,” Braden-Harder similarly generates an
intermediate-level tag sequence for an input sentence in which a sequence of
lower-level-tagged noun phrases are replaced by a single tag “NP,” resulting
in a final sequence tag at the intermediate-level sequence (i.e., NP-VERB-
NP). Id. at 42–43 (citing Ex. 1020, 12:40–45, 12:66–13:10, Table 1).
As further part of the “keyfact extracting step for” limitation, claim 6
recites, “respectively extracting keyfacts of said document collection and
said user query from said keywords.” Petitioner contends that Braden-
Harder teaches processing the syntactic parse tree using a set of semantic
rules (“keyfact pattern rules”) to yield a logical form graph (“keyfact
pattern”) and incorporates Heidorn for the specific details of the logical form
processing. Id. at 43 (citing Ex. 1020, 14:50–51, Figs. 5A–5D). Petitioner
asserts that, like the ’908 patent’s disclosure, the final tag sequence is
applied to a keyfact pattern rule to extract a keyfact pattern (Ex. 1001, 6:33–
36), and Heidorn similarly applies semantic rules to the tag sequence to
generate a logical form graph. Id. at 44–45 (citing Ex. 1022, 2:22–30, Fig.
43; Ex. 1003 ¶ 209). Petitioner further asserts that the extracted logical form
graph captures the structure of arguments and adjuncts for each input string
and provides a pattern representing the sentence in logical form. Id. at 45
(citing Ex. 1022, 13:56–58; Ex. 1003 ¶ 210). Petitioner contends, and we
agree, that Braden-Harder applies rules to the logical form graph to generate
a list of logical form triples, which are extracted based on semantic
relationships of the nouns (keywords) with other words in the sentence. Id.
IPR2020-00755
Patent 6,366,908 B1

32
at 45–47 (citing Ex. 1020, 14:3–4, 14:18–24, 14:28–33, 14:37–41, Figs. 5B–
5D; Ex. 1003 ¶¶ 211–219).
Important to this Decision, Petitioner argues, and we agree, that to the
extent that “extracting keyfacts of said document collection and said user
query from said keywords” requires generating keyfacts that have forms of
[object, property], “a POSITA would have understood that Braden-Harder
generates logical form triples for ‘the case of a constituent with noun
modifiers’ (see Table 3), such as ‘Nadj’ and ‘Mods’ relations, which exhibit
the semantic relationship of the form [object, modifier].” Id. at 47.15
Alternatively, Petitioner contends, and we agree, that it would have been
obvious to generate other logic forms that do not include relations to enable
structural paraphrasing. Id. (citing Ex. 1020, 11:51–55, 25:49–60).
Petitioner further contends, and we agree, that a POSITA would have been
motivated “to generate logic forms that do not include the relations for ‘noun
appositive’ or ‘relative clause’ constructions to increase matches with
‘sufficiently similar semantic content.’” Id. (citing Ex. 1020, 25:51–52; Ex.
1003 ¶¶ 221–222).

15 During the oral hearing, Petitioner stated, “[t]he difference between
property and modifier is inconsequential. We can see from the ’908 patent
at [column 1, lines 6–12] . . . that the property is the modifier, those are the
same things so that is a distinction without a difference.” Tr. 17:14–18. We
agree that the property described in the ’908 patent is the same as a modifier
because the ’908 patent describes it as such. Ex. 1001, 1:7–12 (“In
particular, the methods describe the formalized concept of a document as a
pair comprising an object that is the head and a property that is the modifier,
and uses the information described by the pair as index information for
efficient document retrieval” (emphasis added)). Patent Owner does not
contend otherwise. See generally PO Resp.
IPR2020-00755
Patent 6,366,908 B1

