Speech and Language Processing: Where have we been and where are we going

Value of the Information & Strength of the Argument

The talk provides valuable insights into the historical development of speech and language processing, drawing on personal experience and well-known examples. The argumentation is solid, using Moore’s law as a lens to analyze progress and differentiate between fields limited by physics versus investment. Church effectively uses anecdotes to illustrate the dangers of overpromising, making the talk both informative and engaging. However, the talk is primarily an opinion piece, and some claims are based on personal recollection rather than rigorous data.

Scientific Rigor, Source Quality, Title Accuracy

The talk is scientifically rigorous in its use of historical data and examples, but it lacks formal citations. The title accurately reflects the content, which is a retrospective and prospective analysis. The sources cited are mostly informal, such as personal communications and borrowed slides, but they are credible given Church’s expertise. The talk does not include a formal bibliography, but it references NIST evaluations and other community benchmarks.

The title accurately reflects the content, which reviews the history and future directions of speech and language processing.

8/10

Talk by a renowned expert in speech and language processing, based on decades of experience and historical data. The content is well-structured and grounded in real examples, but it is an opinion piece rather than a peer-reviewed study.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction and context: after-dinner talk vs. after-breakfast talk, setting expectations.
• 05:00Historical demos: machine translation hype and the 'Casper' demo, illustrating success catastrophes.
• 10:00Discussion of Moore's law and its application to speech and language processing.
• 15:00Examples of progress: speech coding, disk seek vs. capacity, PCs vs. supercomputers.
• 20:00Analysis of error rate reductions in speech recognition over 15 years.
• 25:00Comparison of speech and text rates, and the gap between them.
• 30:00Discussion of periodic and aperiodic signals in progress, and the role of physics vs. investment.
• 35:00Conclusion: the importance of understanding mechanisms to support future research.

• CLSP Seminar Page — Official page for the seminar where this talk was given.

• NIST Speech Recognition Evaluations — Reference to NIST evaluations mentioned in the talk for error rate reductions.