The School of Culture, History and Language in the College of Asia and the Pacific is offering Summer Research Internships (SRI) in 2025. ANU students are eligible to apply for a Summer Internship. Summer Research Internships are designed for talented students from ANU considering undertaking postgraduate research in the future. They are an excellent opportunity to carry out research under the supervision of researchers based in the College of Asia and the Pacific, utilising facilities and materials not readily available elsewhere.
The scholarships are available each year for prospective or continuing ANU Student who are currently enrolled in their 2nd or 3rd year of study or postgraduate studies. The program will be run for 9 weeks, commencing Monday 24 November 2025 and finishing on Friday 30 January 2026 (breaking during the Christmas/New Year period spanning Wednesday 24 December to Friday 2 January).
Successful SRI candidates are awarded a weekly stipend of $300 (total to $2700) to cover for the off-campus rental and other associate costs.
All students participating in the program will be invited to participate in College events such as , workshops and seminars during the summer period.
Applications can be lodged online between 1 October and 31 October 2025 only.
View the application form here. Late applications will not be considered.
Equity-based adjustments may be made for Indigenous students or for other reasons aligned to the Education Access Scheme (Equity) Adjustment Schedule. The College of Asia and the Pacific will offer fully funded positions to successful applicants.
For more information about entry requirements and how to apply please email: Shunichi Ishihara.
Forensic analysis of linguistic evidence
Supervisor:
Shunichi Ishihara
Background:
Online communication has become an integral part of daily life, and with it comes the ever-present risk of malicious actors. In the cyber domain—where individual identities can be easily anonymised—the only available evidence in forensic investigations and legal proceedings often consists of the written texts or recorded speech themselves. Meanwhile, advances in generative artificial intelligence (AI) have delivered unprecedented benefits, streamlining tasks and enhancing convenience. Yet these same technologies can be weaponised by bad actors. Across social media, cybercrime, forensic science and
cybersecurity, “deepfake” techniques have emerged as a critical concern, representing potential threats to both personal and national security. Building on this context, each summer scholar will undertake research within the forensic analysis of linguistic evidence. The broad topic categories are:
- Investigation of generative AI’s ability of mimicking individuals’ writing styles
- Investigation of spectral sub-bands in forensic voice comparison (FVC) and related areas
- Investigation of the impact of external factors on forensic text comparison (FTC)
See below for the details.
Applicant requirements (e.g. prerequisites, assumed knowledge): Depending on the project, students are required to have strong coding skills in Python or R, knowledge of phonetics and speech processing using Praat, or experience with natural language processing. These projects are suitable for later-year students in data science, computing, or linguistics who are interested in pursuing honours in the future, master’s students seeking a project topic, or students who would like to work on real language-related problems.
Investigation of generative AI’s capacity to mimic individual writing styles.
Natural language generation (NLG) systems powered by generative AI have raised serious concerns about enabling large-scale, personalised social-engineering attacks. By emulating the writing style of a trusted individual, these systems can generate highly persuasive phishing emails or fabricate credible-looking news. Under this theme, students may explore:
- Comparative impersonation performance across different pre-trained large language models (LLMs).
- Techniques to enhance impersonation beyond one-shot learning.
- The extent to which generative AI can replicate the distinctive style of a well-known author.
Investigation of spectral sub-bands in forensic voice comparison (FVC) and related areas
Different spectral sub-bands may convey varying degrees of speaker-specific information and exhibit differing resilience to external influences (e.g., background noise or emotional variation). Understanding these will fundamentally enhance the adaptability of FVC to various conditions. Under this topic, students might investigate:
- Encoding of speaker information: Where, and to what extent, speaker characteristics are encoded in various spectral regions beyond vowels (students may select one or more consonants for analysis)?
- Robustness to emotional variation: Which spectral sub-band(s) remain most stable across changes in speaker emotion?
- Cross-language FVC: How can speaker-specific information be effectively extracted while minimising the influence of phonetic content across languages?
- Spoofing detection: In which spectral regions is synthetic information most prominently encoded, and to what extent?
Investigation of the impact of external factors on forensic text comparison (FTC)
The performance of an FTC system is vulnerable to a range of external factors. Students may quantify the degree to which performance is affected under the following conditions:
- Data sparsity: the extent to which performance declines as the available text samples decrease.
- Data mismatch: the deterioration in performance when training and test texts differ in topic and data quantity.
- Sampling variability: the fluctuation in system performance attributable to random differences in selected text samples.
- Estimating the weight of evidence while accounting for the uncertainty that the incriminating message may have been AI-generated.