33
Patent Owner argues that the “Petition fails to prove obviousness of
claim limitations reciting the ‘keyfact’ term, under a proper interpretation
that accurately reflects the intrinsic evidence.” PO Resp. 20. Patent Owner
further argues that Petitioner focuses on Braden-Harder in addressing
“analyzing a document collection and a user query” by alleging that Braden-
Harder discloses using a lexicon to recover various classes of words. Id.
at 20–21. Patent Owner argues that “a mere distinction of ‘classes or words’
. . . is distinguishable on its face from the distinct concept of a ‘keyfact’ that
is an extracted expression of fact, much less one that represents a semantic
relation between words of a sentence in the form of [object, property], as
disclosed and claimed in the ’908 patent.” Id. at 21. Patent Owner argues
that in the ’908 patent, the class of noun determines the object or the
property of the keyfacts and that the “Braden-Harder scheme makes no
determination between a class of noun, much less one used to generate a
‘keyfact’ having the form [object, property].” Id. at 22. Patent Owner
further contends that Braden-Harder’s triples are distinguishable from the
keyfact extraction disclosed and claimed, because “the ’908 patent
disparages the sort of phrase-based scheme Braden-Harder employs as being
subject to certain technical disadvantages.” Id. (citing Ex. 1001, 1:19–28,
1:42–45).
Petitioner responds that Braden-Harder meets the claim limitation
under Patent Owner’s construction for “keyfact” presented in its Response.
Pet. Reply 10 (citing Pet. 47; Ex. 1003 ¶¶ 221–222). Petitioner argues, and
we agree, that the Petition explains that Braden-Harder teaches various
special graph walks, including one for the case of a constituent with noun
modifiers. Id. (citing Pet. 45; Ex. 1020, 14:37–41, Fig. 5D; Ex. 1003
IPR2020-00755
Patent 6,366,908 B1

34
¶¶ 216–217). Petitioner points out that the Petition specifically explains that
if keyfacts were construed to have the form [object, property], a person
having ordinary skill in the art would have understood that Braden-Harder
generates logical form triples for the case of a constituent with noun
modifiers such as “Nadj” and Mods” relations which exhibit the semantic
relationship of the form [object, modifier]. Id. at 11 (citing Pet. 47; Ex. 1003
¶ 221). In particular, Dr. Jansen testifies that:
[A] person of ordinary skill in the art would have understood
that Braden-Harder generates logical form triple for “the case of
a constituent with noun modifiers” (see Table 3), such as
“Nadj” and “Mods” relations that exhibit the semantic
relationship of the form [object, Mods, modifier]. Because
Braden-Harder generates logical form triples for noun
modifiers, Braden-Harder generates keyfacts having the forms
of [object, modifier].
Ex. 1003 ¶ 221. Dr. Jansen further testifies, “each of Braden-Harder’s
logical form triples ‘contains two node words … linked by a semantic
relationship there between.” Id. ¶ 176 (alteration in original).16 We give
substantial weight to Dr. Jansen’s unrebutted testimony because it is
supported by the disclosures in Braden-Harder. In particular, Braden-
Harder’s triples are in the form [object, property] because they are in the
form, for example, [object, Mods, modifier17], where each triple contains

16 The Petition lists eight examples of Braden-Harder’s triples that meet the
form [object, property]. Pet. 45–46; Ex. 1003 ¶¶ 216–221; see also Tr.
18:18–24 (explaining that in the example “BOWL – Mods – SHARK” from
Figure 5D, “that ‘BOWL’ would be the object and ‘SHARK’ would be the
modifier for the property”).
17 As explained above in footnote 15, the record supports that there is no
distinction between modifier and property and we treat them the same, as do
the parties.
IPR2020-00755
Patent 6,366,908 B1

35
two node words linked by a semantic relationship. See Ex. 1001, 1:7–12.
We agree with Petitioner that “[i]n the case of noun modifier, Braden-
Harder’s logical form triples represent ‘a paired relationship’ between an
object that is the noun and a property that is the modifier, and are thus in the
paired form of [object, property].” Pet. Reply 12, 18 (citing Ex. 1001, 1:7–
12).
Patent Owner argues that the use of triples in Braden-Harder is
distinguishable from claim language where keyfacts are extracted in a paired
form, i.e. [object, property]. PO Resp. 22. We agree, however, with
Petitioner’s assertion that there is nothing in the ’908 patent “specification or
the prosecution history that would limit ‘keyfacts’ to at most two elements
(e.g., object and property) and nothing more.” Id. at 18–19. Patent Owner
directs us to no intrinsic or extrinsic evidence to support its contention that a
“keyfact” must be in the exact paired form “[object, property]” and nothing
else. Indeed, as pointed out by Petitioner, Table 1 of the ’908 patent
includes, under the “keyfact term list” column “[KEY2 KEY1, NIL].” Ex.
1001, 6:18, Table 1.
In the Sur-reply, Patent Owner argues that “every single ‘keyfact’
example disclosed in the specification expresses semantic relation between
words in the paired form of [object, property].” Sur-reply 2 (citing
Ex. 1001, code (57), Table 1, 1:8–10, 1:16–18, 4:58–60, 6:15–30, 6:38–44)
(emphasis omitted). Patent Owner argues that, in the example from Table 1
of [KEY2 KEY1, NIL], “the ‘property’ portion is ‘NIL’ or empty, such that
KEY2 and KEY1 pair are the only two nonempty keyfact tags.” Id. at 13
(emphasis omitted). We are not persuaded by Patent Owner’s arguments
that the ’908 patent includes a “disavowal expressed in the statement ‘All
keyfacts express semantic relation between words in the form of [object,
IPR2020-00755
Patent 6,366,908 B1

