Google LLC v. Uniloc 2017 LLC

1. Case Identification

• Case #: IPR2020-00755
• Patent #: 6,366,908
• Filed: March 27, 2020
• Petitioner(s): Google LLC
• Patent Owner(s): Uniloc 2017 LLC
• Challenged Claims: 6-12

2. Patent Overview

• Title: Keyfact-Based Text Retrieving Method
• Brief Description: The ’908 patent describes a method for information retrieval based on natural language processing (NLP). The system analyzes documents and user queries to extract "keyfacts," which represent semantic relationships between words, and uses these keyfacts to index documents and retrieve relevant results.

3. Grounds for Unpatentability

Ground 1: Claims 6-12 are obvious over Braden-Harder in view of Grossman

• Prior Art Relied Upon: Braden-Harder (Patent 5,933,822) and Grossman (Information Retrieval: Algorithms and Heuristics, 1st ed. 1998). Petitioner also asserts that Braden-Harder incorporates by reference Heidorn (Patent 5,966,686) and Messerly (Patent 6,076,051), making their teachings part of the primary reference.
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Braden-Harder taught a concept-based NLP information retrieval (IR) system that performed the core functions of the ’908 patent. Braden-Harder disclosed analyzing text to generate "logical form triples," which express semantic relationships (e.g., word-relation-word) and are analogous to the ’908 patent's "keyfacts." This process involved morphological analysis, part-of-speech tagging to resolve ambiguity, and generating a list of triples, mapping to the "keyfact extracting step" of claim 6. For the "keyfact indexing step," Petitioner contended Braden-Harder taught calculating frequencies of logical form triples (term frequency) and using document frequency to enhance selectivity, as further detailed in the incorporated Messerly reference. For the "keyfact retrieving step," Petitioner asserted that Grossman taught a conventional vector-space model that could be applied to Braden-Harder's system. Grossman disclosed creating document and query vectors, using term frequency (tf) and inverse document frequency (idf) to calculate weights, and computing a similarity score to rank documents for retrieval.
  • Motivation to Combine: A person of ordinary skill in the art (POSITA) would combine Grossman's conventional vector-space model with Braden-Harder's NLP-based system to implement Braden-Harder's statistical search engine. This combination would use a known, efficient method to rank documents by relevance based on the semantic data (logical form triples) extracted by Braden-Harder, thereby improving retrieval performance with predictable results.
  • Expectation of Success: A POSITA would have a reasonable expectation of success because it involved applying a standard, well-understood IR model (Grossman) to a known type of semantic data structure (Braden-Harder's triples) to achieve the predictable outcome of improved document ranking and retrieval.

Ground 2: Claims 7-9 are obvious over Braden-Harder and Grossman, further in view of Kucera

• Prior Art Relied Upon: Braden-Harder (Patent 5,933,822), Grossman (1998 textbook), and Kucera (Patent 4,868,750).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground built upon the arguments in Ground 1. Dependent claims 7-9 add further detail to the "keyfact extracting step," requiring specific sub-steps like analyzing morphology, attaching part-of-speech tags, and selecting a final tag sequence to resolve ambiguity. Petitioner argued Braden-Harder taught these steps. However, to the extent Braden-Harder's method for resolving ambiguity was deemed insufficient, Kucera was introduced. Kucera explicitly taught a "tag disambiguation processor" that used probabilistic methods to determine a single, unique sequence of tags for an ambiguously-tagged string of words.
  • Motivation to Combine: A POSITA would be motivated to incorporate Kucera's disambiguation processing into the Braden-Harder system to more robustly resolve part-of-speech ambiguities. This would ensure a single, most probable part-of-speech tag is assigned to each word, improving the accuracy of the subsequent syntactic and semantic analysis and reducing unnecessary processing of incorrect parse trees.
  • Expectation of Success: Kucera was designed to solve the exact problem of ambiguity in NLP systems like Braden-Harder. A POSITA would expect that applying Kucera's probabilistic model would predictably improve the precision of the tag selection process.

Ground 3: Claim 9 is obvious over Braden-Harder and Grossman, further in view of Miller

• Prior Art Relied Upon: Braden-Harder (Patent 5,933,822), Grossman (1998 textbook), and Miller (Introduction to WordNet, 1990).

• Core Argument for this Ground:

  • Prior Art Mapping: This ground added to the combination for claim 9, which required the use of separate dictionaries for nouns, verbs, adjectives, etc., including a stop-word lexicon. Braden-Harder taught using a lexicon with multiple part-of-speech classes and skipping stop-words. To the extent this was insufficient, Petitioner introduced Miller. Miller described WordNet, an electronic lexical database that organized words into synonym sets ("synsets") and was structured as separate sub-dictionaries for different parts of speech (nouns, verbs, adjectives).
  • Motivation to Combine: Braden-Harder already disclosed using synonyms or hypernyms to identify similar semantic content. A POSITA would be motivated to use Miller's WordNet as the lexicon for Braden-Harder's system because WordNet provided a more effective, structured way to find synonyms and related terms compared to a traditional dictionary. Using WordNet's separate sub-dictionaries was a direct implementation of the claimed dictionary structure, improving search time and retrieval precision by matching semantically related words.
  • Expectation of Success: Integrating WordNet (Miller) into an NLP system (Braden-Harder) was a well-understood technique. A POSITA would expect this combination to predictably enhance the system's ability to handle paraphrasing and identify relevant documents that do not use exact keyword matches.

• Additional Grounds: Petitioner asserted additional obviousness challenges based on various other combinations of Braden-Harder, Grossman, Kucera, and Miller to address all challenged claims, relying on similar theories of combining known NLP and IR techniques.

4. Key Claim Construction Positions

• “keyfact”: Petitioner argued this term was indefinite under §112 but proposed adopting the Patent Owner's construction of "fact contained in sentences" for the purposes of the IPR. Petitioner contended the prior art disclosed this concept through "logical form triples" which are facts derived from sentences.
• “keyfact extracting/indexing/retrieving step for...”: Petitioner argued these limitations in independent claim 6 should be construed as step-plus-function limitations under §112, ¶6. Petitioner identified the specific acts and structures from the ’908 patent's specification corresponding to the claimed functions and argued that the prior art disclosed the same or equivalent acts/structures for performing those functions.
• Order of Steps: Petitioner argued that the prior art taught the claimed method steps in the same order recited in the claims, making the parties' dispute over whether the order is required non-determinative.

5. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under §325(d), asserting that the primary reference (Braden-Harder) and the asserted combinations were never presented to or considered by the USPTO during the original prosecution. It contended the new art was materially different from the prosecution art.
• Petitioner also argued against discretionary denial under §314(a) (Fintiv factors), noting that the parallel district court litigation was stayed. It further argued that the specific prior art combinations asserted in the IPR, particularly those including Grossman and Miller, had not been proposed or briefed in the district court, meaning the IPR presented issues that would not be resolved in the parallel case.

6. Relief Requested

• Petitioner requested institution of an inter partes review and cancellation of claims 6-12 of the ’908 patent as unpatentable.