36
property],’” such that a keyfact must contain a pair and nothing else. Id. at
12–13. We agree with Petitioner that the description from the ’908 patent
(all keyfacts express semantic relation between words in the form of [object,
property]) does not preclude, for example, “more than the pair of node
words,” such as Braden-Harder’s triples that include two node words in the
form of [object, property] linked by a semantic relationship. Ex. 1003
¶¶ 176, 221. Patent Owner responds that in every instance shown in Table
1, “the example keyfacts listed never consist of more than a pair of
nonempty tags.” Sur-reply 13 (emphasis omitted). However, even the form
[KEY2 KEY1, NIL], for example, contains an indicator “NIL” along with
two other terms (KEY2 and KEY1), resulting in a representation that is more
than just a pair of terms, like Braden-Harder. Moreover, Patent Owner’s
argument that Braden-Harder’s triples are “the sort of phrase-based scheme”
that the ’908 patent disparages is unavailing. PO Resp. 22 (citing Ex. 1001,
1:19–28, 1:42–45). As explained above, Braden-Harder applies rules to
extract logical form triples, where each triple “is a factual extraction that
represents the semantic relation between words of a sentence,” which is
distinguishable from phrase-based methods that use “purely syntactic”
approaches. Pet. Reply 21–22 (citing Pet. 1, 34; Ex. 1003 ¶¶ 175–176;
Ex. 1020, 3:46–59, 4:18–20, 4:61–67) (emphasis omitted).
We also are not persuaded by Patent Owner’s arguments that claim 6
requires a “determination between a class of noun” used to generate a
keyfact having the form [object, property]. PO Resp. 21–22 (citing Ex.
1001, 5:35–44). Claim 6 does not require making a determination between a
class of noun. A class of noun is not in the claim language at all. Moreover,
we agree with Petitioner that Braden-Harder nonetheless makes such a
determination between a class of noun, and converts a sequence of nouns
IPR2020-00755
Patent 6,366,908 B1

37
into a single tag, like the ’908 patent. Pet. Reply 16–18 (citing Pet. 42, 45–
46; Ex. 1022, Fig. 22; Ex. 1020, 14:37–41; Ex. 1003 ¶¶ 199–200, 216–217).
Lastly, Patent Owner argues that for claim 6, Petitioner misconstrues
acts as underlying functions and that “[i]t is significant here that claim 6
recites the ‘keyfact extracting step’ in the context of both a document
collection (i.e., more than one document) and a ‘user query’ together, for
each one of the corresponding acts.” PO Resp. 23–24. Patent Owner also
argues that Petitioner makes no distinction between its challenge of claim 6
“and the distinct and additional requirement” recited in dependent claim 7.
Id. (emphasis omitted). For the reasons explained above in the claim
construction section of this Decision, we disagree with Patent Owner that the
dependent claims, such as claim 7, include additional acts, and instead we
agree with Petitioner that the dependent “claims should be properly
construed as describing details required as part of performing the previously
introduced steps.” Pet. Reply 5. Also, as explained above, Petitioner
addresses the acts that Patent Owner argues are claimed in claim 6 by
construing the claim under 35 U.S.C. § 112, ¶ 6.
To the extent Patent Owner contends that the Petition fails to show
how Braden-Harder teaches or suggests the “keyfact extracting step” for
both a document collection and a user query together, Patent Owner’s
argument is without merit. PO Resp. 23–24. First, the claim “keyfact
extracting step” does not recite that the analyzing of a document collection
and a user query be performed “together” or at the same time, as Patent
Owner appears to imply. There simply is no such temporal requirement and
Patent Owner fails to explain why we should construe the claim to include
one. Moreover, Patent Owner fails to identify what is missing from
Petitioner’s showing. Id. at 23–24. For example, the Petition shows, and we
IPR2020-00755
Patent 6,366,908 B1

38
agree, that “the ‘keyfact extracting step’ is performed in the context of both
a document collection and a user query.” Pet. Reply 25–27 (citing Pet. 35–
37; Ex. 1003 ¶¶ 180–183; Ex. 1020, Figure 10A); Pet. 33–34 (citing Ex.
1020, 5:22–24, 5:29–35, 11:44–55, and arguing “Braden-Harder’s IR
method subjects both documents and queries to NLP to produce ‘a list of
logical form triples, with each of the triples being illustratively of a form
‘word-relation-word’”). Indeed, Braden-Harder describes that each
document is “subjected to natural language processing, specifically
morphological, syntactic and logical form, to ultimately produce appropriate
logical forms for each sentence in each document. A user-supplied query is
analyzed in the same manner to yield a set of corresponding logical form
triples therefor.” Ex. 1020, 5:29–35; see also id. at 26:33–61 (claim 1
reciting the same). Petitioner explains, and we agree, that Braden-Harder
teaches a document indexing engine that performs a triple-generation
process to extract logical-form triples for each document being indexed. Pet.
35 (citing Ex. 1020, 6:55–57, 20:62–21:4, Figs. 10A, 10B, 11; Ex. 1003 ¶
180). Petitioner further explains that Braden-Harder performs a similar
retrieval process for analyzing a user query to yield logical form triples. Id.
at 36 (citing Ex. 1020, 15:18–21, Fig. 12A). Patent Owner directs us to no
supporting evidence that rebuts Petitioner’s showing, which we determine is
sufficient. For all of the above reasons, we determine that Petitioner has
shown sufficiently that Braden-Harder in view of Grossman teaches or
suggests the disputed “keyfact” limitation.
As explained above, Petitioner alternatively contends, and we agree,
that as to the claimed “keyfact” limitation, it would have been obvious to
generate other logic forms that would not include relations to enable
structural paraphrasing. Pet. 47 (citing Ex. 1020, 11:51–55, 25:49–60).
IPR2020-00755
Patent 6,366,908 B1

39
Petitioner further contends, and we agree, that a POSITA would have been
motivated “to generate logic forms that do not include the relations for ‘noun
appositive’ or ‘relative clause’ constructions to increase matches with
‘sufficiently similar semantic content.’” Id. (citing Ex. Ex. 1020, 25:51–52;
Ex. 1003 ¶¶ 221–222).18 Braden-Harder describes that its “invention can
readily utilize any other form, such as those noted above, that can portray a
semantic relationship between words.” Ex. 1020, 11:51–55. Braden-Harder
further describes that “the criteria for determining a match, for purposes of
identifying sufficiently similar semantic content across triples, can be
relaxed to encompass paraphrases as matching.” Id. at 25:50–53 (emphasis
added). “A structural paraphrase is exemplified by use of either a noun
appositive or a relative clause.” Id. at 25:53–57. Dr. Jansen testifies that
Braden-Harder’s descriptions would have led a POSITA to “be motivated to
generate logic forms that do not include the relations for ‘noun appositive’
or ‘relative clause’ constructions to increase matches with ‘sufficiently
similar semantic content’” thereby ensuring “that all relevant documents are
retrieved from the document collection.” Ex. 1003 ¶ 221. We give
substantial weight to Dr. Jansen’s testimony and agree with Petitioner’s
contentions that it would have been obvious to generate other logic forms
that do not include relations to enable structural paraphrasing. Pet. 47.
In the Patent Owner Response, Patent Owner did not address
Petitioner’s alternative theory of how the prior art renders obvious the
disputed “keyfact” limitation, but only made arguments regarding the theory
in the Sur-reply. See generally PO Resp.; Pet. Reply 19–20; Sur-reply 14–

18 During oral hearing, Petitioner stated that by not including the relation “it
gets rid of that mod that’s in the middle of the tuple” resulting in “literally . .
. this exact form of object and property.” Tr. 19:6–17.
IPR2020-00755
Patent 6,366,908 B1

40
16 (“Google also falsely suggests it is undisputed that it would have been
obvious to modify Braden-Harder to exclude from its ‘triples’ the
interposing relation label”). In other words, Patent Owner waited until the
Sur-reply to address Petitioner’s alternative showing made in the Petition.
We determine it is too late for Patent Owner to do so. In the Scheduling
Order, we advised Patent Owner “that any arguments not raised in the
[Patent Owner Response] may deemed waived.” Paper 15, 8; see also In re
NuVasive, Inc., 842 F.3d 1376, 1381 (Fed. Cir. 2016) (holding that patent
owner’s failure to proffer argument at trial as instructed in the scheduling
order constitutes waiver). A sur-reply that belatedly raises a new issue may
not be considered. See, e.g., Werner Co. v. Louisville Ladder, Inc.,
IPR2019-00336, Paper 34 at 39 (PTAB Aug. 24, 2020) (Final Written
Decision) (“Patent Owner’s arguments made for the first time in its Sur-
reply are waived because they were not made in the Patent Owner Response,
and Petitioner relied on their absence to its detriment and was deprived of
the chance to present arguments and evidence in its Reply to rebut Patent
Owner’s assertions.”); see also Consolidated Trial Practice Guide, 74
(November, 2019)19. Additionally, Patent Owner provides no explanation
for presenting its arguments late. See Sur-reply 14–16.20 Thus, for the
reasons provided in the Petition, we determine that Petitioner’s unrebutted

19 Available at https://www.uspto.gov/sites/default/files/documents/
tpgnov.pdf?MURL=.
20 Even if we considered Patent Owner’s untimely argument that Petitioner’s
proposed modification would “render Braden-Harder inoperable for its
intended purpose,” the underlying contentions are unsupported by evidence.
Sur-reply 14–16. As stated above, Patent Owner does not introduce
testimony from a declarant in support of its positions. Nor did Patent Owner
cross-examine Dr. Jansen. Thus, Dr. Jansen’s testimony stands unrebutted.
IPR2020-00755
Patent 6,366,908 B1

41
alternative showing is sufficient to show that the disputed limitation with the
“keyfact” term would have been obvious.
c. Claim 6: “keyfact indexing step for calculating the frequency
of said keyfacts of said document collection and generating
a keyfact list of said document collection for a keyfact index
structure”
For the above limitation, Petitioner contends that Braden-Harder
(incorporating Messerly) in combination with Grossman teach keyfact
indexing by calculating keyfact frequencies (tf) and document frequencies
(df) to generate tables used to construct a keyfact index structure. Pet. 47–
48. In particular, Petitioner contends, and we agree, that Braden-Harder
teaches scoring metrics to take into account the frequency of a logical form
triple occurring in a document (tf) which corresponds to the claimed keyfact
frequency. Id. at 48 (citing Ex. 1020, 17:60–18:7; Ex. 1003 ¶ 229).
Petitioner further provides sufficient evidence that to the extent “frequency
of said keyfacts” requires using df to calculate idf, Messerly teaches using df
in combination with tf which was a “well-known approach[] to ranking
documents by relevancy.” Id. at 48–49 (citing Ex. 1020, 2:42–46, 17:60–
18:7; Ex. 1025, 12:33–36, 12:51–59, 13:8–10, 14:8–12; Ex. 1003 ¶¶ 232,
234, 235).
For the “generating a keyfact list of said document collection for a
keyfact index structure” limitation, Petitioner further contends, and we
agree, that, as an example, Braden-Harder describes an indexing engine that
generates and stores a set of logical form triples for each indexed document
in a Master Dataset 1030 (index structure). Id. at 50–51 (citing Ex. 1020,
7:12–16, 15:60–63, 16:61–62, 20:10–13, 20:50–61, 20:64–21:3, 21:7–25,
Fig. 10A; Ex. 1003 ¶ 241). Petitioner further provides a detailed showing as
to how the limitation is met if the claimed “keyfact index structure” requires
IPR2020-00755
Patent 6,366,908 B1

42
each of the specific tables in Figure 4 of the ’908 patent. Id. at 52–54.
Other than arguing that “Petitioner errs by conflating distinct claim
requirements in a manner that impermissibly reads out limitations” (PO
Resp. 14–19), which we address above, Patent Owner does not dispute
Petitioner’s specific showing with respect to the above limitation. See
generally id.
d. Claim 6: “keyfact retrieving step for receiving said keyfact of
said user query and said keyfacts of said document
collection and defining a keyfact retrieval model in
consideration of weigh factors according to a keyfact
pattern and generating a retrieval result”
For the above limitation, Petitioner contends, and we agree, that
Braden-Harder teaches, during the retrieval process, obtaining a user’s full-
text query and generating logical form triples from the user query block and
receiving logical form triples associated with the retrieved document
records. Pet. 55–56 (citing Ex. 1020, 11:19–25, 12:16–29, 15:9–21,
Fig. 12A; Ex. 1003 ¶ 257).
Petitioner further contends that Braden-Harder teaches “defining a
keyfact retrieval model in consideration of weigh factors according to a
keyfact pattern,” as Braden-Harder teaches its retrieval model uses
document weights calculated according to a logical-form-triple pattern. Id.
at 57–58 (citing Ex. 1020, Fig. 8B). In particular, Petitioner contends, and
we agree, that Braden-Harder teaches calculating a document weight by
“weighting of matching logical forms . . . associated with a search query and
each of the retrieved documents.” Id. at 58 (quoting Ex. 1020, 5:8–11)
(alteration in original). Petitioner further contends that the weights are based
on the semantic relationships in the logical form triples, such as the words’
functional roles, semantic roles, or semantic labels. Id. (citing Ex. 1020,
IPR2020-00755
Patent 6,366,908 B1

43
Tables 2–4, 5:41–44 (“different relation type, i.e., such as deep subject, deep
object, operator and the like, that can occur in a logical form is assigned a
predefined weight”), 16:21–25, 16:27–29 (“[t]he weight reflects a relative
importance ascribed to that relation in indicating a correct semantic match
between a query and a document”), Fig. 8A). Petitioner asserts, therefore,
and we agree, that “Braden-Harder defines a retrieval model in consideration
of weight constants according to logical form triples, like the ’908 patent.”
Id. at 58–59 (citing Ex. 1003 ¶ 265). Petitioner further contends that to the
extent the claimed “keyfact retrieval model” requires the use of a vector-
space retrieval model, “then Braden-Harder in view of Grossman renders
this obvious” because Braden-Harder acknowledges that use of a statistical
vector-space model was conventional. Id. at 59 (citing Ex. 1020, 25:35–36).
Petitioner asserts that Grossman teaches a vector-space model and further
provides reasons for combining Braden-Harder with Grossman. Id. at 59–61
(citing Ex. 1010, 13, 16, 18, 80; Ex. 1020, 11:4–7, 17:16–18; Ex. 1003
¶¶ 267, 272–275). Lastly, Petitioner contends that Braden-Harder (with
Messerly) teaches using tf-idf weights augmented by its logic-form-triples
weight constant that when implemented with Grossman calculates a
similarity coefficient using vector-space mode that results in an order of
retrieval results. Id. at 61–62 (citing id. § VI.A.1.c (1), § VI.A.1.d;
Ex. 1020, 8:2–6, 16:35–40; Ex. 1003 ¶¶ 279–281).
Patent Owner argues that Braden-Harder’s “predetermined” weights
do not teach or suggest the claimed “weigh factors according to a keyfact
pattern,” which are described in the ’908 patent as keyfact weight constants
“determined experimentally on the basis of distribution of keyfact pattern of
document collection.” PO Resp. 24–28. Petitioner argues that we
“considered and rejected Uniloc’s previous identical arguments” in the
IPR2020-00755
Patent 6,366,908 B1

44
Decision to Institute and that “Uniloc adds nothing to dispute” our
preliminary determinations. Pet. Reply 27.
We agree with Petitioner that Patent Owner’s arguments directed to
the “keyfact retrieving step” are the same arguments Patent Owner raised in
the Preliminary Response. Compare PO Resp. 24–28, with Prelim.
Resp. 27–31. For the same reasons discussed in the Decision to Institute,
those arguments are not persuasive. See Dec. 33–34. That is, we do not
agree that the “weigh factors” must be determined experimentally or cannot
be predetermined weights as Patent Owner asserts. Moreover, the ’908
patent describes determining the weight constants “on the basis of
distribution of keyfact pattern of document collection” (Ex. 1001, 7:62–64)
like Braden-Harder, which similarly describes that its weights “reflect[] a
relative importance ascribed to that relation in indicating a correct semantic
match between a query and a document.” Ex. 1020, 16:27–29. In other
words, Braden-Harder’s weights are assigned based on the keyfact pattern,
like in the ’908 patent. Ex. 1003 ¶ 265.
e. Claim 6: Motivation and Rationale to Combine
Petitioner argues a person of ordinary skill in the art would have been
motivated and found it obvious to combine the teachings of Braden-Harder
and Grossman. See, e.g., Pet. 52–53. Petitioner asserts that Grossman
teaches that tables (or their equivalents), such as those shown in the table of
Figure 4 of the ’908 patent21, were known for implementing document-index
structures such as the indexed database in Braden-Harder. Id. (citing
Ex. 1003 ¶ 244). Petitioner contends, and we agree, that “[s]uch a technique

21 Petitioner makes this assertion assuming that the “keyfact index structure”
is in the form of each of the specific tables in Figure 4 of the ’908 patent.
See Pet. 52–53.
IPR2020-00755
Patent 6,366,908 B1

45
for generating the index structure would have been an obvious
implementation to a POSITA at the time of the ’908 patent, as taught in
Grossman.” Id. at 53 (citing Ex. 1010, 134–137). Patent Owner does not
dispute Petitioner’s showing regarding the reasons to combine Braden-
Harder and Grossman. See generally PO Resp.
For all of the reasons above, we determine that Petitioner has
demonstrated by a preponderance of the evidence that claim 6 would have
been obvious over Braden-Harder in view of Grossman.
f. Dependent Claims 7–12
Each of claims 7–12 depend directly or indirectly from independent
claim 6. Petitioner identifies supporting portions of Braden-Harder and
Grossman, along with arguments and supporting evidence as to how the
combined teachings of Braden-Harder and Grossman account for the
additional limitations of these dependent claims. Pet. 63–84.
For example, claim 7 depends from claim 6 and recites, “wherein said
step of keyfact extracting comprises the steps of analyzing morphology of an
input sentence and obtaining tag sequences of part-of-speech by attaching
part-of-speech tags.” Petitioner references its showing for claim 6 and
argues, and we agree, that Braden-Harder analyzes morphology of input
sentence to parse text into its constituent words using a stored lexicon. Id.
at 63 (citing Ex. 1020, 5:33–35, 11:62–12:1, 12:30–34). Petitioner further
argues, and we agree, that during its morphological analysis, “Braden-
Harder (as explained in Heidorn) attaches each parsed word with a part-of-
speech tag to obtain tag sequences.” Id. (citing Ex. 1020, 12:41–42, 43;
Ex. 1022, 3:36–40; Ex. 1003 ¶¶ 286–287). For the “selecting a tag sequence
of part-of-speech out of said tag sequences of part-of-speech” limitation of
claim 7, Petitioner again refers to its showing for claim 6 and argues, and we
IPR2020-00755
Patent 6,366,908 B1

46
agree, that Braden-Harder (with Heidorn) uses syntactic analysis to select a
final tag sequence out of a set of tag sequences. Id. at 63–64 (citing Ex.
1022, 4:20–22, 4:56–58; Ex. 1003 ¶¶ 289–292). Claim 7 also recites,
“extracting a keyfact pattern by applying said tag sequence of part-of-speech
to a keyfact pattern rule.” Petitioner refers to its showing for claim 6, and
persuasively argues, with supporting testimony, that Braden-Harder uses
semantic analysis to extract a logical form graph by applying its syntactic-
parse tree (with the associated part-of-speech tag sequence) to at least one
semantic rule. Id. at 64 (citing Ex. 1003 ¶¶ 294–295). Lastly, claim 7
recites, “applying said keyfact pattern to a keyfact pattern generation rule
and generating a keyfact list.” Petitioner refers to its showing for claim 6,
and persuasively argues, with supporting testimony, that Braden-Harder
applies the logical form graph to a series of graph-walk rules to generate a
list of logical form triples. Id. at 65 (citing Ex. 1003 ¶¶ 212–218, 296–298).
Patent Owner argues that the “Petition is keyed to incorrect claim
constructions which improperly attempt to incorporate limitations from
dependent claims into independent claim 6.” PO Resp. 29. We have
addressed that argument in our claim construction discussion above. Patent
Owner does not otherwise address separately Petitioner’s arguments and
supporting evidence as to how the combined teachings of Braden-Harder
and Grossman account for the additional limitations of dependent claims 7–
12. See Id. at 28–29; Pet. Reply 30; Sur-reply 1–22. We have reviewed
Petitioner’s arguments and supporting evidence as to how the combined
teachings of Braden-Harder and Grossman accounts for the additional
limitations of dependent claims 7–12, and we agree with and adopt
Petitioner’s analysis. See Pet. 63–84. Accordingly, Petitioner has
demonstrated by a preponderance of the evidence that the subject matter of
IPR2020-00755
Patent 6,366,908 B1

47
dependent claims 7–12 would have been obvious over the combined
teachings of Braden-Harder and Grossman.
E. Remaining Grounds Challenging Claims 6–12
For the reasons discussed above, Petitioner has shown, by a
preponderance of the evidence that claims 6–12 are unpatentable as obvious
over Braden-Harder and Grossman. In addressing this ground, we have
addressed all of the challenged claims. See 35 U.S.C. § 318(a) (requiring the
Board to “issue a final written decision with respect to the patentability of
any patent claim challenged by the petitioner and any new claim added
under section 316(d)”); see also SAS Inst. Inc. v. Iancu, 138 S. Ct. 1348,
1359 (2108) (holding that a petitioner “is entitled to a final written decision
addressing all of the claims it has challenged”). Accordingly, we need not
and do not decide whether Petitioner has shown by a preponderance of the
evidence that claims 6–12 are unpatentable based on the remaining
challenges. Cf. In re Gleave, 560 F.3d 1331, 1338 (Fed. Cir. 2009) (not
reaching other grounds of unpatentability after affirming the anticipation
ground); see also Beloit Corp. v. Valmet Oy, 742 F.2d 1421, 1423 (Fed. Cir.
1984) (holding that once a dispositive issue is decided, there is no need to
decide other issues).
III. CONCLUSION
For the foregoing reasons, we determine that Petitioner has shown by
a preponderance of the evidence that claims 6–12 of the ’908 patent are
unpatentable, as summarized in the following table:
IPR2020-00755
Patent 6,366,908 B1

48
Claims 35 U.S.C. § References
Claims
Shown
Unpatentable
Claims Not
Shown
Unpatentable
6–12 103(a)22 Braden-Harder,
Grossman
6–12
6–12 103(a) Braden-Harder,
Heidorn,
Messerly,
Grossman

7–9 103(a) Braden-Harder,
Grossman,
Kucera

7–9 103(a) Braden-Harder,
Heidorn,
Messerly,
Grossman,
Kucera

9 103(a) Braden-Harder,
Grossman,
Miller

9 103(a) Braden-Harder,
Heidorn,
Messerly,
Grossman,
Miller

9 Braden-Harder,
Grossman,
Kucera, Miller

9 Braden-Harder,
Heidorn,
Messerly,
Grossman,
Kucera, Miller

22 As explained immediately above, we need not and do not decide whether
Petitioner has shown by a preponderance of the evidence that claims 6–12
also would have been obvious based on the remaining grounds not addressed
in this Decision.
IPR2020-00755
Patent 6,366,908 B1

49
Claims 35 U.S.C. § References
Claims
Shown
Unpatentable
Claims Not
Shown
Unpatentable
Overall
Outcome
6–12

IV. ORDER
Accordingly, it is:
ORDERED that claims 6–12 of the ’908 patent have been shown to be
unpatentable; and
FURTHER ORDERED that, because this is a Final Written Decision,
parties to the proceeding seeking judicial review of the Decision must
comply with the notice and service requirements of 37 C.F.R. § 90.2.
IPR2020-00755
Patent 6,366,908 B1

50
For PETITIONER:
Erika Arner
Joshua Goldberg
FINNEGAN, HENDERSON, FARABOW, GARRETT & DUNNER, LLP
erika.arner@finnegan.com
joshua.goldberg@finnegan.com

For PATENT OWNER:
Brett Mangrum
Ryan Loveless
James Etheridge
Brian Koide
Jeffrey Huang
ETHERIDGE LAW GROUP
brett@etheridgelaw.com
ryan@etheridgelaw.com
jim@etheridgelaw.com
brian@etheridgelaw.com
jeff@etheridgelaw.com